Topic Subtopic

Science & Mathematics Engineering

Robotics
Course Guidebook

Professor John Long

Vassar College
 PUBLISHED BY:

THE GREAT COURSES

Corporate Headquarters
4840 Westfields Boulevard, Suite 500
Chantilly, Virginia 20151-2299
Phone: 1-800-832-2412
Fax: 703-378-3819
www.thegreatcourses.com

Copyright © The Teaching Company, 2015

Printed in the United States of America

This book is in copyright. All rights reserved.

Without limiting the rights under copyright reserved above,

no part of this publication may be reproduced, stored in
or introduced into a retrieval system, or transmitted,
in any form, or by any means
(electronic, mechanical, photocopying, recording, or otherwise),
without the prior written permission of
The Teaching Company.
 John Long, Ph.D.
Professor of Biology
and Professor of Cognitive Science
Vassar College

P
rofessor John Long is a Professor of Biology
and a Professor of Cognitive Science on the
John Guy Vassar Chair of Natural History
at Vassar College. He also serves as the Director
of Vassar’s Interdisciplinary Robotics Research
Laboratory, which he helped found in 2003. He has taught 27 different
courses in four departments and programs, including Perception and Action,
a course in the Cognitive Science Department that features robotics and
laboratories in which students study and program mobile robots. Professor
Long received his Ph.D. in Zoology from Duke University, where he
specialized in biomechanics and received an excellence-in-teaching award.

Professor Long is known internationally for his work in the burgeoning

¿HOGVRIELRURERWLFVDQGHYROXWLRQDU\URERWLFV$VDQDVVRFLDWHHGLWRUKHKDV
KHOSHGODXQFKWZRVFLHQWL¿FMRXUQDOVLQURERWLFVFrontiers in Robotics and
AI in 2014 and Soft Robotics in 2013. He creates self-propelled, autonomous
models of animals, both living and extinct, in order to study how the animals
work, behave, and evolve. He also designs and builds bioinspired robots
in collaboration with computer scientists, electrical engineers, mechanical
engineers, physicists, and development-stage robotics companies.

Professor Long’s research currently is funded by the National Science

Foundation (NSF), and he has received previous research funding awards
IURP WKH 16) WKH 2I¿FH RI 1DYDO 5HVHDUFK WKH 86 6PDOO %XVLQHVV
$GPLQLVWUDWLRQ DQG WKH 'HIHQVH $GYDQFHG 5HVHDUFK 3URMHFWV $JHQF\
+HDOVRVHUYHVDVDQH[SHUWUHYLHZHURIURERWLFVSURMHFWVIRUWKH(XURSHDQ
Commission. Many students working with Professor Long present
WKHLU UHVHDUFK DW QDWLRQDO DQG LQWHUQDWLRQDO VFLHQWL¿F PHHWLQJV DQG HDUQ
FRDXWKRUVKLS RQ VFLHQWL¿F UHVHDUFK SDSHUV DQG PRVW JR RQ WR FDUHHUV LQ
science, technology, engineering, mathematics, or medicine.

i
Professor Long is the author of Darwin’s Devices: What Evolving Robots
Can Teach Us about the History of Life and the Future of Technology, and
along with his students and collaborators, he has published more than 50
SDSHUV LQ VFLHQWL¿F DQG HQJLQHHULQJ MRXUQDOV )RU GHWDLOV VHH KLV *RRJOH
6FKRODUSDJHKWWSELWO\I)H0UI3URIHVVRU/RQJDQGKLVURERWVKDYHEHHQ
featured in the international press; on radio, television, and podcast news
programs; and in science documentaries, including Through the Wormhole
with Morgan Freeman on the Science Channel and Evolve and Predator X
RQ+,6725<+HDOVRKDVEHHQSUR¿OHGLQWKHMRXUQDOScience for a special
issue on robotics. Ŷ

ii
 Table of Contents

INTRODUCTION

Professor Biography ............................................................................i

Disclaimer .......................................................................................... vi
Safety ............................................................................................... vii
Course Scope .....................................................................................1

LECTURE GUIDES

LECTURE 1
The Arrival of Robot Autonomy ...........................................................4
LECTURE 2
Robot Bodies and Trade-Offs ...........................................................18
LECTURE 3
Robot Actuators and Movement .......................................................27
LECTURE 4
Robot Sensors and Simple Communication .....................................35
LECTURE 5
Robot Controllers and Programming ................................................45
LECTURE 6
Human-Inspired Robot Planning ......................................................55
LECTURE 7
Animal-Inspired Robot Behavior .......................................................64
LECTURE 8
Basic Skills for Making Robots .........................................................71
LECTURE 9
Designing a New Robot ....................................................................79

iii
 Table of Contents

LECTURE 10
A Robot for Every Task? ...................................................................89
LECTURE 11
Robot Arms in the Factory ................................................................97
LECTURE 12
Mobile Robots at Home ..................................................................107
LECTURE 13
Hospital Robots and Neuroprosthetics ........................................... 115
LECTURE 14
Self-Driving Vehicles.......................................................................124
LECTURE 15
Flying Robots: From Autopilots to Drones ......................................133
LECTURE 16
Underwater Robots That Hover and Glide .....................................141
LECTURE 17
Space Robots in Orbit and on Other Worlds ..................................150
LECTURE 18
Why Military Robots Are Different...................................................159
LECTURE 19
Extreme Robots ..............................................................................168
LECTURE 20
Swarm Robots ................................................................................176
LECTURE 21
Living Robots? ................................................................................184
LECTURE 22
Social Robots .................................................................................192

iv
 Table of Contents

LECTURE 23
Humanoid Robots: Just like Us? ....................................................201
LECTURE 24
The Futures of Robotics .................................................................210

SUPPLEMENTAL MATERIAL
Timeline ..........................................................................................220
Glossary .........................................................................................227
Answers ..........................................................................................238
Bibliography ....................................................................................253

v
 Disclaimer

This series of lectures is intended to increase your understanding of the

SULQFLSOHV RI URERWLFV 7KHVH OHFWXUHV LQFOXGH H[SHULPHQWV LQ WKH ¿HOG RI
robotics, performed by an experienced professional. These demonstrations
may include dangerous materials and are conducted for informational
purposes only, to enhance understanding of the material.

WARNING: THE DEMONSTRATIONS PERFORMED IN THESE

LECTURES CAN BE DANGEROUS. ANY ATTEMPT TO PERFORM
THESE DEMONSTRATIONS ON YOUR OWN IS UNDERTAKEN AT
YOUR OWN RISK.

The Teaching Company expressly DISCLAIMS LIABILITY for any

',5(&7 ,1',5(&7 ,1&,'(17$/ 63(&,$/ 25 &216(48(17,$/
'$0$*(625/267352),76WKDWUHVXOWGLUHFWO\RULQGLUHFWO\IURPWKH
use of these lectures. In states that do not allow some or all of the above
limitations of liability, liability shall be limited to the greatest extent allowed
by law.

vi
 Safety

As you work through activities in robotics, it is essential that you follow

sound safety procedures. If you are under 18 years of age, you should only
proceed with the supervision of an adult. Some of these activities involve
SRWHQWLDOKD]DUGVWKDWLQFOXGHEXWDUHQRWOLPLWHGWRÀ\LQJSDUWVWKDWFRXOG
harm your eyes; sharp parts that could abrade or cut your skin; mild electric
current that could cause momentary discomfort; spinning motors that could
entangle loose hair or clothing; solder, soldering irons, hot glue, and hot glue
JXQVWKDWFRXOGEXUQ\RXUVNLQRULJQLWHD¿UHDQGKDQGWRROVWKDWFRXOGSRNH
holes, abrade, or cut your skin.

7RLPSURYH\RXUFKDQFHVRIDYRLGLQJLQMXU\DOZD\VVWDUWZLWKDFOHDUZRUN
surface. Any electric tools, such as a hot glue gun or soldering iron, should
never be left plugged in and unattended. Any electric appliance, such as a
hot glue gun or soldering iron, should never be used on or placed next to
anything combustible, such as paper, clothing, or solvents.

Also, before you begin to work, tie back loose hair and secure loose clothing.
5HPRYHDQ\QHFNODFHVRUEUDFHOHWV.HHSDFKDUJHG¿UHH[WLQJXLVKHUDQGD
¿UVWDLGNLWQHDUE\

Never work alone. With a partner, you will be assured of help in case of an
emergency. Make sure that you have a phone handy and that you know the
number for emergency help.

When you are done, make sure to unplug all tools and appliances.

vii
viii
 Robotics

Scope

L
ike computers and self-propelled vehicles of the 20th century,
robots are a technological revolution of the 21st century that impact
nearly every aspect of our lives, businesses, and security. Robotics
XQLWHVPDQ\¿HOGVRIVFLHQFHDQGHQJLQHHULQJDQGWKHVHV\QHUJLHVKHOSXV
FUHDWHDQHZNLQGRIGHYLFHWKDWZH¶YHEHHQGUHDPLQJDERXWIRUPLOOHQQLD
a driverless, self-controlled, goal-driven machine that moves itself or
REMHFWVLQRUGHUWRDFFRPSOLVKHYHU\WKLQJIURPVLPSOHUHSHWLWLYHWDVNVWR
complex missions.

Robotics is the principled study of these remarkable machines with the

goal of being able to design, build, test, and operate them. This course
EHJLQVE\LQWURGXFLQJWKHPDQ\GLIIHUHQWVKDSHVDQGVL]HVRIURERWVJLDQW
MHW SODQHV WLQ\ QDQRERWV XQZLHOG\ ZHOGLQJ DUPV EDUNLQJ WR\V DQG ORZ
slung circular appliances.

The principle that forms the backbone of the course across the many kinds
of robots is autonomy. An autonomous robot can act on its own to achieve
its goals. Those goals are built into its body and programmed into its
brain-like computer.

7RXQGHUVWDQGDXWRQRPRXVURERWVZHVSHQGWKH¿UVW¿YHOHFWXUHVDQDO\]LQJ
them, experimenting to see what they can do and how they do it. We take
URERWVDSDUWDQGSXWWKHPWRJHWKHUDQGZHVXUYH\¿YHHVVHQWLDOFRPSRQHQWV
bodies, actuators, sensors, energy supply, and controllers.

We learn how to create and program a simple behaviorally autonomous

URERW7KH NH\ LV WR FUHDWH D SHUFHSWLRQDFWLRQ ORRS WKDW OLQNV VHQVRUV DQG
motors. That linkage is physical, electrical, and information-based, mediated
through the robot’s computer and the physical interactions with the world
in which the robot is moving. We use an open-source computer controller
called an Arduino, a simple sensor, and a servomotor to build a swimming
robot called a Tadro.

1
 More complex robots also take their inspiration from nature. In Lectures 6
and 7, we explore how humans and other animals offer solutions for basic
functions that nearly all robots must accomplish. Navigation is central,
and robots that use maps and models of the world can make and enact
SODQV 6XFFHVVIXO QDYLJDWLRQ DOVR UHTXLUHV WKDW URERWV DGMXVW UDSLGO\ WR WKH
XQH[SHFWHG LQFOXGLQJ DYRLGLQJ REMHFWV IROORZLQJ ZDOOV DQG HVFDSLQJ
IURP WLJKW VSRWV 7KHVH DUH UHÀH[LYH EHKDYLRUV LQVSLUHG E\ WKH VXFFHVVIXO
workings of insects.

Roboticists need to know some basics of working with wires and electronics,
and some simple techniques are introduced in Lecture 8. The do-it-yourself
approach is great for putting the principles of robotics to work to build
simple robots.

For robots to accomplish complicated tasks, we need to turn to the formal

design processes of engineering (Lecture 9), which involves understanding
the task that the robot needs to accomplish. Task analysis is explained in
Lecture 10, laying the groundwork for breaking down any task into the steps
that are needed to build and program a robot.

In Lectures 11 through 18, we explore the diverse environments where we

use robots—from factories, homes, hospitals, and the road to the air, the
ZDWHUVSDFHDQGWKHEDWWOH¿HOG

In the last six lectures, we address the latest research and designs for robots.
([WUHPHURERWVDUHWKHIRFXVRI/HFWXUHLQZKLFKZHDQDO\]HWKHELJJHVW
the smallest, and the fastest robots, with an emphasis on robots with legs.

In Lecture 20, we examine swarms of robots and how they coordinate their
collective movements. We push the limits of the idea of modularity, the
semi-independent functioning of elements within a single robot or across
multiple robots.
Scope

2
,Q /HFWXUH  ZH H[DPLQH ZKDW HOVH OLIH GRHV ZHOO DQG ¿QG WKDW HDWLQJ
developing, self-replicating, and evolving provide further goals for
robotics—four additional kinds of autonomy. We discover robots that
attempt to harvest their own energy, assemble themselves, build copies of
themselves, and redesign themselves in response to feedback from the world
about their performance.

One of the most exciting uses of robots is to have robots cooperate with and
work alongside humans. This requires attention from humans, and the design
RI DWWHQWLRQJUDEELQJ URERWV WDNHV XV WR WKH QHZ ¿HOG RI VRFLDO URERWLFV
(Lecture 22), in which emotional interactions can make or break a human-
robot relationship.

Lecture 23 focuses even more directly on humanoid robots. We explore

WKH YDOXH RI KXPDQRLGV LQ HQYLURQPHQWV WKDW FDQQRW EH PRGL¿HG WR
DFFRPPRGDWH WKH URERW )RU H[DPSOH ¿UVW UHVSRQGHUV GXULQJ D GLVDVWHU
confront degraded spaces and equipment typically designed for humans, not
robots, creating a situation in which humanoid robots might succeed where
other robots would fail.

5RERWLFV LV D FDSVWRQH ¿HOG ZLWK PXOWLSOH RULJLQV DQG WKHUH DUH PXOWLSOH
futures to be expected, too, including miniaturization, cloud robotics
(in which robots share what they learn through the Internet), Watson-
style cognitive robotics, biohybrids, and evermore-powerful modularity.
Autonomous robots are amazing machines, and the study of robotics allows
XVWRXQGHUVWDQGDQGEHQH¿WPRUHIXOO\IURPKRZWKH\ZRUN²DQGWRGHVLJQ
DQGEXLOGRXURZQŶ

3
 The Arrival of Robot Autonomy
Lecture 1

T
KH ¿HOG RI URERWLFV RIIHUV DQ HQGOHVVO\ H[FLWLQJ ZD\ RI VHHLQJ RXU
world. Almost everything around you—any machine, any electronics
component, any animal, anything that humans do—has potential
implications for robotics, which has implications for you. Robotics is a
¿HOG ZKHUH PDQ\ GLVFLSOLQHV FRPH WRJHWKHU VR WKH VRXUFHV RI SRWHQWLDO
inspiration are endless. In this course, you will discover how autonomous
robots work and learn the science and engineering behind how they are
designed and built.

Robots in Our World

z Mars has robots exploring the planet’s surface on our behalf.
NASA’s Curiosity, also known as the Mars Science Laboratory, is
140 million miles away. With a generator powered by nonexplosive
plutonium 238, Curiosity uses its camera, along with lasers and
abraders for pulverizing rocks, a grabber for taking samples, and
a miniature onboard laboratory for analyzing samples, to explore
Mars. Curiosity is an explorer and a scientist, and if any signs of
Lecture 1—The Arrival of Robot Autonomy

life are found on another planet, that discovery may come from
a robot.

z In our oceans, we also have robots at work. These robots search

IRU ORVW DLUOLQHUV ¿[ OHDNLQJ RLO SLSHOLQHV DQG GLVFRYHU ORQJ
lost shipwrecks. Just like with our explorers on other planets,
underwater robots work in dangerous conditions and for periods of
time that humans cannot.

z For decades, tens of thousands of robots have been added every

\HDU WR PDQXIDFWXULQJ SODQWV 0XOWLMRLQWHG URERWLF manipulators
move heavy car doors into place for assembly. Robotic arms weld
parts together, and robotic sprayers paint the body. Increasingly
complicated teams of robots even assemble cars.

4
z 5RERWVDUHQRORQJHUZRUNLQJMXVWLQFDJHGRIIDUHDVWKH\QRZZRUN
VLGHE\VLGHZLWKKXPDQV:HFDOOWKLV¿HOGFROODERUDWLYHrobotics.

z Some robots can drive themselves. Thirty percent of Curiosity’s

GULYLQJRQ0DUVGXULQJLWV¿UVWPRQWKVZDVVHOIFRQWUROOHG8VLQJ
stereovision from two cameras, Curiosity created a drivability
map, an estimate of the danger of the path ahead. By navigating on
its own, Curiosity can cover more terrain than would be possible if
it had to rely entirely on remote control by humans.

z 2Q (DUWK ZH KDYH GULYHUOHVV FDUV JHWWLQJ OLFHQVHG WR FDUU\
passengers on ordinary roads. They navigate and drive themselves.
The whole car is the robot. There are sensors on top, helping the
robot navigate, and a computer
on board makes decisions about
which way to steer, how fast to
go, and how wide to go when
passing another vehicle.

‹'DULR/R3UHVWLL6WRFN7KLQNVWRFN
z Robots, such as Husqvarna’s lawn-
mowing robot, can automatically
cut your lawn, and there are robots
available to wash windows, such
DVWKH:LQERWE\(FRYDFV5RERWV
Lawn-mowing robots make
DOVRFOHDQRXUÀRRUVDQGPLOOLRQV
landscaping simpler and less
of homes have iRobot’s Roomba time consuming than it would
RURWKHUURERWYDFXXPVRQWKHMRE be if a human had to do it.

z Robots perform other work in extreme environments, for example,

search and rescue, after disaster strikes. Robot snakes can crawl
through tight spaces, carrying cameras and sensors to show topside
humans what’s inside and to search for survivors.

z 5RERWV ZHUH RXU ¿UVW H\HV LQ WKH )XNXVKLPD QXFOHDU SODQW DIWHU
disaster struck. The iRobot PackBot brought in a live video feed
and took temperature and radiation measurements, and it can even
FOLPEXSÀLJKWVRIVWDLUV

5
 Robotics in Popular Culture

  DULD WKH ¿UVW URERW WR DSSHDU LQ D PRYLH VWDUV LQ WKH
0
silent classic Metropolis.

1941 The word roboticsLV¿UVWXVHGLQWKHVKRUWVWRU\³/LDU´E\

writer Isaac Asimov.

1942 The three laws of robotics are created by Isaac Asimov in

KLVVKRUWVWRU\³5XQDURXQG´

1950 I, Robot, a collection of robotics short stories by Isaac

Asimov, is published.

  ,77 D ¿FWLRQDO DXWRQRPRXV WDONLQJ FDU DSSHDUV LQ WKH

.
television series Knight Rider.

  LFWLRQDO ³UHSOLFDQWV´ DSSHDU LQ Blade Runner, a science

)
¿FWLRQ ¿OP DGDSWHG IURP 3KLOLS . 'LFN¶V Do Androids
Dream of Electric Sheep? (1968).

1987 Star Trek: The Next Generation, a television show, debuts

with android lieutenant commander Data.
Lecture 1—The Arrival of Robot Autonomy

z Robots are in hospitals, helping medical personal treat patients.

Teamwork between surgeons and their robotic assistants, such as
da Vinci’s surgical robot, allows surgeons to perform minimally
invasive surgery.

z Robots are also taking education by storm, thanks to efforts from

companies like ArcBotics, Arduino/(*25RERPDWWHUDQG9(;
Robotics. Robotics clubs and competitions are everywhere and
entice people of all ages, from elementary schools to colleges, and
even to million-dollar-challenge events attracting some of the best
companies in the world.

6
The Development of Robots
z After World War II ended in 1945, the technological advances
EURXJKWE\WKDWFRQÀLFWKDGHYHU\RQHORRNLQJIRUSHDFHIXOXVHVIRU
ZHDSRQV 6FLHQFH ¿FWLRQ WRRN RII DV D ZD\ WR WKLQN DERXW D QHZ
future—to prototype a future.

z Isaac Asimov led the charge, coining the word robotics in 1941 and
SXEOLVKLQJDVKRUWVWRU\LQFDOOHG³5XQDURXQG´LQZKLFKKH
introduced his now-famous three laws of robotics 'RQ¶W LQMXUH
KXPDQV REH\ KXPDQV ZKHQHYHU WKDW FRPPDQG GRHVQ¶W LQMXUH
humans; protect yourself, the robot, as long as you obey and don’t
LQMXUHKXPDQV

z 6FLHQFH¿FWLRQKHOSHGXVGUHDPDZRUOGRIIDQWDVWLFURERWV5REE\
WKH 5RERW VWDUUHG LQ WKH  ¿OP Forbidden Planet as a good
robot, helping humans. Of course, the real Robby was a metal suit
worn by a human; we didn’t have real, capable humanoid robots
in 1950.

z Flying cars, personal helicopters, and trips to the Moon were all
being promised. In 1956, a group of computer scientists met to
discuss how to make machines that were humanly intelligent. Nine
years later, fueled by the advent of electronic computers, future
Nobel Prize winner Herbert Simon optimistically predicted that
³PDFKLQHVZLOOEHFDSDEOHZLWKLQWZHQW\\HDUVRIGRLQJDQ\ZRUN
D PDQ FDQ GR´ 7KH SXUVXLW RI WKLV JRDO FUHDWHG WKH ¿HOG WKDW ZH
NQRZDVDUWL¿FLDOLQWHOOLJHQFH$,

z Half a century later, we know that what Simon said was wrong,
at least in terms of a delivery date for the promise. But while
personal aviation may have stalled around 1980, at least in terms
of the number of private pilots in the United States, robots have
been a very different story. We now have humanoid robots, such as
Honda’s ASIMO.

7
Lecture 1—The Arrival of Robot Autonomy

‹.RLFKL.DPRVKLGD*HWW\,PDJHV1HZV7KLQNVWRFN

Honda’s humanoid robot ASIMO is able to react to its environment and interact
with humans.

8
z It’s only in the 21st century that we’ve seen case after case of robots
ZRUNLQJQRWMXVWLQWKHODERUDWRU\EXWDOVRDVUHOLDEOHFRPPHUFLDO
products, such as the Roomba. We even have robot aircraft, called
drones, which are used by the militaries of many countries and, as
fast as permits are issued, by everyone from delivery services to
Hollywood movie makers.

z But why has this process of creating intelligent machines taken so

much longer than the 20 years that Simon predicted? The gestation
period for the working technology of robots has been much longer
than the memory of anyone living today.

z 0HFKDQLFDO ¿JXUHV ZH FDOO automatons have been around since

WKH0LGGOH$JHVLQ(XURSHDQGZHUHNQRZQLQ&KLQDDQGDQFLHQW
*UHHFHORQJEHIRUHWKDW7KH¿UVWVHULRXVGHVLJQIRUDKXPDQRLG
robot comes from Leonardo da Vinci working in the late 1400s
and early 1500s.

z Leonardo never had a chance to build his humanoid. But from his
notebooks, we can tell that his robot had an exoskeleton of armor.
Inside, pulley, gears, and cables were connected to move the hands,
wrist, elbow, and shoulder. These actuators, which are what create
movement, might have been able to connect with a mechanical
FDUWPDNLQJLWSRVVLEOHWRUHSURJUDPWKH¿JXUH7KLVFRPELQDWLRQ
is what made Leonardo’s ideas an advance over the clockwork
automatons that had already been livening up public squares in
PHGLHYDO(XURSH

z There are three ingredients that are missing from Leonardo’s

KXPDQRLG WKDW ZH VHH LQ WRGD\¶V DXWRQRPRXV URERWV HOHFWULF
sensors, electric motors, and digital electronic computers. These
three ingredients combined result in robotic complexity that we
don’t see in purely mechanical machines.

9
 z The great virtue of a digital electronic computer is the algorithms
we can put into the computer itself. Those algorithms, which can
be reprogrammed easily in software, take the place of hardware
circuits that have to be rebuilt.

z This kind of potential to make machines more intelligent is what

Simon and other computer scientists got so excited about—
and rightfully so. But what caused Simon’s timing to be overly
optimistic is that we knew less about how humans and animals
work than we thought.

Types of Robots
z The moment we started to try to create machines that could
VHH²D¿HOGFDOOHGFRPSXWHUYLVLRQ²ZHUHDOL]HGWKDWLGHQWLI\LQJ
DQREMHFWLQDYLVXDO¿HOGZDVH[WUHPHO\FRPSOLFDWHG7KHURERW
named ShakeyDPRELOHURERWIURPWKHODWHVZDVWKH¿UVWWR
KDYHDQGXVHREMHFWLGHQWL¿FDWLRQWRPRYHDURXQGDURRP6KDNH\
had rudimentary vision and the algorithms to turn a pattern of
OLJKWLQWHQVLW\LQWRLQIRUPDWLRQDERXWREMHFWVLQWKHZRUOG

z ,W ZDV LQ 6KDNH\ WKDW HYHU\WKLQJ FDPH WRJHWKHU HOHFWULF PRWRUV
electric sensors, and electronic computers. Shakey was a mobile
Lecture 1—The Arrival of Robot Autonomy

robot that could sense its world, reason about the state of the world
and its place in the world, make plans about how to move in the
world, and then enact those plans.

z Shakey had a computer, VRQDU UDQJH ¿QGHUV, a video camera,

and bump detectors. Shakey was built to navigate an internal
VSDFH PXFK OLNH D VLPSOL¿HG RI¿FH EXLOGLQJ ,W ZDV JLYHQ D PDS
RI WKDW VSDFH WKDW LQFOXGHG REMHFWV OLNH ZDOOV DQG EORFNV *LYHQ
WKH SUHVHQFH RI ZDOOV DQG REMHFWV 6KDNH\ FDOFXODWHG D SDWK WR LWV
GHVWLQDWLRQDQGPRYHGIRUZDUGDQGDURXQGREMHFWV

z Shakey was a robot that had intelligence, as seen through its

behavior. Its intelligent behavior set the standard for modern
mobile robots, and nearly every mobile robot today owes
something to Shakey.

10
 Robotics Firsts

~1495 Leonardo da Vinci designs a mechanical knight

using pulleys and cable, as well as a mechanical cart
capable of serving as a programmable controller.

  KHUHPRWHFRQWUROLV¿UVWSDWHQWHGDQGGHPRQVWUDWHG
7
on a model ship by Nikola Tesla.

1956 $UWL¿FLDO LQWHOOLJHQFH is coined and launched as a

research area by John McCarthy and others at a
conference at Dartmouth College.

± 6KDNH\ WKH ¿UVW DXWRQRPRXV PRELOH URERW ZLWK

a digital electronic computer, is built by Stanford
Research Institute.

1984 Vehicles, a book by Otto Braitenberg, lays the

groundwork for the design of the simplest-possible
autonomous robots.

1990 The iRobot Corporation is founded by Rodney

Brooks, Colin Angle, and Helen Grenier.

2000 ASIMO, a humanoid robot, is introduced by Honda

Motor Corporation.

2002 Roomba, an animal-inspired robotic vacuum cleaner,

is introduced commercially by iRobot.

2005 Arduino, an inexpensive open-source microcontroller,

is created by Interactive Design Institute in Ivrea, Italy.

z In addition to mobile robots, the other great class of robots is

stationary, and we call them robotic manipulators. While they don’t
move around to get their work done, they move an appendage, often
DQDUPLQRUGHUWRSLFNXSDQGPRYHDQREMHFWLQWKHZRUOG

11
 z There is more than one way to put a manipulator on a mobile
robot—that is, there is more than one type of possible robot body
IRUDOPRVWDQ\JLYHQWDVN,Q¿FWLRQURERWVPRVWRIWHQKDYHERGLHV
like humans. Humanoids are mobile, and they have manipulators.

z But robots do not have to be built to resemble humans to take

advantage of biology. Robots can be inspired by the bodies of other
animals. The robot BigDog, built by Boston Dynamics, is a famous
nonhumanoid biomorph,W¶VEHHQPRGL¿HGWRKHOSFRPEDWWURRSV
transport huge loads of equipment over rough terrain. It’s built like
a dog, with four legs that have the basic anatomy of dog legs.

z By contrast, robots with no biological inspiration can be called

mechanoids, a category that would include wheeled vehicles,
such as Shakey and the
Mars rovers.

z Other robotic devices

are worn by the user and
can be strapped to the
outside of your body,
leading to the expression
Lecture 1—The Arrival of Robot Autonomy

wearable robotics. For

example, there’s a walk-
assist device that can
‹.RLFKL.DPRVKLGD*HWW\,PDJHV1HZV7KLQNVWRFN
help people with muscle
weakness from disease.
It was a spin-off from
technology developed
by Honda to build the
humanoid ASIMO. It’s a
robot in the sense that it
has sensors, a computer,
and motors. Some people
like to call these wearable The Land Walker robot is an example of
robots exoskeletons. what some might call an exoskeleton.

12
z In still other cases, robotics are being implanted in humans in a more
SHUPDQHQWZD\VXFKDVWKH'(.$$UP7KHWHUP³ZHDUDEOH´URERW
GRHVQ¶WZRUNIRUWKH'(.$$UPLW¶VDVPDUWneuroprosthetic. It
uses electric signals from the human to grasp, lift, and release. What
sets apart robotic prostheses is that they have onboard intelligence,
FRXSOHGZLWKVHQVRUVDQGPRWRUVFRPSRQHQWVWKDWZHZRXOG¿QG
in any robot.

z Some of the greatest advances in robotics during the past few

decades have come from learning to simplify robots—to simplify
demands on their computers in particular—making them less like
a human brain. The Roomba vacuum robot moves much faster
WKDQ6KDNH\LQSDUWEHFDXVH5RRPEDGRHVQRWWU\WR¿JXUHRXWLWV
environment before making a move. Instead, Roomba gets right to
ZRUN WDNLQJLWVLQVSLUDWLRQIURP DQLPDOV¿JXULQJ RXWZKDW WR GR
next while already moving.

z Roomba is an example of behavior-based robotics; Shakey is

an example of model-based robotics. These terms refer to the
different strategies that we take in the software of the controller.
And they are not mutually exclusive; most modern robots use a
combination of these two architectures.

z Intelligent robots, such as Roomba and Shakey, operate on their

own—it’s a closed loop. Shakey has servomotors responding to
feedback from the sensors. Operation on its own is referred to as
behavioral autonomy.

z Compare that autonomy to a robot that is completely remote

controlled, with a human sensing the world and then determining
the actions of the robot. In this case, most or all of the intelligence
of the robot is located in the human, not the robot. A behaviorally
autonomous robot, acting on its own, has its own intelligence. It
doesn’t need a human in the control loop.

13
 Activities

Moving the Robot

&KHFN RXW WKLV WXWRULDO IURP $UF%RWLFV PDNHU RI 6SDUNL KWWS
DUFERWLFVFRPOHVVRQVPRYLQJWKHURERW ,W¶V D JUHDW LQWURGXFWLRQ
QRWMXVWWR6SDUNLEXWDOVRWRDQ\URERWWKDWPRYHVXVLQJGLIIHUHQWLDO
drive. The principles here will work on any robot with two actuators
moving in a plane.

Building a Tutebot
Tutebot is a very simple robot, making it a great place to start
building your own. Mobile Robots by Jones, Flynn, and Seiger has
the full instructions. Chapter 2 also includes an introduction to the
electronic components and how they work.

Tutebot does not have a programmable digital microcontroller.

7R ³UHSURJUDP´ LW \RX KDYH WR FKDQJH WKH HOHFWURQLFV ,W LV DQ
analog computer. If you build your circuit on a breadboard, then
you can rearrange components more easily than if you create a
soldered circuit.
Lecture 1—The Arrival of Robot Autonomy

<RX FDQ DOVR ³UHSURJUDP´ WKLV DQDORJ FRPSXWHU E\ DOWHULQJ WKH
settings on the potentiometers using a small screwdriver. The circuit
has four potentiometers; they are the little blue boxes with the white
screw on top. By changing the resistance of the two potentiometers
linked directly to the capacitors in the back, you control how quickly
the capacitors release stored charge, which determines how long the
motors operate in reverse.

Capacitors are a special type of energy storage device. They are

charged quickly when the switch on the bumper is activated and
Tutebot begins to back up. If the potentiometers are set to low
resistance, the capacitors discharge rapidly when the bumper is
released. If we set the potentiometers to a high resistance, then the
capacitors drain more slowly and the motors will keep spinning in

14
 reverse for a longer period of time. If you set the two potentiometers
at different resistances, then Tutebot will back up and turn in an arc.
This reverse-and-turn motion is what gives Tutebot its ability to
perform wall-following behavior.

Important Terms

actuator 7KH PRYLQJ SDUWV RI D URERW WKDW DOORZ LW WR DFW DQ\ SDUW
appendage, or mechanical system that uses motors to move a robot or
manipulate the world through movement.

Arduino $Q RSHQVRXUFH KDUGZDUH DQG VRIWZDUH FRPSDQ\ NQRZQ IRU

microcontrollers.

ASIMO7\SHRIKXPDQRLGURERWHPSKDVL]LQJELSHGDOPRELOLW\RYHUXQHYHQ
terrain; in development since 2000 by Honda.

automaton$PHFKDQLFDOPDFKLQHVKDSHGOLNHDKXPDQRUDQLPDOWKDWZRUNV
automatically without feedback from sensors or direct control by humans.

behavior-based robotics 'HVLJQ RI URERWV WKDW HOLPLQDWHV RU PLQLPL]HV

WKHXVHRILQWHUQDOZRUOGPRGHOVDQGPD[LPL]HVWKHXVHRIUHÀH[OLNHVHQVH
act modules.

biomorph$Q\URERWPRGHOHGDIWHUDOLIHIRUP

controller7KHPHFKDQLFDOHOHFWURQLFRUFRPSXWHUL]HGSDUWRIDURERWWKDW
converts information provided by sensors into instructions—whether to
DGMXVWDFWXDWRUEHKDYLRURURYHUDOOPRWLRQXSGDWHPDSVDQGRWKHULQWHUQDO
models, or provide information for human operators.

drivability map$SODQQLQJPRGHOWKDWLVFRQWLQXDOO\XSGDWHGDQGXVHGWR
plot the immediate course for a robot.

15
 drone $Q\ XQPDQQHG DHULDO YHKLFOH 8$9 HVSHFLDOO\ RQH WKDW FDQ À\
autonomously (using GPS or other navigational data) and beyond the line of
sight needed for radio-controlled (RC) aircraft.

humanoid A robot designed to look and function like a human. See

Actroid, android, and cyborg.

manipulator $ URERWLF DUP WKDW JUDVSV DQG PRYHV REMHFWV DOVR DQ\
stationary robot that has one or more such arms.

mechanoid$W\SHRIURERWEXLOWZLWKRXWLQVSLUDWLRQIURPELRORJ\

model-based robotics'HVLJQRIURERWVWKDWPD[LPL]HVWKHXVHRILQWHUQDO
world models, ongoing and sophisticated planning algorithms, and complex
goals and tasks.

neuroprosthetic7\SH RI URERWLF PDQLSXODWRU XVHG DV D UHSODFHPHQW OLPE

for human amputees and controlled by muscular or neural signals.

UDQJH ¿QGHU $Q DFWLYH VHQVRU WKDW EURDGFDVWV XOWUDVRQLF RU RWKHU VRXQG
waves and then uses the time it takes for the echo to return to measure the
GLVWDQFHWRWKHUHÀHFWLQJVXUIDFH6HHsonar.
Lecture 1—The Arrival of Robot Autonomy

robot A type of machine that can be remote controlled, partially autonomous,

RUIXOO\DXWRQRPRXVDVLWPRYHVLWVHOIRUREMHFWVLQRUGHUWRFDUU\RXWWDVNV
While robots always have controllers and actuators, remote-controlled robots
may lack onboard sensors.

robotics 7KH ¿HOG RI VWXG\ DQG LQTXLU\ WKDW GHYHORSV SULQFLSOHV DQG
approaches for the design, fabrication, operation, and control of robots.

Roomba7KH ÀRRUFOHDQLQJ URERW LQWURGXFHG LQ  E\ L5RERW WKH ¿UVW
fully autonomous home robot to achieve commercial success.

16
sensor $Q\ GHYLFH WKDW GHWHFWV FKDQJHV LQ SK\VLFDO SURSHUWLHV RU HQHUJ\
patterns in the world or the robot and converts those into electric, chemical,
RU PHFKDQLFDO VLJQDOV XVDEOH E\ D FRQWUROOHU WR DGMXVW DFWXDWRU EHKDYLRU
and overall motion, update maps and other internal models, or provide
information for human operators.

servomotor $Q HOHFWULF '& PRWRU WKDW XVHV VHQVRU\ IHHGEDFN IURP DQ
internal potentiometer to precisely control position and movement.

Shakey7KH¿UVWPRELOHDXWRQRPRXVURERWFRQWUROOHGE\DUHSURJUDPPDEOH
digital electronic computer.

sonar6WDQGVIRUVRXQGQDYLJDWLRQDQGUDQJLQJW\SLFDOO\DQDFWLYHVHQVRU
system that broadcasts sound waves and then uses the pattern of the returning
HFKRWRPHDVXUHWKHGLVWDQFHDQGVL]HRIREMHFWV

stereovision 9LVXDO VHQVLQJ WKDW XVHV WZR FDPHUDV IRFXVHG RQ WKH VDPH
REMHFW RU VFHQH WR SURYLGH LQIRUPDWLRQ DERXW GHSWK UDQJH DQG WKUHH
dimensional shape.

three laws of robotics ,QWURGXFHG E\ ,VDDF $VLPRY LQ WKH VKRUW VWRU\
³5XQDURXQG´  )LUVW ODZ$ URERW PD\ QRW LQMXUH D KXPDQ EHLQJ RU
WKURXJK LQDFWLRQ DOORZ D KXPDQ EHLQJ WR FRPH WR KDUP 6HFRQG ODZ $
robot must obey the orders given it by human beings, except where such
RUGHUVZRXOGFRQÀLFWZLWKWKH¿UVWODZ7KLUGODZ$URERWPXVWSURWHFWLWV
RZQH[LVWHQFHDVORQJDVVXFKSURWHFWLRQGRHVQRWFRQÀLFWZLWKWKH¿UVWRU
second laws.

Suggested Reading

Bekey, Autonomous Robots, chap. 1.

Jones, Flynn, and Seiger, Mobile Robots, chap. 2.
Rosheim, Leonardo’s Lost Robots. chap. 3.
6PLWKHUV³$XWRQRP\LQ5RERWVDQG2WKHU$JHQWV´

17
 Robot Bodies and Trade-Offs
Lecture 2

B
y taking apart and building, or analyzing and synthesizing, robots, we
learn about how robots work. When we do this, we see that we need
WKH¿YHFDWHJRULHVRISDUWVVHQVRUVDFWXDWRUVFRQWUROOHUVDQHQHUJ\
supply, and a body. And there is a sixth category for the support system,
which supports, and may change, the way that the robot body interacts
physically with the world. Robot bodies teach us a universal lesson about
KRZURERWVZRUN7KHUHDUHDOZD\VWUDGHRIIV7KHQHHGIRUPRYHPHQWDQG
the kinds of movement needed, drive the design of robot bodies.

Robot Parts
z There are two different methods that scientists and engineers
HPSOR\ WR ¿JXUH RXW KRZ D URERW ZRUNV DQDO\VLV DQG V\QWKHVLV
Roughly, analysis translates into taking stuff apart. Scientists dissect
and destroy in order to build a broader understanding. Synthesis
WUDQVODWHV LQWR SXWWLQJ VWXII WRJHWKHU (QJLQHHUV GHVLJQ DVVHPEOH
build, and construct.

z (YHQWKRXJKDQDO\VLVDQGV\QWKHVLVDUHGLIIHUHQWWKH\DUHHQWLUHO\
Lecture 2—Robot Bodies and Trade-Offs

FRPSOHPHQWDU\ZD\VWR¿JXUHRXWKRZVRPHWKLQJZRUNV5RERWLFV
QHHGV ERWK 1HLWKHU WKH VFLHQWL¿F QRU HQJLQHHULQJ DSSURDFK LV
EHWWHU (QJLQHHUV QHHG WKH VFLHQWLVWV¶ XQGHUVWDQGLQJ RI ZKDW WKH
bits and pieces are and how they work in order to build something.
Scientists need to have instruments and devices designed and built
by engineers in order to take new stuff apart, and engineers can tell
the scientists when they need to know more about the parts.

z 7KHSDUWVRIDURERWFDQEHRUJDQL]HGLQWR¿YHIXQFWLRQDOFDWHJRULHV
WKDW DOO URERWV KDYH VHQVRUV DFWXDWRUV FRQWUROOHUV DQ HQHUJ\
supply, and a body.
ż Sensors detect changes in the world. For example, Roomba
has a bump sensor, which can detect when the robot touches
something.

18
 ‹(NDWHULQD0LQDHYDL6WRFN(GLWRULDO7KLQNVWRFN
iRobot’s Roomba has sensors, actuators, a controller, a battery, and a chassis
that enable it to vacuum independently.

ż Actuators create movement. Anything with a motor is an

DFWXDWRU 7KH DFWXDWRUV IRU 5RRPED IDOO LQWR WZR FDWHJRULHV
actuators that move Roomba and actuators that spin brushes or
fans to create movement of the dirt.

ż The controller is an electronic computer and the software

loaded onto the memory of that computer. The controller links
and coordinates the functions of the sensors and the actuators.
The controller is located on Roomba’s motherboard.

ż The energy supply is the battery. And the battery gets recharged
at the home base, so even though the home base isn’t part of
the robot, it is part of the energy supply chain for Roomba.

ż 7KHGXVWELQ¿OWHUDQGWRSDQGERWWRPFRYHUVRI5RRPEDDUH
parts of the body. They aren’t sensors, actuators, controllers, or
the energy supply. The chassis is the main frame of the body.

19
 z Sometimes, these anatomical categories are a bit fuzzy. For
H[DPSOH VRPHWLPHV LW¶V KHOSIXO WR WKLQN RI WKH ERG\ DV QRW MXVW
anything that isn’t a sensor, an actuator, a controller, or the energy
supply but as all of the parts put together—the whole robot. And
roboticists sometimes switch between these different meanings
ZLWKRXWWHOOLQJ\RXIRUH[DPSOHZKHWKHU³ERG\´LVWKHZKROHURERW
RUMXVWWKHFKDVVLVDQGUHODWHGSDUWV

z (LWKHU ZD\ WKHUH¶V HYHQ PRUH WKDQ WKH ZKROH URERW WR FRQVLGHU
There are some other items that get sold with Roomba, for example,
that are not part of Roomba’s body in either sense of the word.
These are parts of what we call the robotic system, including the
home base, which charges Roomba’s battery; the virtual wall
lighthouse, which communicates with Roomba about where to
go and where to avoid; and the wireless command center, which
is the remote control. The support system is really a sixth kind of
component for an autonomous robot.

z As soon as the human remotely controls the robot, it’s no longer

autonomous; a human is in the functional loop, making decisions
in place of the robot’s controller. In fact, a human is almost
always in the background, maybe only in the deep background, as
another part of a robot’s support system. When and how a human
Lecture 2—Robot Bodies and Trade-Offs

enters the control loop is always important. That’s why many

URERWV KDYH D EXLOWLQ DELOLW\ WR GR ERWK EH DXWRQRPRXV DQG EH
remotely controlled.

Trade-Offs
z As with any animal or robot, evolving or designing it for a
particular task involves trade-offs. You can’t do everything well.
6R ZKLOH 5RRPED GRHV LWV MRE²FOHDQLQJ ÀRRUV UHDOO\ ZHOO HYHQ
WUDQVLWLRQLQJIURPEDUHÀRRUVWRFDUSHW²LWLVQRWEXLOWIRUPRYLQJ
up and over, or down and over, steps or drops. In fact, it has cliff
detectors to keep it from tumbling down stairs. The trade-off is that
it can clean but not climb.

20
z You might think that this is not really a problem; it’s not a trade-
off if Roomba never encounters stairs or other obstacles. But many
domiciles have stairs that lead to other places you might want
Roomba to clean. The short-term solution is to get a Roomba for
HYHU\ÀRRURI\RXUKRXVHEXWWKDW¶VH[SHQVLYHDQGLWVWLOOOHDYHVWKH
stairs uncleaned.

z With stairs and other obstacles in mind, iRobot, the maker of

5RRPED KDV GHVLJQHG D GLIIHUHQW NLQG RI URERW 3DFN%RW  ,W
has a very different body when compared to Roomba. PackBot has
WUDFNVLQVWHDGRIZKHHOVDQGLWKDVDFRPSRXQGWUDFNZLWKDMRLQW
in the middle that lets it get up stairs. The design of PackBot’s body
converts stairs into an elevated track. However, PackBot is too large
to clean under furniture. So, the trade-off for PackBot is that it can
climb but not clean—the opposite of Roomba.

z Unlike wheels, tracks do not get caught in the pits and valleys of
uneven terrain. Just like with stairs, PackBot’s tracks are great for
moving over debris. So, we can think about a trade-off at the level
RIWKHDFWXDWRUV:KHHOVFDQEHVPDOODQGWDNHXSMXVWDVPDOOSDUW
of the body, while tracks tend to be larger and take up a large part
of the body. So, one trade-off for these actuators is between size
and traversability.

z We see trade-offs like this in other robots; trade-offs are built into
any design. With drones, or aerial robots, the body’s actuation
system has propellers, and the body is very lightweight. So, it is a
V\VWHPEXLOWIRUÀ\LQJ

z A quadcopter RU TXDGULFRSWHU KDV IRXU SURSHOOHUV 4XDGFRSWHU

GURQHV DUH KLJKO\ PDQHXYHUDEOH7KH\ FDQ KRYHU ÀLS DQG WZLVW
with ease. That’s why a drone can be so good for taking aerial
pictures. It may even have a whole cinematography mode
built in.

21
 ‹LYDQVPXNL6WRFN(GLWRULDO7KLQNVWRFN
Drones are useful in aerial photography due in part to their maneuverability.

z &RPSDUHG ZLWK ¿[HGZLQJ DLUFUDIW²ZKDW ZH FDOO DLUSODQHV²

helicopters suffer in terms of energy requirements. A plane
generates lift by moving its wings forward, while a helicopter uses
propellers. For a robot body designed as a helicopter or quadcopter,
Lecture 2—Robot Bodies and Trade-Offs

WKHWUDGHRIILVPDQHXYHUDELOLW\YHUVXVHI¿FLHQF\

Robotic Manipulators
z We can see differences in robot bodies that relate to the
environment. Terrestrial robots have wheels and tracks. Aerial and
aquatic robots usually have propellers. Robots that move quickly
must be streamlined, whereas slow-moving robots will have a shape
determined by other considerations, such as how to move through
FRQ¿QHGVSDFHV

z 6R IDU WKH IRFXV KDV EHHQ RQ GLIIHUHQFHV LQ ORFRPRWLRQ WUDFNV
ZKHHOVSURSHOOHUVDQGÀLSSHUV)RUPRELOHURERWVZKLFKDUHEXLOW
to move themselves around, the needs of that mobility drive the
design of the body.

22
z But there are other robots, called static robots, stationary robots,
RU PDQLSXODWRUV WKDW VWD\ LQ SODFH DQG PRYH REMHFWV UDWKHU WKDQ
themselves. This is typically a robotic arm, so the movements of the
arm dictate the design of the body.

z Baxter is a type of robotic manipulator that has two arms and can
GRDW\SHRIWDVNWKDWLVYHU\LPSRUWDQWLQPDQ\LQGXVWULDOVLWXDWLRQV
SLFNXSREMHFWVDQGSODFHWKHPVRPHZKHUHHOVH7KLVLVFDOOHGpick
and place3DFNLQJXQSDFNLQJVRUWLQJDQGVXSSO\LQJDUHDOOMREV
that involve picking and placing. Picking and placing is what we
KXPDQVGRZKHQZHUHDFKIRUJUDVSDQGPRYHDQREMHFW

z 9HU\ ¿QH DQG DFFXUDWH SLFNLQJ DQG SODFLQJ LQ WKUHH GLPHQVLRQV
UHTXLUHVPDQ\MRLQWV)RU%D[WHUHYHU\UHDFKDQGHYHU\PRYHPHQW
RI DQ REMHFW LQYROYHV WKH FRQWUROOHG PRWLRQ DQG FRRUGLQDWLRQ RI
WKH MRLQWV )RU URERWLFLVWV WKH KXPDQ DUP KDV EHHQ D VRXUFH RI
LQVSLUDWLRQ,IDMRLQWLVDKLQJHMRLQWVXFKDVDQHOERZLWFDQRQO\
EHQGLQRQHSODQH,QHQJLQHHULQJWHUPVZHVD\WKDWWKHHOERZMRLQW
has one degree of kinematic freedom.

z 7KH VKRXOGHU MRLQW KRZHYHU LV D EDOO MRLQW DQG LW FDQ HOHYDWH
and depress, abduct and adduct, and supinate and pronate very
FRPSOH[ PRYHPHQWV ZLWK D VLQJOH MRLQW 7KH VKRXOGHU MRLQW KDV
three degrees of freedom; each type of rotation operates at 90
degrees to the other.

z :KHQ ZH WKLQN DERXW WKH MRLQWV LQ WKH KXPDQ ERG\ ZH DOPRVW
DOZD\V WDON DERXW URWDWLRQ %XW ZH FDQ GH¿QH PRWLRQ DORQJ D
straight line as translation. Baxter’s manipulator, or grabber, has
¿QJHUVWKDWPRYHDORQJDOLQHDUWUDFNLQRUGHUWRFORVHDURXQGDQ
REMHFW7KH\KDYHDVLQJOHGHJUHHRIIUHHGRPLQWUDQVODWLRQ

z More degrees of freedom mean more maneuverability, more space

WKDWFDQEHUHDFKHGDQGPRUHDQGGLIIHUHQWNLQGVRIREMHFWVWREH
KDQGOHG %XW WKHUH DUH WUDGHRIIV (DFK GHJUHH RI IUHHGRP QHHGV
its own motor, and more motors mean more power consumed.
Also, more degrees of freedom mean that the mathematics of

23
 coordinating the motion of the whole arm becomes more complex.
So, more degrees of freedom mean that you need a bigger and
faster controller.

z No matter whether you are looking at the body of a robotic

manipulator or a mobile robot, you always see trade-offs. Taking
a robot apart helps us see additional trade-offs. We can see how
actuators and sensors work to give certain kinds of behaviors, and
we see how, for example, actuators themselves have trade-offs. But
another way to think about trade-offs is to recognize them in the
URERWVWKDW\RXDUHEXLOGLQJ\RXUVHOI7KLVLVWKHYDOXHRIV\QWKHVLV
understanding by building.

Activity

To build a Tadro for yourself (see Lecture 5), you will need the
following electronics, all of which you can order from SparkFun
(OHFWURQLFVZZZVSDUNIXQFRP

Ŷ $UGXLQR8QRPLFURFRQWUROOHUGHY

Ŷ 86%FDEOHUWO
Lecture 2—Robot Bodies and Trade-Offs

Ŷ 9EDWWHU\KROGHUSUW

Ŷ 6HUYRPRWRUJHQHULFKLJKWRUTXHURE

Ŷ 0LQLSKRWRFHOOSKRWRUHVLVWRUVHQ

Ŷ )L[HGUHVLVWRUN:FRP

You also need a circular, plastic food container; a short length of

wooden dowel; duct tape; a hot glue gun; hot glue; and a 9V battery.

24
 Important Terms

Baxter$WZRDUPHGURERWLFPDQLSXODWRULQWURGXFHGE\5HWKLQN5RERWLFVLQ
2012 for light manufacturing tasks, featuring rapid reprogramming and safe
interactions with humans.

chassis7KHSULPDU\VWUXFWXUDOVXSSRUWV\VWHPRIDURERW¶VERG\

degrees of freedom 7KHQXPEHURILQGHSHQGHQWPRWLRQVDYDLODEOHLQDMRLQW

structure, or robot, whether the motion is linear (prismatic, translational),
DQJXODUUHYROXWHURWDWLRQDORUVSKHULFDOOLNHDEDOOMRLQW

motherboard7KHPDLQDQGODUJHVWSULQWHGFLUFXLWERDUGLQDFRPSXWHU

pick and place )XQGDPHQWDO WDVN IRU URERWLF PDQLSXODWRUV LQYROYLQJ

JUDVSLQJPRYLQJDQGWKHQUHOHDVLQJDQREMHFW

quadcopter (quadricopter)$URWDU\ZLQJHGDLUFUDIWZLWKIRXUSURSHOOHUV

Suggested Reading

Braitenberg, Vehicles, chaps. 1–3.

Pfeifer and Bongard, How the Body Shapes the Way We Think, chaps. 1–2.

Other Resources

%XJZRUNVKWWSZZZVXVVH[DFXN8VHUVFKULVWEXJZRUNV7KLVVLPXODWRU

models Braitenberg’s vehicles. Wire and rewire from sensors to motors and
see how rearranging the body creates different behaviors.

95(3KWWSZZZFRSSHOLDURERWLFVFRP7KLVURERWVLPXODWRUDOORZV\RX

to quickly test different types of bodies and see how they impact behavior
and function. Simulate a huge array of robots, including Baxter. Try the free
educational version.

25
 Questions to Consider

1. 7KLQNDERXWIXQFWLRQDOWUDGHRIIVLQWKHGHVLJQRIWKHKXPDQERG\:KDW
do we trade off for our upright, bipedal form of locomotion?

2. <RX FDQ KDYH D URERW EHKDYH GLIIHUHQWO\ MXVW E\ FKDQJLQJ LWV ERG\
and nothing else. If you want a two-eyed Tadro to swim away from a
stationary overhead light, which of the following bodies would do the
trick? Note that Tadros swim continuously and that their eyes (looking
like pies with a piece missing) can only pick up light in the direction of
the opening.
Lecture 2—Robot Bodies and Trade-Offs

Decide how the information from the two light sensors is used to
calculate where the light source is relative to the robot. How could you
convert that sensor information into an instruction for Tadro?

26
 Robot Actuators and Movement
Lecture 3

A
FWXDWRUV RI DOO NLQGV GH¿QH KRZ D URERW PRYHV KRZ LW PRYHV WKH
world, and how it can change the world by its movements. Movement
GH¿QHVZKDWDURERWLVEXWDFWXDWRUV²WKHPRWRUVDQGWUDQVPLVVLRQV
underlying robot movement—can do other things to change the world
besides move. Servomotors, with their embedded and tightly linked internal
sensors, steer us toward an understanding that actuators need sensors. For an
autonomous robot, all of its motors, and all of the actuators that they move, are
responding at some level to information provided by sensors. Many machines
move, but autonomous robots need both movement and sensors.

Actuators
z Robots are machines that move with purpose, to achieve their goals
DQGWRJHWZRUNGRQH7KH\PRYHREMHFWV7KH\PRYHWKHPVHOYHV
Roomba, the home robot from iRobot, moves itself to clean the
ÀRRU5RRPEDLVDPRELOHURERWWKDWPRYHVDQGPDQHXYHUVXVLQJ
two wheels that are independently motorized. By independently
controlling the speed of each motor—and, hence, each wheel—
Roomba is able to go forward, reverse, and turn in what is called
differential steering.

z Baxter, the manufacturing robot from Rethink Robotics, moves

REMHFWVWRGRWKLQJV%D[WHULVDURERWLFPDQLSXODWRUEXLOWWRZRUN
alongside humans in a manufacturing environment. For any task
you might design for Baxter, the key is that Baxter moves its
ERG\LQRUGHUWRPRYHREMHFWV%D[WHU¶VPRYHPHQWVDUHFUHDWHGE\
PRWRUV%D[WHUKDVPDQ\GLIIHUHQWMRLQWVHDFKRIZKLFKLVSRZHUHG
E\LWVRZQPRWRU(DFKPRWRUHLWKHULVDQDFWXDWRURULVSDUWRIDQ
actuator. It’s the full actuator that moves the robot.

z ,Q URERWV ZH WDNH WKH FRPSOH[ PRWLRQ RI D VLQJOH MRLQW DQG
decompose it into its fundamental elements. For the shoulder, three-
GLPHQVLRQDOPRWLRQWDNHVWKUHHVHSDUDWHMRLQWVHDFKUHVWULFWHGWRD

27
 © Rethink Robotics, Inc.
Baxter is a robotic manipulator built to work alongside humans in a
manufacturing environment.

single degree of freedom—in this case, angular motion in a plane

that can be described by a single variable, such as the angle between
Lecture 3—Robot Actuators and Movement

the skeletal elements.

z 5HWKLQN5RERWLFVFDOOVMRLQWVZLWKDKLQJHDEHQGMRLQWEHFDXVHLW
FKDQJHVWKHDQJOHRIWKHHOHPHQWVWRZKLFKWKHMRLQWLVDWWDFKHG7KH
RWKHU NLQG RI MRLQW WKH\ FDOO D WZLVW MRLQW EHFDXVH WKH LPPHGLDWH
elements don’t change orientation relative to each other.

z :H FUHDWH WKUHH GLIIHUHQW MRLQWV²WZR WZLVW MRLQWV DQG RQH EHQG
MRLQW²LQWKHVKRXOGHUIRUWZRUHDVRQV)LUVWPRVWPRWRUVDUHEXLOW
to power angular rotation with a single degree of freedom. For
example, when we connect a direct current (DC) electric motor
to an energy supply, we create an electric circuit, and the motor
URWDWHV7KHPRWRUVWKDWSRZHU%D[WHU¶VMRLQWVDUHDVSHFLDONLQGRI
DC electric motor called a brushless servomotor, which includes a
sensor so that the motor’s motions can be carefully controlled.

28
z 7KHVHFRQGUHDVRQWRFUHDWHWKUHHVHSDUDWHMRLQWVLVFRQWURO:KHQ
we decompose complex three-dimensional motions into separate
two-dimensional ones, then the motions are easy to control and
FRRUGLQDWHZKHQZHOLQNXSWKHPRWLRQVRIPXOWLSOHMRLQWV

z ,Q%D[WHU¶VVKRXOGHUHDFKSODQHRIHDFKMRLQWLVQHDUO\DWGHJUHHV
ZLWKUHVSHFWWRDQRWKHUMRLQW6RLIZHZDQWWRFRQWUROWKHSRVLWLRQ
of the end effector—the hand—at the tip of the arm, then we have
MXVWWKUHHPRWRUVWRRSHUDWH

z Compare that with your shoulder, which has more than 10 different
muscle groups, depending on how you count them, and every
muscle group has its own innervation, or neural control, that is
independent and has to be coordinated with other muscles. Baxter’s
VKRXOGHULVDQHOHJDQWVROXWLRQWKUHHGHJUHHVRIIUHHGRPLQURWDWLRQ
and full three-dimensional motion.

z %D[WHU¶VJUHDWUDQJHRIPRWLRQLVPDGHSRVVLEOHE\WKHFRQ¿JXUDWLRQ
RILWVMRLQWVDQGWKHPRWRUVWKDWGULYHWKHP:KDWPDNHV%D[WHUVDIH
for humans to work with is that it has compliant actuators. The
motors transmit their force to the segments of the arms through a
transmission system that has springs.

z 7KHVH VSULQJV PHDQ WKDW WKH MRLQWV JLYH ZD\ ZKHQ WKH DUP KLWV D
human or runs into something. Also, we can back-drive the motors,
RUSXVKWKHMRLQWVEDFNZDUGDQGJHWWKHPRXWRIRXUZD\)LQDOO\
Baxter moves at speeds that don’t create too much kinetic energy.
Combine that with Baxter’s relatively lightweight arms and
padding, and it is a safe companion for side-by-side cooperation.

z The end effector and the motor are connected by the transmission—
gears and shafts that not only transfer the motion from the motor
to the end effector but also further transform it by changing
the motion’s speed, leverage, and torque. Together, the motor,
transmission, and end effector make an actuator.

29
 Automatons
z An automaton is a machine with a hidden mechanism that operates
automatically. Centuries ago, even a clock might be called an
automaton, but by the 18th and 19th centuries, an automaton came to
PHDQDQ\PHFKDQLFDO¿JXUHWKDWVLPXODWHVWKHPRYHPHQWVRIOLYLQJ
beings, especially humans.

z But while their movements may be lifelike, these machines aren’t

DXWRQRPRXV LQ D URERWLF VHQVH 7KH\ GRQ¶W XVH VHQVRUV WR UHDFW
to changes in the world, and they don’t use controllers to change
their behavior in response to sensors. They are automatic in the
sense that you turn them on, they move, and they run through
their routine.

z The springs in toys, automatons, and mechanical clocks are very

useful motors, but they have to be wound up regularly. In terms
of physics, when we wind up or load a spring, we are adding
potential energy to the machine by using the muscular energy
RI RXU KXPDQ ERGLHV (QHUJ\ WUDQVIHU LV DOO DURXQG XV DOO WKH
time, and the trick, for robots, is to put the energy in a form that
creates movement.
Lecture 3—Robot Actuators and Movement

z When it comes to creating movement in robots, probably the most

important kind of motor is the electric motor. And all motors work
RQDVLPLODUSULQFLSOH0RWRUVXVHHOHFWULFLW\WRVSLQPDJQHWV

Electric Motors
z Batteries are a common source of energy for actuators in robots, so
it’s important to understand how batteries work. When we connect
the poles of the battery to the wires of the motor, the DC motor and
the battery create a physical connection called an electric circuit.
The name circuit is used because of the idea of tiny electric particles
traveling in a circuit from one end of the battery, through the motor,
and back to the other end of the battery.

30
z The particles are charges in two different ways, positive or negative,
and opposites attract. You put a bunch of positive charges on one side
of the circuit and let them run to a negative attractor on the other side
of the circuit. This attraction of opposites is the force that drives the
ÀRZRIHOHFWULFSDUWLFOHVDÀRZWKDWZHFDOOWKHHOHFWULFFXUUHQW

z ,QD'&PRWRUWKHHOHFWULFFKDUJHVÀRZGLUHFWO\IURPRQHSROHRI
the battery, through the motor, to the other pole of the battery. In
contrast, in an alternating current (AC) motor, the charges vibrate—
or alternate—back and forth rather than traveling like water in a
pipe, which describes a direct current.

z Batteries contain two different chemicals. The battery is designed

to keep those chemicals physically apart. You can think of one
chemical living at the negative end and the other chemical living
at the positive end. One of those chemicals provides the charged
particles, and the other chemical collects them. And they exchange
those particles through the electric circuit.

z Voltage is a way to measure the strength of the attraction between

the different chemicals in the battery. The higher the voltage, the
greater the attraction. And the greater the attraction, the faster the
ÀRZRIWKHSDUWLFOHVWKURXJKWKHFLUFXLW

z When a fully charged battery is not part of an active electric circuit,

we say that it has potential energy—that is, the separation of the
charges gives the battery the potential to make the circuit work. As
soon as we turn the circuit on by throwing the switch or connecting
the wires, then we turn the potential energy into actual energy as the
charged particles get moving.

z In other words, when we start to use the potential energy, it isn’t

potential anymore; it’s the energy in the motion of the charged
particles. That electric current, the movement of charge, causes the
motor to move.

31
 z (OHFWULFPRWRUVKDYHWZRPDLQSDUWV7KHSDUWWKDWVSLQVLVFDOOHG
the rotor, and the part that stays still is called the stator. An electric
PRWRU ZRUNV E\ FUHDWLQJ PDJQHWLF ¿HOGV RI FKDUJHG SDUWLFOHV WKDW
VSLQ WKH URWRU 7KH VWDWRU FUHDWHV D PDJQHWLF ¿HOG DQG WKH URWRU
GRHVWRR$QG\RXFDQUHYHUVHRQHRIWKHPDJQHWLF¿HOGVEDFNDQG
forth so that you can pull and then push the rotor around.

z An electric motor, as part of an actuator, converts the movement of

molecules into the movement of a robot.

Servomotors
z For roboticists, one of the great innovations in electric motors has
been the servomotor. Unlike the DC motor, which has two wires,
the servomotor has three. Two of the wires are for the electric
circuit, and in fact, they go to a small DC motor inside. But the
third wire is for a control signal.

z The signal, called a pulse-width-modulated (PWM) signal, is

sent to the servomotor from a computer. The width of the pulse,
which is really its duration in time, tells the servomotor where to
position its rotor on a circular dial. And that signal is sent every
20 milliseconds—the attention to positional detail that gives
Lecture 3—Robot Actuators and Movement

servomotors their precision.

z How does the motor know when it has gotten to the right position?
The servomotor has to have a sensor. A servomotor needs feedback
from the world, and it gets that feedback from a sensor that can tell
it where it is, in a rotational sense. In the case of a servomotor, the
sensor is a potentiometer.

z A potentiometer is a variable resistor. A small version of a

potentiometer is built into the servomotor. The potentiometer
rotates with the rotor of the motor. The potentiometer then provides
feedback to the motor, letting it know where it is rotationally.

32
 z You can think of the feedback as the answer to the question of
whether the motor is in the correct rotational position. If it is, then
the motor should hold its position as long as the PWM signal tells it
to. If it is not, then the motor should try to move into that position.
Servomotors are so useful in robotics because of their precise
control of motion.

Important Terms

compliant actuators0RWRUVDQGOLQNDJHVPDGHRIÀH[LEOHVRIWPDWHULDOV

direct current (DC)$W\SHRIHOHFWULFFXUUHQWWKDWLVXQLGLUHFWLRQDO

end effector$WRRORURWKHUGLVWDOHOHPHQWRQDURERWLFPDQLSXODWRUWKHSDUW
of an actuator that interacts directly with the world.

potentiometer7\SHRIVHQVRUWKDWFRQYHUWVDPHFKDQLFDOURWDWLRQLQWRDQ
electric change in resistance; also known as a variable resistor.

pulse-width modulation (PWM) 0HWKRG RI HQFRGLQJ LQIRUPDWLRQ LQ DQ

electric signal by varying the duration of square-wave pulses.

rotor7KHSDUWRID'&PRWRUWKDWVSLQV

stator7KHVWDWLRQDU\SDUWRIDQHOHFWULFPRWRU

torque Force applied through a moment arm to rotate or twist.

Suggested Reading

Jones, Flynn, and Seiger, Mobile Robots, chaps. 6–7.

Monk, Hacking Electronics, chaps. 1–2.
Scherz and Monk, Practical Electronics for Inventors IURQW À\OHDYHV DQG
chaps. 1–2.

33
 Other Resources

*RRGVRXUFHVIRUSXUFKDVLQJDFWXDWRUVDQGHOHFWURQLFV

'LJL.H\KWWSZZZGLJLNH\FRP

0DNHUVKHGKWWSZZZPDNHUVKHGFRP

6SDUN)XQ(OHFWURQLFVZZZVSDUNIXQFRP

/HDUQ DERXW GLIIHUHQW NLQGV RI PRWRUV DW 0LFURPR ZZZPLFURPRFRP

WHFKQLFDOOLEUDU\WHFKQLFDOOLEUDU\

Questions to Consider

1. How is a servomotor different from a DC motor?

2. 7KHWKUHHPRVWFRPPRQPRWRUVWKDW\RX¶OO¿QGLQVLPSOHURERWVDUH'&
brushed motors, servomotors, and a special kind of brushless DC motor
known as a stepper motor. Which motor is the best choice for each of
the following applications?
Lecture 3—Robot Actuators and Movement

a. Power a tank track drive on a mobile navigator.

b. Precisely move and then hold position of a robotic arm.

c. Flap the tail of a Tadro.

3. Some roboticists think of an actuator very broadly as any device that

FKDQJHV WKH ZRUOG E\ DOWHULQJ SDWWHUQV RI HQHUJ\ %\ WKDW GH¿QLWLRQ
which of the following devices is an actuator?

a. light-dependent photoresistor.

b. electrical switch.

c. light-emitting diode.

34
 Robot Sensors and Simple Communication
Lecture 4

A
sensor is any device or mechanism that registers something
happening in the world and converts that event into a signal that can
be transmitted to other parts of the robot. There are two main types of
VHQVRUVSDVVLYHDQGDFWLYH3DVVLYHVHQVRUVDUHEXLOWRQO\DVUHFHLYHUVZKLOH
DFWLYHVHQVRUVDUHEXLOWWRVHQGDQGUHFHLYH)RUURERWVWKHVSHFL¿FVHQVRUV
deployed—whether single, in arrays, or of different types—determine what
they can know about the world. What they do with that information, how
they act in response to their sensor readings, marks the origin of intelligent
behavior in robots.

Sensors
z Sensors respond to many different kinds of events, everything from
changes in light to the presence of dangerous gases. In this modern
age of electronic systems, sensors convert their response into
D FRPPRQ FXUUHQF\ DQ HOHFWULF VLJQDO WKDW WKH URERW¶V FRPSXWHU
can read.

z For robots, sensors determine what they know about the world.
Because the common currency of sensors is electricity, we look
at how sensors change the resistance, current, or voltage in an
electric circuit.

z Photoresistors, also called light-dependent resistors, are sensitive

to light. Ping sensors DOVR NQRZQ DV UDQJH VHQVRUV FDQ ¿JXUH
RXWKRZIDUDZD\DQREMHFWLV,QIUDUHGSUR[LPLW\VHQVRUVFDQDOVR
detect distance.

z A potentiometer is a variable resistor that can be used to signal

rotational position. This is how it is used in a servomotor, as part of
the feedback system to allow the motor to achieve precise control
of its motion.

35
 z Accelerometers convert mechanical vibrations caused by
acceleration into an electric signal. Tiny microelectromechanical
accelerometers are common in smartphones and robots. They work
E\ FRQYHUWLQJ WKH GHÀHFWLRQV RI D WLQ\ FDQWLOHYHUHG EHDP ZLWK D
mass at the end into a change in electric voltage.

z Microphones convert sound pressure waves into electricity by

vibrating a membrane that oscillates a small magnet that induces a
FKDQJHLQDQHOHFWURPDJQHWLF¿HOG

z Digital compasses are tiny magnetometers that are sensitive to

WKH (DUWK¶V PDJQHWLF GLSROH DQG FRQYHUW WKDW ¿HOG LQWR RULHQWDWLRQ
relative to magnetic north.

Examples of Types of Sensors

z We can attach the two wires of a photoresistor to the positive and
negative leads of a device called a multimeter, which can measure
multiple features of an electric circuit or electric parts, such as the
Lecture 4—Robot Sensors and Simple Communication

YROWDJHRIDEDWWHU\WKHDPSHUDJHRIFXUUHQWÀRZRUWKHUHVLVWDQFH
of the circuit.

z If we set the multimeter to measure resistance, then if we cover

up the photoresistor, the resistance increases, but if we let light
fall on the sensor, the resistance decreases. This sensor gives a
continuously variable response to continuously variable changes in
the world.

z ,Q FRQWUDVW D WRXFK RU EXPS VHQVRU JLYHV D VLPSOH \HVQR

response to touch. We can turn this circuit on by turning on a
switch. While the touch sensor is a mechanical switch, with an on
or off position, the photoresistor creates a continuously variable
response by using the energy of photons to knock electrons onto
the band through which electricity is conducted. More light means
PRUH HOHFWURQV 0RUH HOHFWURQV PHDQ HDVLHU ÀRZ RI HOHFWULFLW\
and, hence, less resistance.

36
z To take the change in resistance and get it to produce an electric
signal for the robot, we have to incorporate the photoresistor into
an electric circuit. We can use the photoresistor to build a simple
circuit that acts as a light meter. We couple this circuit with a
computer, the Arduino microcontroller that is used in many robots.

z Because of the way that the analog electric circuit is put together,
WKH/('light-emitting diode) lights get brighter when more light
IDOOVRQWKHSKRWRUHVLVWRU,QRWKHUZRUGVWKH/('OLJKWVPXVWEH
getting more voltage supplied to them in response to the change in
resistance of the photoresistor.

z Another type of sensor makes use of the Hall effect, which is a

VLGHZD\V GLVUXSWLRQ RI WKH ÀRZ RI HOHFWULF FXUUHQW WKDW ZDV ¿UVW
discovered in 1879. That disruption may be caused by a magnet,
and it creates a change in voltage that can be detected and measured
by a Hall effect sensor. By the careful placement of magnets, Hall
HIIHFWVHQVRUVFDQEHXVHGWRPHDVXUHHOHFWULFFXUUHQWÀRZDVZHOO
as the position and velocity of motors and other moving parts.

z We can understand the importance of sensors if we think about

ZKDWOLIHLVOLNHIRUDURERWZLWKRXWDQ\VHQVRUV7KH¿UVWIDFWRU\
robots had none. They were robotic manipulators, or arms, that
would go through the motions with very high accuracy and speed,
without having any feedback. Movement without sensing is
not intelligent.

z The counterexample is all sensing and no movement. A weather

buoy, located in the ocean, is one such device. It carries a variety
of sensors. It has an anemometer, up on the tip of a pole, that
measures the speed of the wind. That same sensor can measure the
direction of the wind using a weather vane. The buoy can also be
RXW¿WWHGZLWKVHQVRUVWRPHDVXUHWKHWHPSHUDWXUHRIWKHDLUDQGWKH
water, barometric pressure, humidity, wave action, currents, water
chemistry, environmental factors, and pollution.

37
 ‹%DORQFLFLL6WRFN7KLQNVWRFN
Although robotic arms are precise and fast, they do not receive feedback and
therefore cannot be considered intelligent.
Lecture 4—Robot Sensors and Simple Communication

z :HDWKHU EXR\V PD\ ÀRDW IUHHO\ RU PD\ EH WHWKHUHG 6RODU SDQHOV
charge onboard batteries that power the sensors, the data logger,
and the communication system. Weather buoys communicate via
satellite to computers on land.

z Is a weather buoy intelligent? Movements with purpose, such as

seeking light, are the hallmark of an intelligent robot. Robots are
EXLOW WR PRYH WKHPVHOYHV RU REMHFWV :LWKRXW PRYLQJ LWVHOI RU
REMHFWV LQ UHVSRQVH WR LWV VHQVRU UHDGLQJV WKH ZHDWKHU EXR\ LVQ¶W
intelligent. To be fair, it isn’t a robot, either.

Bump Sensors
z 7XWHERWLVDURERWWKDWKDVDYHU\VLPSOHW\SHRILQWHOOLJHQFHIROORZLQJ
walls. As soon as you turn Tutebot on, the action starts. Tutebot hits a
wall. Then, Tutebot backs up and turns. Then, it goes forward. It hits
the wall again. Tutebot is working its way along the wall. It is doing a
behavior that we see in robots called wall following.

38
 z Tutebot has a bump sensor, which is made up of two spring-
ORDGHG SXVK VZLWFKHV 3XVK HLWKHU DQG WKH VDPH WKLQJ KDSSHQV
The motorized wheels reverse direction momentarily. This is a
case where the switch of a bump sensor is switching the current
from running in one direction through the motor to running in the
opposite direction.

z Rotation of what we call a direct current motor depends in part

RQWKHGLUHFWLRQRIWKHFXUUHQWÀRZWKURXJKLWVFRLOV5HYHUVHWKH
direction of the current by switching the positive and negative poles
in the connection to the battery, and the motor reverses its rotation.

z You can think of the bump sensor as asking whether Tutebot has
run into something. The answer is either yes or no. We can also see
simple switches working as bump sensors in a sophisticated robot
like Roomba. Roomba’s bump sensor is really an array of 11 separate
sensors. By having so many, Roomba can know where on its front
EXPSHULWKDVKLWDQREMHFWDQGWKHQWXUQLQWKHRSSRVLWHGLUHFWLRQ

z When Roomba is in action, if it bumps on the right side of its body,

it backs up and turns to the left. If it bumps on the left side, it backs
up and turns to the right. For Roomba, the question this sensor array
LVDVNLQJLVPRUHVRSKLVWLFDWHG+DYH,UXQLQWRVRPHWKLQJDQGLI,
have, where, in relationship to my body, is it?

Active Sensors
z Sparki, made by ArcBotics, does not have bump sensors. To
DYRLG REMHFWV 6SDUNL KDV DQ XOWUDVRQLF UDQJH ¿QGHU VRPHWLPHV
called a ping sensor, and it uses this sensor as a kind of remote
REMHFWGHWHFWLRQ V\VWHP 7KLV VHQVRU ZRUNV LQ WKH VDPH ZD\ WKDW
echolocation does in dolphins and bats.

z %DWVÀ\DWQLJKWDQGKXQWPRWKV7KH\FDQ¶WVHHVRLQRUGHUWRFKDVH
GRZQDPRELOHLQVHFWDQGDYRLGÀ\LQJLQWRWUHHVDQGEXLOGLQJVWKH\
basically shout out and listen for the echo of their shout. We don’t
hear bats do this because the frequency of the sound is much higher
than our human ears can hear—ultrasound.

39
 z The ping sensor works the same way, with ultrasound. One side of
the sensor is the voice box, a speaker that produces the sound. The
other side is the ear, the receiver.

z Using sound for active sensing in engineered systems has been

around for 100 years. Sonar, which is short for sound navigation
ranging, was originally developed for sensing underwater, but now,
it is used in air, too, such as in a pinger for a robot.

z There are both passive and active sonar systems. The passive
VRQDU V\VWHP MXVW OLVWHQV 7KH DFWLYH RQH HFKRORFDWHV 7KH DFWLYH
system, such as the one in Sparki, is very common in robotics. The
passive system is basically a microphone. In fact, some robots have
microphones, and they will respond to a clap or a loud noise.

z ,QSUDFWLFHLW¶VGLI¿FXOWWREXLOGDURERWWKDWPDNHVYHU\LQWHOOLJHQW
use of a simple microphone. The problem is that the world is really,
UHDOO\QRLV\6R¿QGLQJDVLJQDOWKDWPHDQVVRPHWKLQJWRWKHURERW
Lecture 4—Robot Sensors and Simple Communication

LVGLI¿FXOW

z You can overcome the noise problem by broadcasting your own

signal, at a particular wavelength, and listening for its return. Active
sensors—send and receive sensors—are very popular in robotics,
VXFKDV6SDUNL¶VXOWUDVRQLFUDQJH¿QGHU

z Another very common sensor in robotics that works in the same

ZD\ DV DQ XOWUDVRQLF UDQJH ¿QGHU LV DQ LQIUDUHG ,5 SUR[LPLW\
detector. Infrared refers to a place on the electromagnetic spectrum.
The electromagnetic spectrum includes what we call visible light—
visible to humans, that is. Just outside that range, as measured by
WKH ZDYHOHQJWK RI WKH OLJKW LV WKH LQIUDUHG ³OLJKW´ 1LJKWYLVLRQ
VFRSHVXVHLQIUDUHGWR³VHH´LQWKHGDUN

z An IR proximity detector has an IR emitter and an IR receiver.

6LPLODUWRDQXOWUDVRQLFUDQJH¿QGHUDQ,5SUR[LPLW\GHWHFWRUVHQGV
out a signal and then waits for the echo. With light, we call the echo
WKHUHÀHFWLRQ

40
Communication
z One way to think about active sensors working is that they are
communicating with the world. They send out a signal and wait
for its return. The only difference in practice is that for what
we normally call communication, the return signal comes from
another person.

z We normally think about communication using radio waves,

including ultrahigh-frequency radio waves such as Wi-Fi and
Bluetooth. But any part of the electromagnetic spectrum can be
XVHGQRWMXVWUDGLRZDYHV

z Microwaves have a smaller, tighter wavelength that is good for

sensors to detect motion, such as motion detectors on a house or so-
FDOOHG³UDGDU´JXQVXVHGE\SROLFH

z Infrared is good for even shorter distances—for example, inside a

single room—if you have a clear line of sight. For a traditional TV
UHPRWHWRZRUN\RXKDYHWRDLPLWULJKWDWWKH797KDW/('OLJKW
\RXVHHLVDOVRDQLQIUDUHG/('EXOE

z Visible light has even shorter wavelengths, making it great for

communication using tightly focused laser beams but also for wide-
beam communication in swarms, too. Visible light can convey
information in more complex ways, too. For example, consider the
spatial and temporal patterns of light recorded by a camera.

z 7KLV TXLFNO\ JHWV LQWR D ZKROH ¿HOG LQ FRPSXWHU VFLHQFH DQG
DUWL¿FLDO LQWHOOLJHQFH PDFKLQH YLVLRQ &DPHUDV DUH D JUHDW
example of a passive sensor that provides a large amount of
FRQWLQXRXV LQIRUPDWLRQ %XW WKH IDFW WKDW WKHUH LV D ZKROH ¿HOG
of machine vision should tell you that this gets complicated
very quickly.

41
 Important Terms

accelerometer $ VHQVRU WKDW FRQYHUWV FKDQJHV LQ YHORFLW\ LQWR HOHFWULF

signals that can be read by the controller.

echolocation$FWLYHVHQVLQJLQZKLFKDQDQLPDORUURERWEURDGFDVWVDVRXQG
senses the returned echo, and uses the difference between the signal and its
echo to calculate the distance and composition of the target. See sonar.

Hall effect sensor$VHQVRUWKDWFRQYHUWVFKDQJHVLQDQH[WHUQDOPDJQHWLF

¿HOGLQWRFKDQJHVLQDYROWDJHRURWKHUHOHFWULFRXWSXW

light-emitting diode (LED)6HPLFRQGXFWRUEDVHGOLJKWVRXUFH

multimeter $Q LQVWUXPHQW XVHG WR PHDVXUH HOHFWULF YROWDJH UHVLVWDQFH

continuity, and current.

photoresistor7\SH RI VHQVRU WKDW FRQYHUWV FKDQJHV LQ OLJKW LQWHQVLW\ LQWR
Lecture 4—Robot Sensors and Simple Communication

changes in electric resistance.

ping sensor An ultrasonic UDQJH¿QGHU

Suggested Reading

Jones, Flynn, and Seiger, Mobile Robots, chap. 5.

Karvinen and Karvinen, Make, chaps. 1–2.

Other Resources

7LPHWRJHW\RXURZQ$UGXLQRVWDUWHUNLWKWWSVWRUHDUGXLQRFFFDWHJRU\

42
 Questions to Consider

1. The following robot has a bump sensor on its right and left front corners.
Only the left sensor is activated when the robot hits the wall.

ll bump
wa sensors

Which of the following statements best describes how a small Roomba-

like robot might react?

,)WKHOHIWEXPSVHQVRULVWXUQHGRQ7+(1BBBBB(/6(BBBBB

a. then keep moving forward, else reverse direction and turn right.

b. then reverse direction and turn right, else keep moving forward.

c. then reverse direction and turn left, else keep moving forward.

43
 2. (DFKVHQVRUDVNVDTXHVWLRQ7KHVDPHVHQVRUFDQDVNGLIIHUHQWTXHVWLRQV
depending on how it is situated on the body, what other sensors it’s
working with, and what task the robot is trying to accomplish.
Lecture 4—Robot Sensors and Simple Communication

This is a Tadro with two light-dependent sensors. We use those two

VHQVRUVWRDVN:KDWLVWKHGLUHFWLRQRIWKHOLJKWUHODWLYHWR7DGUR"7KLV
is also called the bearing of the light. How do the two sensors work
together to give an answer about the bearing of the light?

44
 Robot Controllers and Programming
Lecture 5

A
controller is a special kind of computer whose program modules
give autonomy to a robot. The biggest questions in robotics often
revolve around what combination of autonomous self-control and
remote control by humans is most appropriate for a given situation or robot.
Both in the design and operation of a robot, the controller is how any robot’s
DXWRQRP\LVFUHDWHGDQGGH¿QHG²LQWKHOLQNLQJRIVHQVRUVZLWKDFWXDWRUV
The controller is where a robot’s autonomy is overruled by humans or
extended even further.

Controllers
z Self-control is what turns a remotely controlled machine into
an autonomous robot. Where does that self-control come from?
Sensors collect information that is coordinated and used to tell the
actuators what to do next. These sensor-guided movements depend
on a controller, the part of the robot that makes sense of the sensory
information and then decides how the robot is going to react. A
robot controller or microcontroller gives control, or autonomy, to
the robot.

z ,Q WKH YDVW PDMRULW\ RI PRGHUQ URERWV WKH FRQWUROOHU LV D
computer that is specially designed to take as input sensor data
and communications from a human operator. The controller is
specially designed to produce as output signals for actuators and
communications for a human operator. Because the controller runs
computer code, it is programmable.

z To understand the controller, and how it works with sensors and

actuators, we can build a robot using the simplest sensor-actuator
circuit. Tadro is a robot with only one sensor and only one actuator.
And we can use a class of microcontrollers based on Arduino.

45
 z Let’s start with the basic model, the Arduino Uno. You can buy one
for about 50 dollars. Arduino uses free software that is available
at Arduino.com. Both the Arduino hardware and the software are
open-source systems, which means that you can copy, modify,
and share your own software and hardware versions for free and
without violating any patents or trade secrets.

z $PLFURFRQWUROOHUOLNH$UGXLQRLVPRUHWKDQMXVWDPLFURSURFHVVRU
A microcontroller is like an entire computer on a single chip,
because it also includes inputs, outputs, a read-only memory,
and buffers.

z We can get our sensor and actuator working together by way

of the Arduino. The sensor is a light-dependent photoresistor, a
very simple sensor that changes its resistance to electric current
depending on how much light is falling on it.

z In addition to the input ports, where you connect sensors, Arduino

also has output ports. We’re interested in the ports through which
we can power and control a servomotor.
Lecture 5—Robot Controllers and Programming

z Once we have those plugged in, we need the code to run it. We’re
going to use a program called Knob that can be found on the
Arduino website. We connect the photoresistor to the 5V power port
and the A0 analog input port. Then, we connect the servomotor to
the 5V power port, the ground, and digital port 9. The servomotor’s
shaft rotates.

z We have compiled and loaded the KNOB program onto the

$UGXLQR 7KH /(' OLJKWV ÀDVK WR OHW XV NQRZ WKDW ZH KDYH
communication. If the position of the servomotor changes as we
shine a light on the photoresistor, then we have a connection from
the sensor to the actuator. As the world changes, the sensor sends
a signal to the microcontroller that is then converted to a signal for
the actuator.

46
z How can we use this to build an autonomous robot with one sensor
and one actuator? The actuator behavior we want is for the Tadro to
DOZD\VEHFUXLVLQJDURXQGÀDSSLQJLWVWDLODWDFRQVWDQWIUHTXHQF\
Where the sensor comes in is that the tail turns, while it keeps
ÀDSSLQJZLWKWKHGLUHFWLRQDQGDPRXQWRIWXUQLQJSURSRUWLRQDOWR
the light on the photoresistor.

z +RZ GR ZH JHW D EHKDYLRU OLNH ÀDSSLQJ" :H FDQ JR EDFN WR WKH
Arduino website. There is a sample program called Sweep, by
BARRAGAN studios, that moves the servomotor back and forth.
2QFHZHORDGWKLVSURJUDPWKHVHUYRPRWRUZLOOMXVWVZLQJEDFNDQG
forth without paying attention to the sensor, because it is open loop.

z 1H[W ZH QHHG WR FRPELQH WKH 6ZHHS SURJUDP IRU ÀDSSLQJ ZLWK
the Knob program for turning. The trick is to add the servomotor
position from the Knob program—think of this as the turning
DQJOH²WRWKHQH[WÀDSSLQJSRVLWLRQWKDWWKH6ZHHSSURJUDPZDQWV
to go to.

z To do this, we compile and download the program and observe

ZKDWKDSSHQV7KHVHUYRPRWRULVÀDSSLQJ:KHQZHVKLQHDOLJKW
on the photoresistor, the servomotor adds the offset, a turning angle,
WR LWV ÀDSSLQJ 1RZ ZH KDYH D VHQVRUDFWXDWRU FLUFXLW WKDW ZLOO
allow Tadro to swim and to react to light.

z For the body, we’re going to put the Arduino into a storage bowl
and use duct tape to secure the photoresistor on the edge. The
VHUYRPRWRUQHHGVVRPHWKLQJWRÀDSVRZH¶UHJRLQJWRXVHDSHQFLO
as the motor shaft, tape it to the servomotor, and then create a tail
with duct tape that will work underwater. We’re going to tape the
VHUYRPRWRUWRWKHERZO:H¶YHMXVWEXLOWDURERWIURPVFUDWFK

z As we move the light, Tadro slowly spirals its way toward the light.
:H¶YH EXLOW D URERW ZLWK OLJKWVHHNLQJ EHKDYLRU 7KLV VKRZV MXVW
how simple self-control can be.

47
 z We can use this simple sensor-actuator circuit to point us in
the direction of bigger, better, and more complex kinds of self-
controlled robotic behavior. When we think about the purposeful
movement of the whole Tadro, we can now put the sensory-actuator
circuit into its place in the functional loop of an autonomous robot.

z The movement of the Tadro changes the amount of light that falls
on the sensor, and the whole functional loop starts all over. Sensor
reading changes actuator movement; actuator movement changes
sensor reading. Because the conversion processes that take place
on board the robot give the robot self-control, it makes sense to call
this part of the system the controller.

Shakey
z 2QHRIWKH¿UVWURERWVWKDWKDGDGLJLWDOFRPSXWHUIRUDFRQWUROOHU
was Shakey, which was a mobile robot built by researchers led
by Charles Rosen at the Stanford Research Institute from 1966 to
$WWKHWLPHGLJLWDOFRPSXWHUVZHUHODUJHDQGKHDY\7KH¿QDO
YHUVLRQRI6KDNH\XVHGD'LJLWDO(TXLSPHQW&RUSRUDWLRQ3'3
PDP-15 mainframe.
Lecture 5—Robot Controllers and Programming

z The PDP-10 looms large in hacker lore because it helped get

the Internet to take baby steps as part of the Internet’s precursor,
$53$1(7,WZDVYHU\DGYDQFHGIRUWKHWLPHDQGZDVXVHGDWRQO\
DIHZXQLYHUVLWLHV%XWWKH3'3FRXOGQHYHU¿WRQWRDURERWWKH
way an Arduino can now. So, to serve as the controller of Shakey,
the PDP-10 had to be in constant radio communication in order to
be in Shakey’s sensory-actuator loop.

z :KLOH 6KDNH\ ZDV UHYROXWLRQDU\ LQ SXWWLQJ DUWL¿FLDO LQWHOOLJHQFH

into a sensor-actuator circuit, the idea of using radio waves to
communicate with a mobile vehicle was not. The concept of remote
control was patented in 1898 by Nikola Tesla. In his patent, Telsa
GHVFULEHVZKDWKHFDOOHGD³WHOHDXWRPDWRQ´ZKDWWRGD\ZHFDOOD
radio-controlled or remote-controlled vehicle, and he demonstrated
his idea at Madison Square Garden with a four-foot boat that he
controlled using a telegraph signal box.

48
 z Tesla did not discover radio waves or invent the devices to produce
them. But he did build actuators that could be controlled remotely.
Propulsion and steering were his targets, and he demonstrated a
remotely controlled boat to the public.

z What we see in many different kinds of robots, by contrast, is neither

Shakey nor the Tesla boat, but something we call partial autonomy
and partial remote control. The robot has some autonomy, and a
human operating remotely shares control with the robot.

z Sparki has a remote control that works a lot like a TV remote

controller, sending a few simple signals to its onboard computer.
We can signal Sparki to move forward, turn right, or turn left.

z Roomba, too, has a remote control that works in the same way, with
an infrared signal from the controller to the same sensor on Roomba
that receives the signal from the docking station.

z So, you can also see why old-fashioned remote control is still
with our modern robots even though they have the capacity to
control themselves. By allowing humans to insert themselves into
the functional loop, we can take control, as needed. This involves
communication between robot and human.

Behavior-Based Robotics
z We can treat sensor-motor circuits as discrete building blocks,
where each block is a behavior module. These modules can function
independently. They may function together, their coordination
decided upon by the controller.

z This focus on behavior modules driving thanks to sensor-actuator

circuits was developed and made famous by Rodney Brooks when
he was a professor at MIT. This approach goes by the name of
behavior-based robotics.

49
 z The hallmark of behavior-based robotics is that the controller
makes decisions about which behavior module gets to operate
when. Because it’s an all-or-none proposition, we think about this
as the behavior modules competing to have the controller decide
to let it operate, a winner-takes-all scenario. There is no sharing of
the actuators; you either are the behavior that is operating or you
are not.

z Rodney Brooks very quickly realized that the great value of behavior-
based robotics is that you can use very simple programming, such
as what we did with the Arduino, to get robust self-control out of
your robot.

z Roomba is one of Brooks’s creations. Roomba has several sensors,

such as cliff detectors to signal when Roomba is going over the
edge of the stairs. Another infrared sensor on top of Roomba is an
infrared detector that is paired with the infrared emitter on the top
of the recharging station. Together, they give Roomba the ability to
¿QGLWVZD\EDFNKRPHWRVOHHSDQGWRUHFKDUJH
Lecture 5—Robot Controllers and Programming

z What happens when we put them together? How do the modules

interact? What decisions is the controller making? We can do pair-
wise comparisons to see which behavior is programmed to win
in which situation. For example, cliff detection and avoidance
overrule docking. We are always comparing our current perception
RIWKHVWDWHRIWKLQJVZLWKRXUGHVLUHGVWDWHRIWKLQJVRXUJRDO$UH
we there yet? If not, then keep going. If so, then stop.

z This comparison of your desired state to your actual or goal state

is at the heart of self-control. The general principle is what’s called
negative feedback control—one of the most important principles
that governs the control and regulation of mechanical or organic
V\VWHPV,QIDFWWKH¿HOGVRIcybernetics and engineering control
theory are built around issues of feedback control.

50
Activity

/HDUQ WR UHDG WKH FRGH IRU WKH RQHH\HG 7DGUR ZKLFK \RX¶OO ¿QG
KHUHKWWSSDJHVYDVVDUHGXGDUZLQVGHYLFHVFRGH

The following are a few tips to help you read this code.

LQGLFDWHVDFRPPHQWOLQHWKDWZRQ¶WEHUHDGE\WKHFRPSXWHU<RX
can leave your self-important notes this way.

6HWXSLVDIXQFWLRQWKDWLVUHDGRQFHZKHQWKHSURJUDP¿UVWVWDUWV
running. Variables are initialized, and libraries containing important
IXQFWLRQDO URXWLQHV IRU KDUGZDUH DUH FDOOHG 7KH ZRUG ³YRLG´ LQ
front of setup() indicates its data type; void means that this function
UHWXUQVQRWKLQJ&RQWUDVWWKDWZLWKWKHYDULDEOHVVXFKDV³LQWSRV´
WKDWXVHLQWHJHUV³LQW´IRUWKHYDULDEOH³SRV´

Loop() is a function that every Arduino program must have.

The instructions inside are read over and over until you turn the
Arduino off.

Curly braces { } are always found in facing pairs, and they have
different uses. They are used to circumscribe the code used
by a function. They indicate the code that belongs to a loop or a
conditional statement. It’s important to note that functions, loops,
and conditionals can be nested and that nesting is organized with
the braces.

The #include is an instruction to use a library outside of the

program—in this case, the control language for a servomotor.

The semicolon (;) must be included at the end of every line, with the
exception of the lines for setup() and loop(). Forgetting a semicolon
is an easy way to have your program not compile. Compiling is the
process that converts the words and symbols you’ve typed into the
machine code that the computer actually uses for instructions.

51
 $ YDULHW\ RI PDWKHPDWLFDO RSHUDWRUV FDQ EH XVHG  í
PXOWLSOLFDWLRQGLYLVLRQ&RPSDULVRQVFDQEHPDGHZLWKWKH
IROORZLQJ RSHUDWRUV  HTXDO WR  OHVV WKDQ ! JUHDWHU WKDQ
 OHVVWKDQRUHTXDOWRDQG! JUHDWHUWKDQRUHTXDOWR

<RXFDQ¿QGDGGLWLRQDOLQIRUPDWLRQDERXWWKHODQJXDJHRI6NHWFKDW
WKH$UGXLQRVLWHKWWSDUGXLQRFFHQUHIHUHQFHKRPH3DJH

Important Terms

cybernetics7KHVWXG\RIKRZG\QDPLFV\VWHPVDUHUHJXODWHGIRFXVLQJRQ
issues of feedback, control, and communication.

negative feedback control Regulatory process to maintain stable output

of the robot using the difference between the desired output and the
actual output.

Suggested Reading
Lecture 5—Robot Controllers and Programming

Banzi, Getting Started with Arduino, chaps. 1–4.

Monk, Programming Arduino, chaps. 1–3.

Other Resources

Arduino Development Environment

:LWKDQ$UGXLQRVWDUWHUNLWLQKDQGKWWSVWRUHDUGXLQRFFFDWHJRU\
get software that you’ll need to make your Arduino microcontroller
ZRUNKHUHKWWSDUGXLQRFFHQ0DLQ6RIWZDUH

RobotC

Many programming languages for robots are variations on the language

called C. RobotC (free trial) is one, and it can be downloaded at
KWWSZZZURERWFQHWZKLFKLQFOXGHVPDQ\KHOSIXOWXWRULDOV

52
PartSim

You can design and then simulate a robot circuit—before you try
to build it—using one of many simulators available on the Internet.
3DUW6LPKWWSZZZSDUWVLPFRPLVIUHHDQGHDV\WRXVH

Questions to Consider

1. Here is one way to think about the functional loop of a behaviorally

autonomous robot.

Actuators Instructions for actuators

to make changes
to the world.
Change the world. Controller
Decide to make changes
to the world based on
Sensors sensor readings.

If you wanted your robot to work even if the sensors didn’t detect any
changes to the world, what kind of functionality would you need to add
to the program operating the microcontroller?

,)\RXGHWHFWQRFKDQJHVLQVHQVRUUHDGLQJV7+(1BBBB

a. keep giving the same instructions to the actuators.

E K DOWDQGRUWXUQRIIDFWXDWRUVDQGZDLWIRUWKHVHQVRUVWRGHWHFW

a change.

c. go into a random search pattern.

53
 2. :KHQ ZH ZULWH DQ DOJRULWKP IRU D FRPSXWHU LQ (QJOLVK WKLV LV FDOOHG
pseudocode. While pseudocode can’t be used directly to program the
microcontroller, it is great way to start designing the real code. Using
pseudocode, how would you create an algorithm that lets a Roomba
with two bump sensors and two infrared (IR) detectors navigate around
the house?
Lecture 5—Robot Controllers and Programming

54
 Human-Inspired Robot Planning
Lecture 6

S
hakey was a mobile robot built by researchers at the Stanford
Research Institute from 1966 to 1972. This is the robot that Life
PDJD]LQHLQKHUDOGHGDV³WKH¿UVWHOHFWURQLFSHUVRQ´$OWKRXJK
Shakey didn’t look like a human, it had the ability to reason and plan with
what was considered a brain-like machine—a digital electronic computer.
In the controlled but real-world conditions of a laboratory, Shakey could
successfully navigate, move through the world from one place to another,
DYRLGREMHFWVDQGPRVWLPSUHVVLYHO\DGMXVWWRFKDQJHVLQWKHZRUOGWKDWLW
hadn’t known about. Shakey could learn and update its plans.

Navigation
z Key to Shakey’s thoughtful, deliberative navigation is a computer-
reasoning construct for the robotic controller that we call an
internal world model. Model-based architectures are inspired by
humans and our ability to visualize, plan, and simulate our actions
before we undertake them. Some of the most important actions that
we and our robots take are the goal-directed movements that we
call navigation. If you use a world model to navigate, we call that
model a map.

z In robotics, navigation is the most important problem that you

must get your mobile robots to solve. How do you get safely and
HI¿FLHQWO\ IURP SRLQW$ WR SRLQW %" 1DYLJDWLRQ LV D IXQGDPHQWDO
example of movement with a purpose, and purposeful movement is
at the heart of self-control and autonomy in mobile robots.

z There are different solutions to navigating depending on the

workplace and task of your robot. Shakey shows us that maps can be
useful indoors, and modern robots working indoors often use maps,
WRR²PDSV RI ÀRRU SODQV IRU H[DPSOH 0DSV DUH DOVR H[WUHPHO\
helpful for sustained and long-distance navigation outside.

55
 ‹,JRU0RM]HVL6WRFN7KLQNVWRFN
Global positioning systems have made the work of navigation quite easy for
humans.

z Over the past 2,000 years, we humans have developed a whole set of
formal rules for navigation. Navigation is not a trivial problem, even
Lecture 6—Human-Inspired Robot Planning

though we take it for granted, thanks to the computers and global

positioning systems (GPSs) that do the heavy lifting these days.

Homing
z Navigation begins with needing to know where you are, relative to
some reference position or landmark, so that you can set a course
to go where you want to go. In robotics, you can have your robot
know where it is without using a map, provided that the workplace is
VWUXFWXUHGSUHGLFWDEOHDQGFRQ¿QHG<RXFDQVHWXSDEHDFRQVRWKDW
the robot can always return home. This process is called homing.

z Homing is the ability to go out and get back to where you started.
Once you are back to where you started, you know where you are.
For robots, homing is important for returning from a mission or
recharging batteries.

56
z )RU 5RRPED KRPLQJ LV FDOOHG GRFNLQJ EXW LW¶V WKH VDPH DFWLRQ
returning home to where it started. Roomba stops cleaning and
VWDUWVVHDUFKLQJIRUWKHLQIUDUHGVLJQDOIURPWKHGRFN2QFHLW¿QGV
it, its behavior changes; it heads straight in, slowly, to the dock.

z Roomba is able to navigate without a map because it can sense the

signal from the infrared beacon on the dock. Using a beacon is an
example of what we call landmark navigation. A beacon is one type
of landmark that actively emits a signal. The classic navigational
beacon is a lighthouse.

z For Roomba, operating without an internal world model, all we

need its computer code to do is to read the intensity of the beacon’s
LQIUDUHG VLJQDO ZLWK LWV LQIUDUHG VHQVRU DQG PDNH DGMXVWPHQWV WR
minimize the difference between the maximum value measured and
WKHYDOXHVMXVWPHDVXUHG7KLVLVDQH[DPSOHRIQHJDWLYHIHHGEDFN
control, where you minimize the error between a desired and actual
state in order to regulate your movements.

z We can use negative feedback control to make an algorithm to

DOORZ5RRPEDWRQDYLJDWHWRLWVKRPLQJEHDFRQZLWKMXVWDVLQJOH
infrared sensor. We measure the strength of the infrared signal to
the homing beacon at a particular moment in time. Then, the robot
moves forward and to the right, for example, and we take another
measurement. If we take the difference between the maximum
value we’ve measured so far and the current value, the difference is
called an error term in control theory.

z But with only a single sensor, we don’t yet know what that error
tells us. Is the robot going too far to the right or too far to the left?
We can see that the robot is actually turning away from the beacon
to the right, for example. But the robot doesn’t know that; it needs
more information.

z One way that robots, including Shakey, get more information

is to move and see, or sense, what happens. This is called
active perception.

57
 z The robot moves to the right again, and we measure the infrared
value again, along with the difference with the maximum value. We
can compare the two errors. If the most recent error is greater than
the earlier error, the conclusion is that the error is increasing. And
because we kept track of how the robot was turning, we know that
it was turning to the right and that movement increased the error.
So, if the goal of negative feedback is to decrease the error, then the
robot needs to turn in the opposite direction.

z The robot turns to the left now as it moves forward. Comparing the
PRVWUHFHQWHUURUWRWKHSUHYLRXVHUURUZH¿QGWKDWE\WXUQLQJWRWKH
left, the robot is decreasing the error. This means that the robot is
heading back toward the beacon—it’s navigating.

z :H¶YH MXVW FUHDWHG DQ DOJRULWKP RU D VHW RI LQVWUXFWLRQV IRU
QDYLJDWLQJWRZDUGDEHDFRQXVLQJMXVWDVLQJOHVHQVRU2QHRIWKH
important tricks of this simple homing algorithm is that we had to
know something about how the robot was moving. In this case, we
were keeping track of which direction the robot was turning.

Holding a Compass Heading

z The algorithm for homing is very similar to an algorithm for
Lecture 6—Human-Inspired Robot Planning

holding a compass heading. Obviously, a compass is used as the

navigational sensor. Like homing or docking, holding a compass
heading is a very basic behavior in robot navigation.

z A digital compass is a common sensor to put on a robot. The

Arduino robot has one built in. It is a Honeywell model HMC 6352,
a two-axis magnetoresistive sensor. These sensors change their
HOHFWULFUHVLVWDQFHLQUHVSRQVHWRFKDQJHVLQWKHPDJQHWLF¿HOG

z One of the great advantages of a compass over a system like

an infrared beacon is that it works at any distance and in any
condition—in snow, in rain, or with obscured vision. The one great
disadvantage of a compass is that it is sensitive to local magnetic
sources, so you need to keep that in mind if you get strange results
or have to operate around a lot of machinery.

58
 z In the computer code for Arduino robot, we’ve programmed
the robot to move along a course set by its internal compass. We
calculate the error term as the difference between the desired and
the actual compass heading. When we turn on Arduino robot, the
¿UVW WKLQJ LW GRHV LV WR WXUQ DQG WKHQ KHDG RXW VWUDLJKW DORQJ WKH
course dictated by our choice of a compass heading. It makes small
turning corrections based on the errors.

z Holding a compass heading is perhaps the most basic behavior

of navigation. If a robot can’t move steadily on a given compass
KHDGLQJ WKHQ LW ZLOO QRW EH YHU\ HI¿FLHQW ZKHQ LW FRPHV WR
navigation using a map.

z One of the great advantages of using a compass is that you don’t

need to use a beacon or a landmark. But holding your compass
heading only gets you so far. With a compass alone, you don’t know
ZKHUH\RXDUHEXWRQO\LQZKDWGLUHFWLRQ\RXDUHKHDGHG(YHQLI
you don’t know where you are, a compass can be useful to know
ZKHUH\RXMXVWZHUHZKLFKLVKRZZHFDQEXLOGXSWRKRPLQJIURP
holding a compass heading.

Using Multiple Sensors

z For navigation with a compass, in addition to knowing your
direction, to know where you are on a map as you travel, you
QHHGVRPHWKLQJWRKHOS\RX¿JXUHRXWWKHGLVWDQFH\RX¶YHWUDYHOHG
from a known starting position. If you are on foot, then you can
count your steps. If you know the length of each step, then you can
PXOWLSO\ WKH QXPEHU RI VWHSV DQG WKH VWHS OHQJWK WR ¿QG WKH WRWDO
distance you’ve traveled.

z You can solve this problem on a wheeled robot by counting the

QXPEHU RI URWDWLRQV RI WKH ZKHHOV (OHFWURQLF VHQVRUV FDQ FRXQW
rotations, such as a sensor called a shaft encoder, which is often
D PDJQHW RU DQ LQIUDUHG HPLWWHUGHWHFWRU WKDW FRXQWV KLJKFRQWUDVW
lines on the wheel. With a stepper motor, you simply tell the
motors how much to rotate.

59
 z If you know the size of your wheel and the rate at which it is
rotating, you can build a speedometer. If you have a clock to keep
track of how long you’ve been going in that direction and at that
speed, then you can calculate the distance you’ve traveled, your
current position relative to where you started, and how to get back
home. The distance traveled is the product of your speed and the
time you’ve been moving at that speed. With the distance and
direction you’ve traveled from your starting point, you know where
you are relative to where you were.

z What’s important in navigation is that data from multiple sensors—

such as compasses, odometers, speedometers, and clocks—are
LQYROYHG$QG FRRUGLQDWLRQ RI PXOWLSOH VHQVRUV VRXQGV OLNH D MRE
for a robotic controller.

z In practice, going straight out and then straight back is rarely very
useful. We often want our robots to explore or to carry passengers
for hours along a course on the road or the sea that may involve
avoiding land, bodies of water, or known obstacles.

z A kind of navigation called dead reckoning involves knowing

where you are and then heading out and moving along your intended
Lecture 6—Human-Inspired Robot Planning

course, making turns after you’ve traveled a given distance and

heading as calculated by your compass, clock, and speedometer.

z One way to know where you are is to have your robot use GPS.
One of the exciting breakthroughs in consumer robotics is that
robots are now fully capable of navigating for themselves outside
ZLWK *36 ZKLFK LV D ÀHHW RI QDYLJDWLRQDO EHDFRQV PRXQWHG RQ
satellites in orbit, that provide us and our robots with what we call
a QDYLJDWLRQDO¿[.

z $ QDYLJDWLRQDO ¿[ UHIHUV WR ¿[LQJ RU ORFDWLQJ \RXU SRVLWLRQ RQ D
map. To do this with GPS, the robot needs a GPS receiver that reads
WKHVLJQDOIURPXSWR¿YH*36VDWHOOLWHV7KHVDWHOOLWHVNQRZH[DFWO\
where they are, even though their position is changing as they orbit.
By knowing where the satellites are and the distance to them, the

60
 robot can triangulate its position. This is done with a GPS module,
and some trigonometry, which is done by a tiny computer on the
GPS module. The GPS module is basically a fancy sensor that gives
the robot its position in latitude and longitude.

z With GPS, we can get around the problems of dead reckoning

by using waypoints. We can then tell the robot to go to the
waypoint. If the robot senses that it’s not in the right position, it
FDQ PDNH DGMXVWPHQWV 7KRVH DGMXVWPHQWV DUH GRQH E\ QHJDWLYH
feedback control.

Shakey: A Pioneer
z Shakey was able to sense, model, plan, and act. In terms of a
sensor-actuator system, the information from the sensors was sent
to the computer controller. The controller updated the model and
then created a new plan for how to best achieve its navigational
goals. Then, it sent the plan back to the robot, which moved the
actuators accordingly.

z In spite of the fun some modern roboticists poke at Shakey, the

URERW ZDV DKHDG RI LWV WLPH 6KDNH\ ZDV WKH ¿UVW URERW WR XVH D
powerful digital computer for its robotic controller. Because of that
FRPSXWLQJ SRZHU 6KDNH\ ZDV DOVR WKH ¿UVW WR XVH ZKDW ZH QRZ
think of as the model-based controller architecture. The use of
rules of reasoning and deliberation in the planning routines of the
FRQWUROOHUOLQN6KDNH\WRWKHHDUO\GD\VRIWKH¿HOG-RKQ0F&DUWK\
FDOOHGDUWL¿FLDOLQWHOOLJHQFH

Important Terms

dead reckoning $ PHWKRG RI QDYLJDWLRQ LQ ZKLFK FXUUHQW SRVLWLRQ LV
calculated based on speed, heading, and time from last known location.

global positioning system (GPS)1DYLJDWLRQDOV\VWHPXVLQJVDWHOOLWHVWKDW

locate a receiver in coordinates of the world geodetic system.

61
 homing 1DYLJDWLQJ WR D SUHYLRXV ORFDWLRQ XVXDOO\ WKH ORFDWLRQ ZKHUH D
robot started.

QDYLJDWLRQDO¿[/RFDWLRQIRXQGXVLQJH[WHUQDOUHIHUHQFHSRLQWV

shaft encoder$VHQVRUWKDWFRQYHUWVWKHURWDWLRQVRIDZKHHOLQWRDQHOHFWULF
signal that represents the speed of the wheel’s rotation.

stepper motor$EUXVKOHVV'&PRWRUWKDWFDQURWDWHSUHFLVHO\ZLWKRXWWKH
need for sensory feedback.

Suggested Reading

Cook, Mobile Robots, chap. 4.

Murphy, Introduction to AI Robotics, chap. 9.

Other Resources

Shakey is incredibly important in the history of robotics. Learn more about it

DWWKH6WDQIRUG5HVHDUFK,QVWLWXWHKWWSZZZDLVULFRPVKDNH\
Lecture 6—Human-Inspired Robot Planning

Questions to Consider

1. Imagine that you have a robotic ship on the ocean. Where I am

If you want your robot to navigate successfully going.
to the next waypoint, what is the minimum
information it would need?

a. Map coordinates of the next waypoint. Where I am.

b. A compass heading to the next waypoint.

c. A compass heading to the waypoint and Where I was.

distance to the waypoint. Fixed starting point.

62
2. ,PDJLQHWKDW\RXKDYHDURERWWKDWZRUNVLQDQRI¿FHEXLOGLQJSLFNLQJ
up and delivering mail. If you want a robot to be able to update its world
PRGHOTXLFNO\ZKDW¶VWKHPRVWHI¿FLHQWPHWKRG"

D 0RQLWRUDOOREMHFWVDQGSHRSOHDWDOOWLPHV

E 0
 RQLWRUDVXEVHWRIREMHFWVDQGSHRSOHZLWKZKLFKDQGZKRP\RX
are likely to interact.

c. Wait until you run into something that’s not on your map and then
add it to your map.

3. Model-based controller architectures focus on planning. In contrast,

EHKDYLRUEDVHG DUFKLWHFWXUHV IRFXV RQ UHDFWLQJ :KDW DUH WKH EHQH¿WV
and costs of these different approaches?

63
 Animal-Inspired Robot Behavior
Lecture 7

I
n principle, building an internal world model is a great idea for a robot,
if it has the sensors and the computing power. It senses, plans, and then
acts. That’s how we thoughtful creatures work, but the trick is that we
have to constantly build our model. Humans have to create and update our
map while we are operating in the world; we have to rapidly, constantly, and
accurately update our internal world model. We have to make and remake
our map, continuously, as we learn about the constantly changing world
we’re in. If we don’t, we could be in trouble.

Removing Planning from the Approach

z In the early 1970s, robot Shakey’s answer to this problem was
twofold. First, Shakey moved short distances, stopped, and then
used its sensors to look for changes in the world. Second, if Shakey
IRXQG REMHFWV WKDW ZHUHQ¶W RQ LWV PDS LW ZRXOG XSGDWH WKH PDS
Then, and only then, would Shakey calculate a new navigational
plan. This took time, and, looking back, it’s easy to criticize Shakey
for being slow. But this was 1970, before small, fast personal
Lecture 7—Animal-Inspired Robot Behavior

computers were invented.

z How do we make Shakey faster, smarter, with more intelligent

behavior for responding quickly to changes in the world? Most
UHVHDUFKHUV LQ DUWL¿FLDO LQWHOOLJHQFH $, WKRXJKW WKH\ NQHZ WKH
DQVZHU6KDNH\QHHGHGDEHWWHUEUDLQDIDVWHUFRPSXWHUZLWKPRUH
memory. That way, Shakey could update its world model faster,
PDNH LW PRUH DFFXUDWH DQG PDS DQG UHPDS PRYLQJ REMHFWV DV LW
kept track of its own movements. They were right, in part. Faster
computers with more memory do help.

z But in the 1980s, this make-the-brain-smarter and use-maps

DSSURDFK KDG ODUJHO\ IDLOHG DQG QRW MXVW IRU WKH UHDVRQ WKDW WKH
FRPSXWHUV ZHUHQ¶W ELJ HQRXJK DQG IDVW HQRXJK 7KH ¿HOGV RI
AI and AI robotics were stuck in a methodological rut. Mobile

64
 map-making robots were still very slow to move and worked
only under conditions where the world was structured—that is,
a world that is predictably not changing much. Throw most real-
world situations at these robots and they simply couldn’t move or
accomplish their tasks.

z One AI researcher, Rodney Brooks, a professor at the Massachusetts

Institute of Technology, began to question the whole sense-plan-act
approach of model-based robotics. Instead of thinking about how
humans work, Brooks was inspired by the intelligent behavior of
QRQKXPDQDQLPDOV+HZDWFKHGLQVHFWVPRYHÀ\QDYLJDWHVWLQJ
and pollinate.

z He observed that insects behaved intelligently—doing things no

robot at the time could do—without much of a brain at all. This got
Brooks thinking that part of the problem with robots like Shakey
was that humans were trying to build them in their own image, with
humanlike intelligence.

z %URRNV ¿JXUHG WKDW E\ IRFXVLQJ RQ KXPDQ LQWHOOLJHQFH²DEVWUDFW

reasoning tasks, such as playing chess, solving math problems, and
EXLOGLQJLQWHUQDOZRUOGPRGHOV²ZHKDGVWDUWHGWKH¿HOGRIPRELOH
robotics backward, evolutionarily speaking. Why not work the way
that evolution had? Start with the simpler animals and then build up
from there.

z Pushing that idea to its limits, Brooks wondered if you could still
get intelligent behavior if you removed planning altogether from
the sense-plan-act architecture of the robotic controller. Don’t
create an internal world model but, instead, rely on sensing and
DFWLQJZLWKRXWSODQQLQJ$QGEHFDXVHWKHSODQQHULVWKHEUDLQMXVW
remove it altogether.

The Biology of Brains and Intelligence

z How do animals that don’t have much of a brain, who don’t do
much planning, manage to live, behave, and be successful? We
humans are animals with a brain that seems large in proportion

65
 to our body size. A ratio calculated across all animals called the
encephalization quotient suggests that our brains are more than four
times larger than would be expected from body size alone.

z However, our big-brain pride tends to blind us when we consider

the cognitive workings of nonhuman animals and their implications
for robots. Most people think that no other animal or robot is as
smart, as intelligent, or as adaptable as a human. Depending on the
situation, they might be right.

z But intelligence depends on the context, the environment in which

your species evolved. Judged by mass, rather than size, small
animals such as birds, mice, and even ants have brains that are at
least as, or more, impressive than those of humans. Sensors and
actuators across the animal kingdom are even more impressive.

z The implication that matters most for building robots is that every
species has its own special set of sensors (and actuators) that are
DGDSWHGIRUZRUNLQJDWDSDUWLFXODUMRELQDSDUWLFXODUHQYLURQPHQW
For example, female mosquitoes are olfactory and thermal
Lecture 7—Animal-Inspired Robot Behavior

‹VWLOOZRUGVL6WRFN7KLQNVWRFN

Sensor-guided movement is apparent in every species and is central to

intelligent behavior.

66
 predators, using an array of chemical sensors on their antennae to
À\XSWKHVWUHDPVRIFDUERQGLR[LGHWKDWZHPDPPDOVPDNHDVZH
live and breathe.

z When thinking about insects, Rodney Brooks realized the following

LPSRUWDQWOHVVRQIURPELRORJ\6HQVRUJXLGHGPRYHPHQWLVFHQWUDO
to intelligent behavior. Movement without sensors may be cool, but
it doesn’t let you navigate. You can’t move toward any particular
JRDO<RXFDQ¶WQDYLJDWHRUJHWSUHVSHFL¿HGWDVNVGRQH

Sensor-Guided Movement
z Whether or not you think a mosquito is intelligent, you have to
appreciate their pesky hunting abilities. Mosquitoes can track down
mammals with a simple system of sensors connected to a simple
nervous system that creates a continuous turning signal for the
wings. This is sensor-guided movement.

z Mosquitoes are incredibly successful survivors, and they win the

game of life without an internal map. They don’t need one. All they
need to navigate to the food is sensor-guided movement.

z ,Q WKH ODQJXDJH RI DUWL¿FLDO LQWHOOLJHQFH PRVTXLWRHV EHKDYH

intelligently because they don’t waste processing power making
PDSVWKDWWKH\GRQ¶WQHHG,QVWHDGVLPSOHUHÀH[OLNHFRQQHFWLRQV
EHWZHHQVHQVRUVDQGPRWRUVJHWWKHMREGRQH

z Instead of sense-plan-act, mosquitoes sense and act—without

planning. They behave without a brain. These were exactly the
kinds of bioinspired insights that Rodney Brooks was after when he
was thinking about the slow progress of the sense-plan-act approach
of classical AI robotics.

z If we wanted intelligent machines that could actually do something

in the real world, Brooks reasoned that we needed to stop focusing
on the computationally intensive problem of building an internal
world map. Instead, we needed to think about what happens in the
world. Behavior happens. Animals move around and get stuff done.

67
 z So, how could robots behave intelligently without planning and
without much of a brain? Once you understand the intelligent
behavior of insects, you can use the simple sense-and-act rule to
SURJUDPDURERW$QGWKDWZDV%URRN¶VÀDVKRIEULOOLDQFH

z %XW D NH\ SUDFWLFDO TXHVWLRQ UHPDLQHG :LOO WKRVH VLPSOH LQVHFW

LQVSLUHG UHÀH[OLNH UXOHV UHDOO\ ZRUN RQ D URERW" 7R ¿QG RXW
Brooks and his graduate students at MIT built a six-legged, insect-
like robot named Genghis. Like a mosquito, Genghis had onboard
sensors to detect mammals. But while mosquitoes sensed carbon
GLR[LGH%URRNVIRFXVHGRQDQRWKHULQVHFWVHQVHKHDWGHWHFWLRQ

z Brooks and his team mounted six pyroelectric heat-detecting

sensors on the front of Genghis. Genghis used these sensors and a
simple rule to follow warm human mammals around the lab. And it
worked beautifully. Finally, in the form of the mobile autonomous
Genghis, here was a mobile robot that behaved intelligently.
Genghis could move around in the world and do something quite
LPSUHVVLYHWUDFNDQG¿QGKXPDQV

z No other robot from the AI community had come close to this kind
of ability, this kind of intelligent behavior. And, best of all, Genghis
Lecture 7—Animal-Inspired Robot Behavior

could track mammals in a dynamically changing world, without a

map—no planning needed, and no brain required.

z Brooks moved robotics from the classic sense-plan-act paradigm,

what we call model-based robotics, to what is now called behavior-
based robotics. In 1988, when Genghis was built, behavior-based
robotics was a breakthrough in the world of robotics, and Brooks
became internationally famous. Brooks and his students went on to
XVHKLVEHKDYLRUEDVHGURERWLFVWRFUHDWHWKH¿UVWFRQVXPHUURERWWR
KDYHZLGHVSUHDGVXFFHVV5RRPED

z Compared to model-based systems, behavior-based systems are

simple and robust. They require a lot less computer processing.
And simple design plus less computer processing, in turn, means

68
 that your batteries last longer. More importantly, Roomba doesn’t
need a map to work. Like an insect, you put Roomba down, and it
gets to work.

z What are the advantages of a behavior-based robotic architecture?

Because it doesn’t have to plan, a behavior-based robot can react
quickly. Because the software that runs the computer on the robot
doesn’t have to be as complex, it means that the whole robot is
OHVVOLNHO\WREUHDNGRZQ$QGWKHPRGXODULW\RIVHQVHDFWUHÀH[HV
means that you can use them like building blocks to easily create
complex behaviors.

z But best of all, the sensor-guided movements of behavior-based

robotics has gotten engineers to focus on building better sensors.
Because behavior-based robots work, they are now a fascinating
part of contemporary life, thanks to pioneers like Rodney Brooks.

Activity

9DVVDU¶V 5RERW (WKRORJ\ &KDOOHQJH KWWSZZZPLQGLOVWXHGX

FXUULFXOXPYLUWXDOBURERWLFVBODEYLGHRSKS

By visiting this website, you can select the priorities of different

behaviors and see how your programming impacts the overall
behavior of a small wheeled robot.

Suggested Reading

Brooks, Flesh and Machines.

Dawson, Dupuis, and Wilson, From Bricks to Brains, chaps. 6–7.
:DOWHU³$Q,PLWDWLRQRI/LIH´

69
 Other Resources

Fast, Cheap & Out of Control  $ GRFXPHQWDU\ E\ ¿OPPDNHU

(UURO 0RUULV DERXW IRXU XQUHODWHG LQGLYLGXDOV LQ XQXVXDO ¿HOGV KDYLQJ
something to do with animals, one of whom happens to be MIT roboticist
Rodney Brooks.
Lecture 7—Animal-Inspired Robot Behavior

70
 Basic Skills for Making Robots
Lecture 8

I
n order to make robots, there are some basic, preliminary skills that you
need to know. The ones you will learn are very basic, but the point is
that it takes very little to get started in the world of making robots. Soon
enough, you will know what you need to do in order to make a circuit, hack
a robot, and even make a toy into a robot. No matter what tools you have in
your toolkit, always make sure to start with some kind of safety protection,
such as safety glasses.

Supplies of the Maker

The following is a list of tools of the trade that you’ll need if you’re
going to be a maker of robots.

Ŷ :LUHFXWWHUV
Ŷ 0LQLDWXUHVFUHZGULYHUV
Ŷ &RQYHUWLEOHVFUHZGULYHU
Ŷ 3OLHUV
Ŷ 6PDOOVFLVVRUV
Ŷ 6ROGHULQJLURQDQGVWDQG
Ŷ 0XOWLPHWHU
Ŷ 6ROGHUOHVVEUHDGERDUG
Ŷ +RWJOXHJXQ
Ŷ 6ROLGFRUHZLUHUHGEODFNDQG\HOORZ
Ŷ 6WUDQGHGZLUHUHGEODFNDQG\HOORZ
Ŷ /HDGIUHHsolder with rosin core
Ŷ %ODFNHOHFWULFWDSH
Ŷ 7ZLVWWLHV
Ŷ 3HUPDQHQWPDUNHU
Ŷ $$DQG9EDWWHULHV

71
 ‹$UWXVKL6WRFN7KLQNVWRFN
Make sure that you have all of the necessary tools handy before you begin
your project.

Skills of the Maker

z Some of the core skills you need to know in order to build robots
include stripping a wire, twisting two wires together, and soldering
two wires together. When it comes to soldering, safety is important,
because you can burn yourself. Some of the safety practices that
Lecture 8—Basic Skills for Making Robots

should be followed include putting on safety glasses and working

in a ventilated area.

z :KHQVROGHULQJWZRZLUHVWRJHWKHU¿UVWPDNHVXUHWKDWWKHVROGHULQJ
iron is fully heated. Then, clean the tip of the hot soldering iron
on a damp sponge. The next step is tinning the tip, which involves
melting a small amount of metal and putting it on the tip of the
soldering iron to prepare it for use.

z The counterintuitive thing about soldering is that you don’t actually

want to melt the solder directly. Instead, you want to heat up
the wires and let the wires melt the solder. To do this, touch the
soldering iron to the wire (not the solder). Then, touch the solder to
WKHKRWZLUHQHDUWKHLURQ/HWWKHVROGHUÀRZLQWRWKHZLUH

72
z Another skill that is good to
have when building robots is
how to test for continuity with
a multimeter. Multimeters are
YHU\ LQH[SHQVLYH \RX FDQ ¿QG
them at hardware stores. You
need to make sure that your
multimeter comes with cables.

‹UREHRL6WRFN7KLQNVWRFN
6SHFL¿FDOO\ WKHUH VKRXOG EH D
red and a black test cable, or test
lead. Multimeters measure direct
current voltage and alternating
current voltage. When we build
Ventilation and eye protection
electronics, we’re working in the are important precautions to take
direct current world. when using a soldering iron.

z A speaker on the multimeter measures what’s called continuity. We

want to know if we have a good connection, or continuity, when we
wire something together. If you touch the two leads together after
they’re plugged in to the multimeter, you should hear a beep, which
is the signal that you have continuity.

z 7KHUH LV DQRWKHU VNLOO WKDW LV XVHIXO ZKHQ ZRUNLQJ ZLWK D FLUFXLW
using a breadboard. A breadboard has outer power strips that are
connected internally. By putting your circuit on a breadboard, you
don’t have to do any wire twisting, and it’s a way to lay out your
circuit and make sure that it works before you do something, such
as solder.

z The core of making a robot is connecting a sensor to an actuator.

That means building and testing a circuit. Voltage dividers show
up everywhere in electronic circuits. They are very handy, and they
can be used in a circuit to make a light-sensitive robot. A voltage
divider involves twisting two resistors together, as well as input
wires and output wires.

73
 ‹VZLIW\XNL6WRFN7KLQNVWRFN
Solderless breadboards are useful for laying out and testing circuits before
committing with solder.

Hacking Robots
z Hacking a robot is where you really make it your own. The hacking
might start with a toy, or a robot, or any other gadget. You might
be changing software code or adding controller intelligence to a
previously dumb machine, or you might be changing the body of
Lecture 8—Basic Skills for Making Robots

the robot. Any and all of these count as hacking.

z Hacking is a design process in the sense that you aren’t following

any set of directions. Hacking requires no specialized training. To
hack, you observe, dissect, tinker, commandeer, recombine, and
repurpose. The whole idea is to get something that you have to do
something that it was never intended to do.

z The problem with hacking is that most of us were told not to break
our toys—but forget about that. Because hacking works by taking
something else apart, many people get nervous about doing so.
But don’t be afraid to tinker and make mistakes. Fail your way to
success, as counterintuitive as that seems.

74
z A tank track is a very common thing that is used in robots. There are
kits that you can buy that let you build it. It’s meant to be the basis
of any kind of robot that you might want to build. You can also
VWDUWZLWKDQ\VRUWRIWR\WKDWMXVWJRHVVWUDLJKWZKHQ\RXWXUQLWRQ
Tracks are even able to get up and over obstacles, but they’re blind.
There are actuators and a body, but there are no sensors attached.

z You can use a circuit that you lay out on a breadboard to make a
robot. A transistor is what allows you to hack the tank track to make
it into a robot. A transistor can be a fancy kind of switch that sends
the data to the voltage that is being divided by the voltage divider in
the circuit. It’s a kind of gate that allows greater amounts of current
to pass through it to power the motor. In fact, with some extra steps,
you can make a photosensitive robot.

z One of the great things about hacking a robot, or building one with
help from a kit, is that anyone can try it. In fact, the robot hackers’
secret is that you can hack any robot—and hackers do. With the
ULVHRINLWVOLNH/(*20LQGVWRUPV9(;5RERWLFV'HVLJQ6\VWHP
and BIOLOID, it has never been easier for beginners to build
sophisticated robots.

z 7KHUHLVDNLWIURP9(;5RERWLFVWKDWOHWV\RXEXLOG&ODZERWZKLFK
is a great way to get started on building a sophisticated robot—in
this case, one that includes a manipulator and a mobile robot base.
Best of all, when you start building robots like this and getting them
to work, you’re getting a shortcut to creating something lifelike
DQGMXVWDVLQWHUHVWLQJDXWRQRPRXVURERWEHKDYLRU

z There’s also a Hacking Roomba website to get you started. It is

full of resources, including a link to Roomba’s company, iRobot.
Moreover, the company not only supports this, but in 2007, they
also created a Roomba that they want you to hack, called Create, so
that you don’t have to worry about wrecking your Roomba.

75
 z Instead of a vacuum cleaner, the Create robot has a cargo bay where
iRobot has added a communications device that allows you to
easily hook in a variety of sensors. iRobot has also made available
software, called Open Interface Commands, to make more readily
available a number of the behavior modules that Roomba uses. One
use of Create has been to mount a camera to perform surveillance.

Activities

Hack a toy!
We’ve seen a simple hack that involves taking a motorized vehicle
and adding a voltage divider with a light-dependent resistor (LDR)
as the main sensor. The following circuit diagram shows what
\RX QHHG WR PDNH WKLV ZRUN WKH /'5 D ¿[HG UHVLVWRU WKLV ZLOO
vary depending on your LDR, but start with a value of 10 kW), a
small motor (DC brush, 1.5 to 3 V rating), and an NPN switching
transistor (type 3904). While two AA batteries (3 V total) were used
LQWKHOHFWXUH\RXPD\¿QGWKDW\RXQHHGWRGRXEOHWKHYROWDJHWR
6 V. A 9V battery can also be used with success, but you need to be
careful not to run the motor for too long or you may burn it out.
Lecture 8—Basic Skills for Making Robots

motor M
LDR
+
battery
−
fixed transistor
resistor

76
 How does this circuit work? When more light falls on the LDR, its
UHVLVWDQFH GHFUHDVHV$V PRUH FXUUHQW ÀRZV WKURXJK WKDW OHJ RI WKH
FLUFXLWLWÀRZVLQWRWKHEDVHRIWKHWUDQVLVWRUDOORZLQJPRUHFXUUHQWWR
ÀRZWKURXJKWKHWUDQVLVWRUIURPWKHPRWRUWRWKHQHJDWLYHVLGHRIWKH
EDWWHU\7KHUHVXOW$VPRUHOLJKWKLWVWKH/'5WKHPRWRUVSLQVIDVWHU

7KH KDUGHVW SDUW RI WKLV KDFN LV ¿QGLQJ WKH ULJKW NLQG RI WR\<RX
need a battery-operated car or truck with a motor that powers a
single drive axel.

Hack the code!

Peter Staten and John Long combined two programs available from
ArcBotics (www.arcbotics.com) for running Sparki. Their hack,
which was demonstrated in Lecture 1, combines edge avoidance and
remote control to force you to drive Sparki safely. It is called Peter’s
Safe Driver Program.

,I\RXKDYHD6SDUNL\RXFDQFRS\DQGHQWHUWKHLUFRGHIURPWKLVVLWH
KWWSSDJHVYDVVDUHGXGDUZLQVGHYLFHVFRGH

Important Terms

solder Metal alloys with low melting points used to fuse together separate
elements and, in electronics, create secure electric connections.

voltage divider A simple circuit that uses two resistors, with resistance Z1
and Z2, to reduce the input voltage, Vin, to a lower output voltage, Vout, by the
IROORZLQJHTXDWLRQ

77
 Suggested Reading

Monk, Hacking Electronics, chaps. 2–3.

Scherz and Monk, Practical Electronics for Inventors, chap. 7.

Other Resources

The Sparki robot by ArcBotics is a great open-source robot to get you

VWDUWHG ,W FRPHV UHDG\ WR ZRUN ULJKW RXW RI WKH ER[ KWWSDUFERWLFVFRP
SURGXFWVVSDUNL

Question to Consider

1. How do you turn a toy into a robot? Many toy vehicles have a body,
actuators, and an energy source. What do we need to add to a toy vehicle
to turn it into an autonomous robot?
Lecture 8—Basic Skills for Making Robots

78
 Designing a New Robot
Lecture 9

H
ow do we move from kits and hacks into designing and building
complex robots for commercial or research purposes? First, you
QHHG SHRSOH &RPSOH[ URERWV DUH EXLOW E\ WHDPV ZLWK GLIIHUHQW
members specializing in different aspects of design, construction, and
testing. Second, you need money—at least to pay for gear and supplies, and
you may even need to pay some or all of the people on your team. Third, you
need a clear design goal and a plan to get you there.

Robot Madeleine
z Robot Madeleine is a complex robot that was built by a team at Vassar
College in partnership with Nekton Research LLC. This combined
team included a biologist, cognitive scientist, conceptual designer,
mechanical engineer, electrical engineer, and systems integrator.

z The group had two goals in building Madeleine. First, they had
D VFLHQWL¿F JRDO 7KH\ QHHGHG D URERW WKDW WKH\ FRXOG XVH WR WHVW
LGHDV DERXW KRZ H[WLQFW DQLPDOV ZLWK IRXU ÀLSSHUV²VXFK DV WKH
SOHVLRVDXUV RU NURQRVDXUXVHV RI WKH 0HVR]RLF (UD²FRXOG VZLP
Of particular interest was why those giant ancient reptiles would
VZLPZLWKDOOIRXUÀLSSHUVEXWWKHOLYLQJVSHFLHVRIYHUWHEUDWHVWKDW
had gone back into the water, such as sea lions and penguins, swam
with only two. To answer this question, they needed to be able to
FRQWUROWKHÀLSSHUVLQGHSHQGHQWO\VRWKDWWKH\FRXOGYDU\WKHJDLW
RUSDWWHUQRIÀLSSHUXVH

z 6HFRQG VWXG\LQJ ÀLSSHUV DOVR FRQWULEXWHG WR D SHUIRUPDQFH

goal. Nekton was a small company, and they wanted to create an
amphibious robot, one that could swim and maneuver in the surf
zone and then crawl its way up the beach.

z Goals and a plan are crucial, but roboticists also keep something
HOVH LQ PLQG (YHU\ HQJLQHHULQJ SURMHFW WR FUHDWH VRPHWKLQJ QHZ

79
 goes through a series of failures before it succeeds. This process
of failing your way to success is sometimes called prototyping and
involves trial-by-error learning as you go.

z 2QH ZD\ WR JHW WULDODQGHUURU OHDUQLQJ VWDUWHG LV WR ¿UVW EXLOG
the robot in computer simulation. Coppelia Robotics has created
VRIWZDUHIRUGRLQJMXVWWKDW&DOOHG95(3ZKLFKVWDQGVIRUYLUWXDO
URERW H[SHULPHQWDWLRQ SODWIRUP LW DOORZV \RX WR FRQ¿JXUH \RXU
robot and then program it in the same computer language that you
would use in your physical robot. Then, you run your simulated
robot in a simulated world.

z $VLPXODWLRQSURJUDPOLNH95(3OHWV\RXPRGHOFRPSOH[URERWLFV
problems, such as path planning for manipulators and vision sensory
simulation. As long as you have a good physics engine for your
world and your robot, simulations can shorten development time
of the physical robot by letting you try out and compare different
designs very quickly.

z It took the team about 18 months to build Robot Madeleine, version

'XULQJWKLVSHULRGWHDPPHPEHUVVSHQWDERXWIRXUPRQWKVMXVW
¿JXULQJRXWKRZWRWDNHWKHLURYHUDOOJRDODQGWXUQLWLQWRFRQFUHWH
VSHFL¿FDWLRQV IRU EXLOGLQJ 0DGHOHLQH 7KH\ ZDQWHG 0DGHOHLQH
1.0 to be self-propelled, fully submersible, and remote controlled
(through a cable coming out of the back).
Lecture 9—Designing a New Robot

z 7KHUHDUHDKXJHQXPEHURIGHFLVLRQVWKDWKDYHWREHPDGH(YHU\
sensor, motor, part, size, and shape involves making a decision. And
it helps to have on your team someone who is the systems integrator
to make sure, for example, that the selected motors are compatible
with the chosen batteries. You don’t want to have batteries that
can’t supply the motors with enough power.

z It took another 12 months to make Madeleine truly autonomous,

a version called 2.0. Part of the issue was that they were applying
a behavior-based controller architecture to an underwater robot.
7KHELJFKDOOHQJHLVFRQWURO7KHURERWPLJKWGHWHFWDQREMHFWDQG

80
 VWRSÀDSSLQJLWVÀLSSHUVEXWLWNHHSVJOLGLQJWRZDUGWKHREMHFW7KH
WHDPQHHGHGWR¿JXUHRXWKRZWRPRYH0DGHOHLQHVORZO\HQRXJK
to avoid damaging the robot but fast enough to make progress. The
systems integrator was the chief engineer on these issues.

z 6FLHQWL¿FDOO\ 0DGHOHLQH ZDV D VPDVKLQJ VXFFHVV 7KH WHDP ZDV

DEOH WR VKRZ WKDW IRXU ÀLSSHUV ZHUH JUHDW IRU DFFHOHUDWLRQ IURP
a stop and breaking. But once Madeleine got to top speed, that
PD[LPXPZDVQRGLIIHUHQWIRUVZLPPLQJZLWKIRXURUWZRÀLSSHUV
6RJLYHQWKDWLWWDNHVPRUHHQHUJ\WRVZLPZLWKIRXUÀLSSHUV\RX
might as well swim with two, if you are a cruiser. But if you are a
sit-and-wait predator, use all four.

z 1HNWRQ¶VFRPPHUFLDOJRDOZDVDOVRDVXFFHVV:LWKVSHFLDOÀLSSHUV
that Nekton patented, Madeleine could swim in the surf and
FUDZO XS WKH EHDFK 0DGHOHLQH XVKHUHG LQ D QHZ HUD RI ÀLSSHUHG
amphibious vehicles called transphibians. Nekton’s transphibian
patent was later purchased by iRobot, the maker of Roomba,
becoming the start of a marine robotics division.

RayBot
z A new underwater robot, RayBot, is in the process of being built.
7KH5D\%RWSURMHFWLVDFROODERUDWLRQEHWZHHQWKH,QWHUGLVFLSOLQDU\
Robotics Research Laboratory at Vassar College and FarCo
7HFKQRORJLHV,QF7KHLPSHWXVIRUWKHPWREXLOGWKLV¿VKOLNHURERW
FDPH IURP WKH 86 1DY\¶V 2I¿FH RI 1DYDO 5HVHDUFK ZKLFK LQ
 FUHDWHG DQG PDGH SXEOLF D IXQGLQJ VROLFLWDWLRQ ³(I¿FLHQW
+LJKO\0DQHXYHUDEOH$UWL¿FLDO)LVKIRU6WHDOWK\6XUYHLOODQFH´

z To come up with a plan, the team needed to get down to

VSHFL¿FDWLRQV :KDW NLQG RI ¿VK" )RU D VKDSHFKDQJLQJ HQHUJ\
HI¿FLHQW VLOHQW ERG\ WKH WHDP FRQVLGHUHG VNDWHV DQG UD\V 7KH\
NQHZ WKDW ¿VK WKDW FDQ JOLGH ZRXOG EH YHU\ HQHUJ\ HI¿FLHQW$Q
HOHFWULF UD\¶V ERG\ LV ÀDWWHQHG DQG LW GRHV QRW ÀDS LWV ERG\ GLVN
7KLV LV GLIIHUHQW IURP VNDWHV VWLQJUD\V DQG PDQWD UD\V (OHFWULF
UD\VXVHWKHLUERG\GLVNWRJOLGHLQWKHPRVWHQHUJ\HI¿FLHQWDQG
stealthy form of swimming possible.

81
 z $QRWKHUEHQH¿WRIWKHHOHFWULFUD\ERG\SODQLVWKDWWKLVW\SHRIUD\
FDQFDUU\DVWDEOHSD\ORDG%RQ\¿VKKDYHWRZLJJOHDQGEHQGWKHLU
bodies in order to swim. The electric ray’s body disk doesn’t bend
up and down for propulsion; instead, the body disk is stable, so its
payload, the electric organ, is, too. The electric organs, which can
]DSDQGVWXQSUH\DUHDFWXDOO\PRGL¿HGPXVFOHV%XWWKHVHPXVFOHV
GRQ¶WFRQWUDFWWKH\MXVWVWRUHDQGUHOHDVHHOHFWULFFKDUJH

z (OHFWULF UD\V KDYH ODUJH HOOLSWLFDO FRPSDUWPHQWV ,Q D URERW EXLOW

like an electric ray, those elliptical compartments are where those
stable payloads can go.

z :KHQ WKH WHDP VWDUWHG EXLOGLQJ 5D\%RW WKH\ ¿UVW FUHDWHG DQ
autonomous surface robot modeled on the electric eel. It had
a body disk and propulsive tail. The body disk carried a payload
WKDWLQFOXGHGDFRQWUROOHU7KHFRQWUROOHUUDQD¿[HGSULRULW\VHWRI
LQVWUXFWLRQVWKDWRSHUDWHGWZREHKDYLRUVQDYLJDWHDQGDYRLGREMHFWV

z Navigation is a behavior created by linking the input from the two

photoresistors to the steering of the tail. RayBot 2.0 is programmed
WRVZLPWRZDUGWKHOLJKW2EMHFWDYRLGDQFHLVDEHKDYLRUFUHDWHGE\
linking the input from the two infrared proximity detectors, located
on the sides, with the rapid turning of the tail.

z 7KH WDLO LWVHOI ÀDSV EDFN DQG IRUWK JHQHUDWLQJ IRUZDUG thrust.
Lecture 9—Designing a New Robot

Depending on whether navigation or avoidance is the selected

EHKDYLRU WKH WDLO FDQ WXUQ JHQWO\ RU GUDPDWLFDOO\ ZKLOH LW ÀDSV
changing the direction of RayBot.

z 7KH WHDP FRQ¿UPHG WKDW WKH VKDSHFKDQJLQJ WDLO ZRUNHG XVLQJ D

servomotor, and then they turned to the problem of creating muscle-
like actuators. They did more work to understand how electric rays
glide. Their detailed understanding of electric rays made it clear
that they needed RayBot to change its posture and the camber, or
shape, of its body disk.

82
 ‹*UHJ$PSWPDQ+HPHUD7KLQNVWRFN
(OHFWULFUD\VXVHWKHLUERG\GLVNWRJOLGHPDNLQJWKHPKLJKO\HI¿FLHQWDQG
stealthy—an ideal model for an underwater robot.

z To get the surface RayBot working under water, the team enclosed
5D\%RW  LQ D ODUJHU ÀH[LEOH DQG ZDWHUSURRI ERG\ GLVN 7KH\
DOVRDGGHGVRPHSHFWRUDO¿QVEHKLQGIRUVWHHULQJXSRUGRZQRUWR
roll—called attitude control.

z %HFDXVH ZDWHU DQG HOHFWULFLW\ FDQ¶W LQWHUDFW WKH ¿UVW VWHS LQ
engineering an underwater RayBot 3.1 was to enclose the
electronics in a waterproof box, called a hull in nautical engineering
terms. While the team needed to keep the electronics dry, at the same
time, they needed to put holes in the hull in order to communicate
electrically or mechanically with the actuators outside of the hull.

z 2QWKHEDFNHQGWKHUHLVDÀH[LEOHVLOLFRQHWDLOIXQFWLRQLQJDVWKH
SURSHOOHU7KHWHDPQHHGHGWREHDEOHWRÀDSWKHWDLOEDFNDQGIRUWK
DQG FRQWURO KRZ IDVW LW ÀDSV %RWK ¿QV DQG WDLO DUH FRQWUROOHG E\
servomotors, located inside the hull.

83
 z 7RWUDQVIHUWKHIRUFHDQGSRZHURIWKHPRWRUVWRWKHH[WHUQDO¿QV
and tail, the team took another chapter out the book of nature and
created tendons—polystyrene cables that connected the oscillating
VHUYRPRWRUV WR WKH ÀDSSLQJ ¿QV DQG WDLO 7R SDVV WKHVH WHQGRQV
WKURXJKWKHKXOOWKH\PDGHWLJKW¿WWLQJUXEEHUSDVVDJHVWKDWFRXOG
be loaded with a viscous lubricant.

z 5D\%RW  XVHG LWV ÀDSSLQJ WDLO WR VZLP 'LIIHUHQW WDLOEHDW
frequencies could be programmed to control speed. When RayBot
WUDQVLWLRQHG LWV ¿QV IURP GRZQ WR XS LWV DWWLWXGH FKDQJHG DQG LW
PRYHGYHUWLFDOO\7KH¿QVFRQWUROOHGSLWFK

z +RZHYHU WKH WHDP GLWFKHG WKH SHFWRUDO ¿QV 7KH ZD\ WKH\ KDG
WKHP FRQ¿JXUHG GLGQ¶W ZRUN YHU\ ZHOO IRU DWWLWXGH FRQWURO 6R
they put that part of the problem on hold and created RayBot 3.2,
ZKLFKKDGMXVWDWDLO7KHELRPLPHWLFWDLODFWHGDVERWKDSURSHOOHU
and a rudder.

z RayBot 3.2 had the ability to detect a light source, navigate toward it,
and hold station around it. This was a simple proof-of-concept test,
showing that they could build a working autonomous underwater
vehicle prototype that was self-propelled and autonomous.

z In parallel with their work on the whole robot, the team also started
trying different kinds of muscle-like actuators, and they analyzed
Lecture 9—Designing a New Robot

whether to make them or buy them. They thought more about the
ÀDSSLQJSHFWRUDO¿QVRQ5D\%RW,WWXUQVRXWWKDWLQUHDO¿VKWKH
¿QVDUHFRPSRVHGRIWKLQURGVRIVNHOHWRQWKDWJLYHWKH¿QVXSSRUW
5HVHDUFKKDVVKRZQWKDWWKH¿QURGVDOVRDFWLYHO\ÀDSWKH¿QDQG
FKDQJHWKHFDPEHURIWKH¿Q

z The team induced a shear using a rack and pinion. But unlike the
commercially available rack and pinion attached to a rotary motor,
theirs was driven back and forth by a servomotor connected to the
pinion. They called this actuator the ¿QVKHDUDFWXDWRU.

84
z This also led them to ideas about how to engineer a better body,
RQHZLWKDERG\GLVNWKDWWKH\FRXOGQRZDFWXDWHZLWKWKH¿QVKHDU
actuator and control the posture and camber.

z To get the details right, they bought a large electric ray and made
a plaster cast of it, which they then could use to make a silicone
version. The problem with making a body this way, though, is that
it had asymmetries, bumps and bulges that were partly a result of
the way they made the cast and partly a result of the imperfections
of nature.

z So, the team used the cast of the ray to take a series of very detailed
measurements of the body shape. From those measurements, they
then created a three-dimensional computer graphic version of
the ray in a software program called Maya. Then, they made the
computer graphic version of the ray symmetric and found that in
order to come up with a mold that their machine could fabricate,
they needed to slightly alter some of the lines and curves of the
ray’s body. The result of this process was RayBot 3.3.

z :KHQWKH\FDVWWKH¿QDOERG\RI5D\%RWWKH\XVHGDÀH[LEOH
silicone-based polymer called Dragon Skin. It took nearly 20
pounds of Dragon Skin and a lot of acrobatics to create a mold and
then peel RayBot out of it. The mold was in nine pieces so that
RayBot could be taken out of it without breaking its body.

z The team also designed into the mold a payload in the belly, on the
ERWWRP RI WKH URERW ERG\ ,QWR WKDW SD\ORDG WKH\ SODFHG WKH ¿Q
shear actuator system with a controller to drive the servomotor to
FDXVHWKHWDLOWRÀDSVLGHWRVLGH7KH¿QVKHDUDFWXDWRUPHFKDQLVP
KDVVXI¿FLHQWSRZHUGHQVLW\WREHDEOHWRSURSHO5D\%RW

z The challenge for RayBot 4.0 is to add sensors to the body of

5D\%RWDQGWRXVHWKH¿QVKHDUDFWXDWRUV\VWHPWRFDPEHUWKHERG\
and take advantage of the winglike properties of the body for even
PRUHHQHUJ\HI¿FLHQF\

85
 Activity

Think of the simplest engineering activity you can imagine, and

write it down. Then, using that design as a plan, build it. Then, test
it to see if it works.

7KHHQJLQHHULQJSURFHVVLVLWHUDWLYH,WLVDF\FOHDFRQWLQXRXVSURFHVV

Important Terms

camber 7KH PHDVXUH RI WKH IURQWWREDFN FXUYDWXUH RI D ZLQJ WKDW KHOSV
create lift.

¿QVKHDUDFWXDWRU)6$$ELRORJLFDOO\LQVSLUHGGHVLJQWKDWFRQYHUWVVKHDU

VWUDLQIURPODWHUDOVWUHVVLQWKHOD\HUVRIWKH¿QLQWREHQGLQJ

rack and pinion$W\SHRIDFWXDWRUWKDWFRQYHUWVDPRWRU¶VURWDWLRQDOPRWLRQ

to a translational output.

thrust$IRUFHWKDWSURSHOVDQDQLPDOYHKLFOHRUURERW

Suggested Reading
Lecture 9—Designing a New Robot

Dym, Little, and Orwin, Engineering Design, chaps. 1–2.

+DQVRQ 5XV &DQYLQ DQG 6FKPLHUHU ³%LRORJLFDOO\ ,QVSLUHG 5RERWLFV
$SSOLFDWLRQV´

Other Resources

7KH,(((5RERWLFVDQG$XWRPDWLRQ6RFLHW\KWWSZZZLHHHUDVRUJ

86
 Questions to Consider

1. In which of the following scenarios might engineering a robot be

necessary instead of hacking?

a. You need a robot that carries your webcam around the house when
you aren’t home.

b. You need a quadcopter robot that can inspect the insides of

industrial chimneys.

c. You need a robot to explore the liquid water under the icy crust of
(XURSDRQHRIWKHPRRQVRI-XSLWHU

2. Looking at the following conceptual design for RayBot 4.0, what features
and trade-offs can you anticipate for operating in the open ocean?

navigation
sensors
proximity
proximity

detector
detector

environmental
flexible sensors
body disk

pectoral fins
control shape of disk

flexible pelvic fins

pelvic fins change shape and function
as control surface

tail
oscillated and
turned by distributed
propulsive piezoelectric motors
tail

87
 3. Hatching new ideas for robotics companies is an entertaining way to
think about designing new kinds of robots. Identify three features of a
robotics company that would make it well poised to succeed.

0DQ\ SHRSOH ¿UVW JHWWLQJ VWDUWHG LQ GHVLJQ RU HQJLQHHULQJ GRQ¶W WKLQN
that the process will be iterative for them. They think that their design
SURFHVV FDQ EH D OLQHDU ÀRZ RI GHVLJQ
build, and test. Design

,I \RXUV ZRUNV WKH ¿UVW WLPH WKHQ \RX

might be deemed a genius, or you
might be accused of having had prior
experience with that particular design.
*RRGOXFN Build Test
Lecture 9—Designing a New Robot

88
 A Robot for Every Task?
Lecture 10

I
f we want a robot to perform a given task, we have choices about which
way to go. We could use a generalist robot, a robot that can do a large
range of tasks, including tasks for which it wasn’t originally designed
or programmed. We could redesign an existing specialist robot, to add to it
the new capability we want. We could create new specialist robots, such as
URERWLFWUDVKFDQVWKDWIRFXVRQRQHMRE7KHIDFWWKDWZHVRRIWHQKDYHWKHVH
FKRLFHVWHOOVXVDORWDERXWKRZWKH¿HOGRIURERWLFVZRUNV

Robots That Do Jobs

z 7RGHVLJQDURERWWRGRDMREWKHNH\LVWR¿UVWXQGHUVWDQGWKHMRE
0RVWRIWKHWLPHDMREORRNVVLPSOHDW¿UVWDQGWKHQWKHGHHSHU
\RX JHW LQWR LW WKH WULFNLHU LW EHFRPHV 8QGHUVWDQGLQJ WKH MRE
LQYROYHVXQGHUVWDQGLQJWZRSDUWVWKHZRUNSODFHDQGWKHWDVNLWVHOI

z 6RPHWLPHVWKHZRUNSODFHLVDVVSHFL¿FDVDEHGURRPLQDKRXVH
or it can be more general, such as any room in any house. The task
LV GRQH LQ WKH ZRUNSODFH \RX¶YH MXVW VSHFL¿HG <RX FDQ¶W KDYH D
task without a workplace. In that context, you need to be able to
break down a task, analyze it, and determine its subcomponents.
This is analogous to what we do when we write an algorithm for
the computer in our robot’s microcontroller. We need a step-by-step
understanding of what the task entails.

z In the case of taking out the trash, start with the workplace.
Imagine a laboratory. Think about the layout with regard to the
generation of trash. There is a small trash can that is kept in a
convenient location and a large trash can into which the contents
of the small trash can are dumped. When the large can is full,
then it is eventually emptied. In addition, there is some clutter
RQWKHÀRRULQFOXGLQJRWKHUURERWVDORQJZLWKDODUJHODEEHQFK
7KH ÀRRU LWVHOI LV KDUG DQG VPRRWK 7KHUH DUH QR VWDLUV RU RWKHU
environmental hazards for a robot.

89
 ‹2OJD6HUG\XNL6WRFN7KLQNVWRFN
%HFDXVHIDFWRU\ÀRRUVDUHVWDEOHZRUNVSDFHVURERWVGRQRWQHHGWREH
programmed to sense humans, freeing up energy and computer space.

z When we look at a workplace, a big deal in robotics is to think about

how stable it is. A stable workplace is called a structured workplace,
and it’s one in which the world doesn’t change. It maintains its
VWUXFWXUH 0RVW REMHFWV GRQ¶W PRYH DQG WKRVH WKDW GR GR VR LQ
predictable and regular ways, and people don’t come and go willy-
QLOO\$IDFWRU\ÀRRUZLWKDQDVVHPEO\OLQHLVDFODVVLFH[DPSOHRID
Lecture 10—A Robot for Every Task?

highly structured workplace. You can map out a structured workplace

on one day and have that map still be accurate on the next.

z On the other end of the spectrum are unstructured workplaces, which

change all the time and change in ways that are often not predictable.
Most busy, crowded places outside are unstructured. People, animals,
and vehicles come and go. Rain falls. Bicyclists ride by.

z A laboratory falls in between. It is structured in the sense that the

ZDOOVWDEOHDQGFRQ¿JXUDWLRQGRQ¶WFKDQJH,WLVXQVWUXFWXUHGLQWKH
sense that people and robots come and go. This kind of intermediate
structure is typical of most workplaces, such as houses, schools,
DQGRI¿FHV

90
 z Knowing the level of structure allows you to make important
decisions about what your robot needs to be able to sense—to keep
track of—and what it can ignore. If you can ignore having to identify
REMHFWVWKHQ\RXVDYHDORWRIWLPHHQHUJ\DQGFRPSXWHUVSDFH$
structured workplace allows you to build into the programming of
your robot implicit assumptions about the world.

Taking Out the Trash

z Consider the task of taking out the trash. This will be a series of
events. On some regular schedule or when it starts to get full, we
ZDQWWKHVPDOOWUDVKFDQWREHHPSWLHG,WVHHPVVLPSOH7RHPSW\
the trash, we pick it up, take it over to the large trash can, empty it,
and return it. The following is a list of subtasks involved in getting
DQDVRI\HWXQVSHFL¿HGURERWWRGRWKLVMRE
ż Turn on when trash needs to be emptied.

ż Navigate to the location of the small trash can.

ż Grasp the can.

ż Lift the can.

ż Move the can to the large trash can.

ż Lift the can higher.

ż Tilt the can over the large can.

ż (PSW\WKHFRQWHQWVLQWRWKHODUJHFDQ

ż Check to make sure that the small can is empty.

ż Return can to transport position.

ż Return can to its location.

91
 ż Home to charging or standby dock.

ż Await next set of instructions.

z What’s useful about task analysis is that we can immediately start

to see what we need the robot to able to do. We can start thinking
about what the robot needs in terms of sensors, actuators, and
sensory-guided movement.

z In terms of kinds of sensory-guided movement, the trash-emptying

robot must be able to navigate around the laboratory, and it must be
able to manipulate the trash can. That navigation and manipulation
DUHUHTXLUHGLVDJRRGVLJQ7KH\DUHWKHWZRIXQGDPHQWDONLQGVRI
EHKDYLRUVWKDWURERWVKDYHWRH[HFXWHLQQHDUO\HYHU\MRE\RXPLJKW
want a robot to do.

z $WWKLVSRLQWLQWKHGHVLJQSURFHVVLWLVFULWLFDOWRDVNWKLVTXHVWLRQ
Now that we know what needs to be done and where it will happen,
LVWKHUHDQH[LVWLQJURERWWKDWFDQGRWKHMRE"7KHSUDFWLFDOYHUVLRQ
RIWKLVTXHVWLRQLVWKLV,VWKHUHDQH[LVWLQJURERWWKDWFDQGRWKLVMRE
that is available for me to use?

z If the answer to either question is yes, then you want to get the robot
DQGKDYHLWGRWKHMREIRU\RX<RXZDQWWRVWRSGHVLJQLQJDURERW
Lecture 10—A Robot for Every Task?

for this task and put your efforts elsewhere, at least for the moment.
Don’t reinvent the wheel unless you have an idea for a better wheel.

z For the task of emptying the trash in the laboratory, the answer to
WKLVTXHVWLRQLV:H¶UHQRWVXUH6ROHW¶VLQYHVWLJDWHDQGVHHZKDW
robots out there might be able to do. If we think about navigation in
WKHODEDFOHDQLQJURERWOLNH5RRPEDGRHVDYHU\JRRGMRERIJRLQJ
out and cleaning.

z However, Roomba, for the most part, is not doing point-to-point

navigation. It’s not until we ask it to return to its dock that is has
D VSHFL¿F SODFH WR JR 5RRPED VKRZV XV KRZ ZH PLJKW VHW XS
our navigation system in the lab, with infrared beacons like what

92
 we have on the docking station. The beacon is like a lighthouse,
VHQGLQJRXWDVWHDG\VLJQDO:KHQ5RRPED¿QGVWKHVLJQDOLWFDQ
move directly toward it.

z 7RKHOS5RRPED¿QGWKHWUDVKFDQZHFRXOGWDNHRQHRIWKHYLUWXDO
wall lighthouses and use it as a beacon to help guide the robot to the
trash, which we could set up with the equivalent of a docking signal.

z :KLOH5RRPEDFRXOGJHWJRLQJDQGQDYLJDWHWRDFDQIXO¿OOLQJWKH
¿UVWRIRXUVXEWDVNVZHVHHWKDWLQWHUPVRIVXEWDVNJUDVSLQJ
WKH FDQ ZH DUH LQ WURXEOH 5RRPED FDQ¶W GR WKH MRE EHFDXVH LW
doesn’t have a manipulator to transport and then empty the can.

z 3HUKDSV LQVWHDG ZH FDQ ¿QG D URERW WKDW LV PRUH RI D JHQHUDOLVW
WKDW ZH PLJKW HDVLO\ SURJUDP WR GR WKH MRE :KHQ ZH WKLQN RI D
generalist, we humans tend to think of humans, because of the huge
number of different tasks that we can accomplish. So, we think of a
humanoid as a machine that could step right in and empty the trash.

z Baxter is an example of a humanoid that is built to be

reprogrammed—and reprogrammed easily, without having to
write any computer code. This is an advantage over Roomba. You
train Baxter by showing it what to do. Baxter’s ability to learn
by doing makes it one of the best multipurpose robots that is
commercially available.

z Baxter is not a mobile robot like Roomba; it doesn’t have legs or

wheels to move around the lab. So, neither Baxter, the versatile
manipulator, nor Roomba, the able-bodied navigator, can pick up
the trash and take it out. We need to combine the two.

z Clawbot has an arm, and it can move around. We can run Clawbot
by remote control to test what it can do. When we try to get it to
empty the trash, we discover that Clawbot lacks the kinematic
degree of freedom that would allow it to twist and get its end
effector, its claw, oriented so that it can pick up the can.

93
 z :KHQHYHU \RXU URERW FDQ¶W GR VRPHWKLQJ \RX KDYH WZR FKRLFHV
rebuild it or change the world (the workplace). In either case, we are
talking about designing a way to allow the robot to be successful. In
the case of Clawbot, we’d want to rebuild it by adding at least one
and probably two degrees of freedom to the wrist.

z A kinematic degree
of freedom in human
ERGLHV LQYROYHV MRLQWV
(DFKVLPSOHKLQJHMRLQW
LQ \RXU ¿QJHU DGGV
one rotational degree
of freedom to your

‹2OL6FDUII*HWW\,PDJHV1HZV7KLQNVWRFN
KDQG (DFK GHJUHH RI
freedom in the hand
means that you can
UH¿QH\RXUPRYHPHQWV
the things that you can
grasp and manipulate.

z The claw needs some Each hinge joint gives hands greater
added dexterity. We freedom and control.
need the claw to roll
GHJUHHVDQGWKHQSLYRWVRWKDWWKH¿QJHUVDUHRULHQWHGYHUWLFDOO\
Lecture 10—A Robot for Every Task?

DQGGRZQ7KHQLWFRXOGSLFNXSWKHFDQ:H¶GQHHGWZRPRUHMRLQWV
with motors, and we’d have to reprogram the software to allow this
to happen. The additional motors would add weight that the motors
support, so the payload maximum in the trash can would have to
decrease. Perhaps there’s a better way to go than rebuilding this robot.

z For either Clawbot or Baxter, this is when we wish that the arms
could telescope if and when we needed them to. This would add
a kinematic degree of freedom to the arm. This linear motion of
telescoping, such as opening a tripod leg, is called translation, and
it’s a very different kind of motion than we see in the vertebrate
body. The closest we get to telescoping in translation is sticking out
our tongue.

94
 z Imagine what it might be like to overcome the shortcoming of each
robot by getting them to work together. Clawbot is mobile with
very rudimentary grasping capabilities. Baxter has great grasping
capabilities but is not mobile. Combining their mobility and
manipulation might result in a great power couple.

z First, Clawbot is working by remote control. We’d want to program

it to operate autonomously like Roomba—to sit in a corner,
charging and waiting until scheduled or called to turn on. Then, it
would navigate to the small can, lift it, and move it to the large can.
Then, Baxter would take the can and empty it and hand it back to
&ODZERWZKLFKZRXOGWKHQGRWKHUHWXUQMRXUQH\

z Navigation and manipulation are central to taking out the trash.

These two abilities are central to a whole host of tasks that we
might want robots to do.

Suggested Reading

$QJHOHVDQG3DUN³3HUIRUPDQFH(YDOXDWLRQDQG'HVLJQ&ULWHULD´
Dym, Little, and Orwin, Engineering Design, chaps. 4–6.

Other Resources

3URIHVVRU 5XV¶V ZHEVLWH DW 0,7 KWWSJURXSVFVDLOPLWHGXGUOZLNLLQGH[

SKS"WLWOH 0DLQB3DJH

Questions to Consider

1. From what you know about robots, what tasks are robots particularly
suited to do—and to do well?

2. Professor Manuela Veloso is working on a type of common sense for

URERWVNQRZLQJ\RXUOLPLWV:KDWPLJKWEHWKHEHVWZD\WRSURJUDPD
robot to know its own limits?

95
 a. Anytime the robot fails to complete a step of a task in a certain
amount of time, it would then categorize that step (or the task) as
beyond its limits.

b. Anytime the robot is unable to carry out steps in the sequence in

ZKLFKWKH\DUHVSHFL¿HGLWZRXOGWKHQFDWHJRUL]HWKDWHQWLUHWDVNDV
beyond its limits.

c. A robot could run an internal model of itself, a simulation of the task

it is trying to complete. The model is updated by the robot’s current
position and current activity, as well as the state of the local world.
With these inputs to the simulation of itself, the robot predicts the
probability of success in conducting the next steps or tasks. Low-
SUREDELOLW\SUHGLFWLRQVDUHMXGJHGWREHEH\RQGLWVOLPLWV
Lecture 10—A Robot for Every Task?

96
 Robot Arms in the Factory
Lecture 11

M
ore than a million robots are at work in factories all over the planet.
Industrial robots help us manufacture cars, electronic and medical
devices, medicines and pharmaceuticals, food and beverages,
metals, chemicals, plastic, and rubber. Virtually all of these robots are not
humanoids; they are specialized robotic manipulators, and their workplace is
the highly structured, highly predictable world of the industrial factory. The
tasks of robotic manipulators involve repetitive, quick, and highly accurate
movements to grasp, move, manipulate, and assemble. With tools in hand,
WKHVHURERWV¿OHSDLQWDVVHPEOHFXWSROLVKÀDPHZHOGERQGJOXHVHDO
inspect, and sort.

The Anatomy of Autonomous Robots

z ,QGXVWULDOURERWVDOOEHJDQDVYDULDWLRQVRQDWKHPHDURERWLFDUP
5RERWLF DUPV KDYH D QDWXUDO DI¿QLW\ ZLWK KXPDQ DQDWRP\ 7KHLU
JUDVSHU FDQ EH WKRXJKW RI DV WKH ¿QJHUV RQ \RXU KDQG 7KH ZULVW
moves the grasper, and there are some elements that look like the
IRUHDUP DQG XSSHU DUP 7KH EDVH MRLQW FDQ EH WKRXJKW RI DV WKH
shoulder, and there also is an elbow.

z In engineering terms, a robotic manipulator is a linkage system,

ZLWK PRWLRQ OLPLWHG WR DQG FRQWUROOHG DW WKH MRLQWV 7KH OLQN
LV D ULJLG ERG\ WKDW PDLQWDLQV WKH GLVWDQFH EHWZHHQ MRLQWV DQG
often contains the motors that power the movement of the
QHLJKERULQJMRLQWV

z 7KH MRLQW RI D URERWLF PDQLSXODWRU LV XVXDOO\ RQH RI WZR W\SHV
UHYROXWH RU SULVPDWLF 5HYROXWH MRLQWV URWDWH RU UHYROYH DQG
SULVPDWLFMRLQWVVOLGH:HKXPDQVKDYHPRVWO\UHYROXWHMRLQWVDQG
PDQ\RIWKRVHMRLQWVUHYROYHDURXQGDKLQJHVRZHFDOOWKHPKLQJH
MRLQWV +LQJH MRLQWV VXFK DV \RXU HOERZ URWDWH WKURXJK D VLQJOH
kinematic degree of freedom.

97
 z Kinematics is the study of
motion without considering
the forces that generate those
motions. Include forces in
your study of motion and it’s
called dynamics. Kinematics
and dynamics are hugely
important in factory robotics
because we want a machine
whose motions can be

‹6WHYHQ)UDPH+HPHUD7KLQNVWRFN
precisely controlled.

z :KHQ D MRLQW KDV D VLQJOH

degree of freedom, in
kinematic terms, we say that
its position can be described
with a single mathematical Hinge joints rotate through a single
variable, which we typically kinematic degree of freedom and
can be described with a single
measure as an angle. For mathematical variable, which, for the
your elbow, its position can elbow, is approximately 150 degrees.
be described by the angle
between your forearm and your upper arm. In other words, that
single degree of freedom can rotate through an angle of about
Lecture 11—Robot Arms in the Factory

150 degrees.

z The heart of autonomous robotics in the factory is the design of

movements guided by sensors. Careful planning (or preplanning)
by humans is key to designing and controlling robots in the factory.
Anatomy is at once an inspiration and a challenge for roboticists.
Many motors make the predictable, accurate control of motion
H[WUHPHO\ GLI¿FXOW HYHQ ZKHQ \RX KDYH VHQVRUV WR PRQLWRU WKH
SRVLWLRQRIHYHU\MRLQW

z Designing robotic manipulators always involves trade-offs between

range of motion and control of motion. One common engineering
WULFN LV WR FRQVWUDLQ HYHU\ MRLQW WR EH D VLQJOH GHJUHH RI IUHHGRP

98
 This allows you to simplify the motion and its control. Then, you
can add degrees of freedom back in, with more single-degree-of-
IUHHGRPMRLQWV

z <RX FRXOG DOVR PDNH PRUH FRPSOH[ MRLQWV VXFK DV WKH EDOODQG
VRFNHWMRLQWEXW\RXZRQ¶W¿QGWKRVHRQPRVWURERWLFPDQLSXODWRUV
That’s because the fundamental function of a robotic manipulator is
VLPSOHSLFNDQGSODFHSLFNVRPHWKLQJXSDQGSODFHLWVRPHZKHUH
else. We humans have been doing this for as long as we’ve been on
two feet, picking fruit and wielding tools.

Industrial Robots
z 7KH ¿UVW UHFRUG ZH KDYH RI D SURJUDPPDEOH SLFNDQGSODFH
URERWLVIURP*ULI¿WK37D\ORU¶VURERWZDVSURJUDPPDEOH
and it picked up a series of bricks and created a circular stack,
automatically, in about 50 minutes. In total, Taylor’s robot had
¿YHGHJUHHVRIIUHHGRPWRDOORZLWWRSLFNDQGSODFHEULFNV6RPH
SHRSOH FRQVLGHU 7D\ORU¶V SLFNDQGSODFH URERW WR EH WKH ¿UVW
manipulator robot.

z As so often happens, it takes years for an idea for a robot to be

WUDQVODWHGLQWRFRPPHUFLDODSSOLFDWLRQV7KH¿UVW86SDWHQWIRUD
commercially successful industrial robot was granted in 1961, to
George Devol. The patent involved tracks along which the transfer
apparatus moved. Parallel to those tracks ran a conveyor belt. The
arm of the transfer apparatus reached across the conveyor belt,
grabbed cartons from the pallets, and then transferred those cartons
to the conveyor belt.

z A farsighted feature of Devol’s patent was a feedback loop

between the sensors detecting position and the program controller.
Devol’s patent envisioned the robot knowing that its arm or gripper
had gotten into the right position because a sensor detected that
motion. When it works, this is proprioception, although reliable
sensor-based movement turned out to be an ongoing challenge
for robotics.

99
 Industrial Robots

  QLPDWLRQ WKH ZRUOG¶V ¿UVW URERW FRPSDQ\ LV

8
IRXQGHGE\*HRUJH'HYRODQG-RVHSK(QJHOEHUJHU

  XPHULFDOO\ FRQWUROOHG 1& PDFKLQHV DUH ¿UVW

1
produced by FANUC in Japan.

 ³ 3URJUDPPHG $UWLFOH 7UDQVIHU´ 86 3DWHQW

2,988,237), key intellectual property for Unimation,
is issued to George C. Devol, Jr.

  QLPDWH WKH ¿UVW LQGXVWULDO URERW IURP 8QLPDWLRQ

8
goes to work unloading die casts at the General
Motors factory in Trenton, NJ.

 )DPXOXV WKH ¿UVW LQGXVWULDO URERW ZLWK VL[

electromechanically controlled axes, is introduced
by German robotic company KUKA.

±   WKH ¿UVW FRPPHUFLDOO\ DYDLODEOH URERW DUP

7
controlled by a minicomputer, is designed by Richard
Hohn and released by Cincinnati Milacron.

1976 Robot arms are used on Viking 1 and 2 space probes.

Lecture 11—Robot Arms in the Factory

1981 SCARA industrial robots are introduced in Japan.

1985 FANUC develops robots to assemble other robots.

z By the time that Devol received his patent in 1961, he had already
IRUPHGDFRPSDQ\LQZKHQKH¿UVWDSSOLHGIRUWKHSDWHQW
7KHFRPSDQ\IRUPHGZLWKVHULDOHQWUHSUHQHXU-RVHSK(QJHOEHUJHU
ZDV FDOOHG 8QLPDWLRQ DQG LW EHFDPH WKH ¿UVW FRPSDQ\ WR
build robots that went to work for industry. Those robots were
called Unimates.

100
 z 7KH¿UVW8QLPDWHURERWZDVLQVWDOOHGLQLQD*HQHUDO0RWRUV
factory. It used the Unimate to lift hot pieces of metal from a die-
FDVWLQJPDFKLQHDQGVWDFNWKHP7KHVH¿UVWLQGXVWULDOURERWVOHGWR
KUKA robots, and many others. These classic industrial robots—all
very much like the original Unimates—are still used in thousands
of factories worldwide.

Range of Motion versus Control of Motion

z Accurate and repeatable control of robotic manipulators was critical
for their development, and two different ways of thinking about
PRYHPHQWV GHYHORSHG IRUZDUG DQG LQYHUVH NLQHPDWLFV ,Q ERWK
approaches, we are interested in the position and orientation of the
end effector, the hand.

z In forward kinematics, we can compute the end effector’s position

E\PHDVXULQJWKHMRLQWDQJOHVDQGNQRZLQJWKHOHQJWKVRIWKHOLQNV
Trigonometry will do the trick mathematically, and it allows us to
WUDQVIRUPZKDWZHFDOOWKHMRLQWVSDFHRIDQJOHVLQWRWKH&DUWHVLDQ
space of the x, y, z position of the effector. With forward kinematics,
it is easy to move the robot and then calculate its position.

z In inverse kinematics, we pose the problem the other way around.

7KLV PD\ EH PXFK PRUH GLI¿FXOW WR GR :H ¿UVW LGHQWLI\ WKH
position that we want the end effector to occupy, and perhaps some
constraints about the path to take, and then calculate the movements
RI WKH DUP WKDW ZLOO GR WKH MRE ,QYHUVH NLQHPDWLFV LV WKH IRUPDO
way to wrap our heads around the problem of the trade-off between
range of motion and control of motion.

z Given this trade-off, one of the most important things to consider

when building or buying an industrial robot is the minimum number
RIGHJUHHVRIIUHHGRPWKDW\RXQHHG(YHU\WLPH\RXLQFUHDVHWKH
range of motion by adding a degree of freedom, you exponentially
increase the computational problem of controlling that motion.
In practice, the range of possible solutions is usually obtained by
imposing additional constraints.

101
 z So far, we have mostly been considering what might be a
numerically controlled machine tool, or a blind robot in an open
control loop. But if we want a robot that knows whether its arm is
in the right place, then this is when sensors come into play. Sensors
FDQPHDVXUHWKHDFWXDOSRVLWLRQRIHDFKMRLQWDQGSURYLGHIHHGEDFN
WRWKHFRQWUROOHUIRUSRVVLEOHDGMXVWPHQWV

z The sensors that robotic manipulators use to measure the position

RIWKHLUMRLQWVDQGHQGHIIHFWRUVYDU\,IWKHMRLQWVDUHSRZHUHGE\
servomotors, the sensory control is actually built into the motor.
Inside the servomotor is the electric motor and the equivalent of
what we call a potentiometer, a variable resistor that signals angular
position by changing voltage in a circuit.

z Many industrial robots use stepper motors, which are more

powerful and precise than servomotors. Stepper motors don’t have
built-in sensors, so you have to use external sensors. Servomotors
DQGVWHSSHUPRWRUVPHDVXUHDQGDGMXVWSRVLWLRQFRQWLQXRXVO\

z From the perspective of range of motion, one of the amazing

things about adding a degree of freedom is that by the laws of
combinatorics, the number of possible positions, or states, of our
system explodes. This increased range of motion, plus control over
PRWLRQWKDWLVPRUHÀH[LEOHJLYHVWKHURERWZKDWZHFRPPRQO\
Lecture 11—Robot Arms in the Factory

call dexterity. The challenge is that as we increase range of

motion and dexterity, the complexity of our control system has to
increase, too.

z In 1973, the KUKA robot company made headlines when they

LQWURGXFHG WKH ¿UVW LQGXVWULDO URERW ZLWK VL[ GHJUHHV RI IUHHGRP
The robot is called Famulus, and it was built for automotive
manufacturing, as about a third of the industrial robots are still
today. Famulus is an articulated robot, which means that it has
DQ\QXPEHURIUHYROXWHMRLQWV<RXZLOOVHHDUWLFXODWHGURERWVLQWKH
food industry.

102
z A very different manipulator is a Cartesian robot, which has three
SULVPDWLFMRLQWV²WHOHVFRSLFMRLQWV²WKDWDUHDUUDQJHGSHUSHQGLFXODU
WRHDFKRWKHU%HFDXVHWKHMRLQWVDUHVOLGLQJWKH\FDQDOVREHPDGH
into very long tracks, into what are called gantry robots.

z $ VSKHULFDO RU SRODU URERW UHSODFHV WKH UHYROXWH HOERZ MRLQW RI DQ
DUWLFXODWHG URERW ZLWK D SULVPDWLF MRLQW 7KH WHUP ³VSKHUH´ UHIHUV
to the shape of the working space of the end effector. The Unimate
URERWV ZHUH VSKHULFDO URERWV ZLWK D WHOHVFRSLF MRLQW WKDW DOORZHG
them to reach for a sample or weld a particular spot and then retract
to get out of the way.

z %HFDXVHVSKHULFDOURERWVKDYHMXVWWKUHHGHJUHHVRIIUHHGRPWKH\
too, are easier to control than articulated robots, which usually have
six. But because they lack a large range of motion, we tend not to
see many spherical robots in operation these days.

z In between the spherical robot and the Cartesian robot is the

cylindrical robot, where the working space is a cylinder. To get
WKLVVKDSHWKH¿UVWMRLQWQHHGVWREHSULVPDWLFUDLVLQJWKHDUPXS
DQG GRZQ$ UHYROXWH MRLQW VSLQV WR FUHDWH D FLUFOH DQG D VHFRQG
SULVPDWLFMRLQWWHOHVFRSHVDUHYROXWHZULVWMRLQW&\OLQGULFDOURERWV
are used in applications like spot welding.

z A big breakthrough for simplicity came in the late 1970s, when

WKH¿UVW6&$5$VHOHFWLYHFRPSOLDQFHDVVHPEO\URERWDUPURERWV
were developed, supposedly inspired by a Japanese-style folding
screen. The trick was in redesigning the work into a series of
simpler tasks to suit the simpler robot.

z 6&$5$ URERWV KDYH WKUHH UHYROXWH MRLQWV EXW XQOLNH RUGLQDU\

DUWLFXODWHGURERWVDOOWKUHHMRLQWVRSHUDWHLQWKHVDPHSODQHVRWKDW
SCARA robots can move the end effector very precisely, more
precisely than articulated robots. The end effector is located on a
IRXUWK MRLQW D SULVPDWLF MRLQW WKDW¶V SHUSHQGLFXODU WR WKH SODQH RI
WKHUHYROXWHMRLQWV

103
 z %HFDXVHWKHWKUHHUHYROXWHMRLQWVDUHQ¶WUHVWULFWHGLQVL]HE\QHHGLQJ
to move perpendicular to each other, SCARA robots can also have
YHU\ ODUJH PRWRUV WR SRZHU WKH MRLQWV PDNLQJ WKHP VRPH RI WKH
fastest robotic manipulators available. SCARA robots are very
JRRG IRU WKLQJV OLNH WKH VXUSULVLQJO\ GLI¿FXOW WDVN RI SODFLQJ DQG
tightening screws.

z :KLOH WKH HDUO\ LQGXVWULDO URERWV GLG WKH VDPH MRE DOO GD\ GD\
DIWHU GD\ D EUHDNWKURXJK IRU ÀH[LELOLW\ FDPH ZLWK WKH 380$
(programmable universal machine for assembly) robots. These
were programmable robots that were developed at Stanford and
supported by General Motors.

z Finally, the world’s fastest robotic manipulators are typically

parallel robots. The idea is to build multiple arms in parallel,
ZRUNLQJ WRJHWKHU OLNH SDLUV RI DUPV 7KH$GHSW 4XDWWUR KDV IRXU
arms in parallel, working for precise, rapid movements.

Collaborative Robots
z Because robotic manipulators work very rapidly, often with a
lot of power, part of the design of the robot is to also design the
workplace. Safety is a big consideration. In 1979, in a Ford Motor
FDVWLQJSODQW5REHUW:LOOLDPVEHFDPHWKH¿UVWKXPDQNLOOHGE\D
Lecture 11—Robot Arms in the Factory

factory robot. Reports say that the robot had begun moving more
slowly, and Williams tried to climb into a storage rack to remove
parts by hand when he was hit in the head by the one-ton robot and
killed instantly.

z The standard way to make it safe for humans and robots to work
together is to keep them from working together, at least physically.
The work areas for industrial robots in traditional factories are very
FOHDUO\GH¿QHGDQGKXPDQVDUHNHSWRXWRIWKRVHDUHDVZKLFKDUH
called safety cages.

104
 z But these restrictions limit the work that robots do. Flexible,
easy-to-use robots that can work alongside humans were goals
for Rodney Brooks when he created the company that is now
called Rethink Robotics, which created the robot Baxter, released
commercially in 2012.

z $VRQHRIWKH¿UVWFROODERUDWLYHURERWVRUFRERWV%D[WHULVGH¿QLQJ
new roles for industrial robots and helping to bridge from industrial
robotics to service robotics, where robots are designed, from the
start, to interact safely with humans.

z Other robotics companies are also working to make robots more

collaborative with humans. KUKA is making slower, lightweight
URERWVZLWKFRPSOLDQWMRLQWVIRUZRUNDVservice robots.

Important Terms

articulated robot$PDQLSXODWRURUDUPWKDWKDVKLQJHUHYROXWHMRLQWV

Cartesian coordinate robot$QLQGXVWULDOURERWZLWKWKUHHSHUSHQGLFXODU

translational degrees of freedom.

gantry robot/DUJHW\SHRI&DUWHVLDQFRRUGLQDWHPDQLSXODWRUURERWZKRVH
degrees of freedom are in translation and at right angles to each other.

kinematics 7KH VWXG\ RI PRWLRQ ZLWKRXW UHJDUG WR WKH IRUFHV JHQHUDWLQJ
them.

proprioception ,QWHUQDO VHQVLQJ RI WKH PRWLRQ DQG IRUFH RI WKH URERW¶V
MRLQWVDQGPRWRUV

service robot Any robot built to assist humans, excluding robots involved
in manufacturing.

105
 Suggested Reading

$QRQ\PRXV³$Q$XWRPDWLF%ORFN6HWWLQJ&UDQH´
Craig, Introduction to Robotics, chap. 1.
Devol, Programmed article transfer, U.S. Patent 2,988,237.

Other Resources

5HWKLQN 5RERWLFV PDNHU RI %D[WHU KWWSZZZUHWKLQNURERWLFVFRP

SURGXFWVED[WHU

Question to Consider

1. The following is a robotic arm produced by the company ABB for the
National Museum of Scotland. Visitors type in their name on a keyboard,
and then the robot picks up blocks, deposits them on the metal shelf in
the lower-right side of the picture, and that spells the person’s name.
Lecture 11—Robot Arms in the Factory

© John Long.

What capabilities must the robotic arm have in order to do this

spelling task?

106
 Mobile Robots at Home
Lecture 12

S
ervice robots can be used for personal or professional use. Most
RI WKH WLPH GLIIHUHQW MREV DUH GRQH E\ GLIIHUHQW URERWV WKDW DUH
VSHFLDOL]HGIRURQHMRE)RUURERWVWKDWDUHGHVLJQHGWRZRUNDWKRPH
three features are proving to be the keys to success of personal service
URERWV 7KH\ DUH VSHFLDOL]HG IRU RQH MRE DUH RI VPDOO VL]H DQG SRVVHVV
autonomous mobility. An important part of being a specialist is that the
URERWGRHVQ¶WQHHGWREHUHSURJUDPPHG6SHFLDOLVWVKDYHRQHMREWRGRDQG
they do it out of the box.

Roomba
z 5RRPEDV ZHUHQ¶W WKH ¿UVW KRPH URERW YDFXXP EXW WKH\ ZHUH WKH
affordable and reliable machine for personal use that transformed
lives. Like any successful commercial product, Roomba has attracted
competitors—including Neato, Navibot, Hom-Bot, bObsweep,
Deebot, and the traditional vacuum makers, and Roomba itself has
undergone a regular series of updates and improvements.

z In 1997, iRobot made a Roomba prototype called Dust Puppy,

which was more like a dust mop than a vacuum cleaner. But Dust
Puppy showed that you could use very simple behavior-based
control algorithms to have a mobile robot navigate around your
KRPHDQGJHWDMREGRQH

z %\WKH¿UVWPRGHORI5RRPEDZDVVZHHSLQJDQGYDFXXPLQJ
One of the important design features is that Roomba goes where you
GRQ¶WVHHRUQRUPDOO\JRXQGHUIXUQLWXUH7KLVH[SODLQVDGHFLVLRQ
WKDWWKH\PDGHDERXWWKHGHVLJQ,WKDGWREHVPDOOLQKHLJKWWR¿W
under furniture.

107
 z In 2004, iRobot introduced the second generation of Roomba,
originally called Discovery but then renamed the 400 series.
Roomba 400s have enhanced autonomous navigation. These
Roombas can also be programmed to clean on a schedule, and they
have a bigger dustbin than the original models.

z In 2007, iRobot introduced the 500 series. In this third generation,

iRobot added automatic docking for autonomous recharging of
batteries, as well as the virtual lighthouse, an infrared beacon that
can speed Roomba’s navigation between rooms or help keep it
from going into a room, depending on the setting. These Roombas
also move faster and need forward-looking infrared sensors to
prevent high-speed collisions. They also avoid getting caught
on cords by reversing the rotational direction of the brushes if
resistance is encountered.

z The fourth-generation 600-series Roombas were introduced in 2008,

and they have a new kind of cleaning head that doesn’t get hair
entangled. Instead of having bristles, the brushes have treads and are
called extractors. These extractors are part of a redesign that includes
DKLJKHUYHORFLW\RIDLUÀRZVXFNLQJXSGLUWDQGGHEULV7KHVDOVR
have an improved acoustic sensor for detecting dirt, which allows
5RRPEDWRVSHQGPRUHWLPHRQUHDOO\VRLOHGSDUWVRIWKHÀRRU

z ,QWURGXFHGLQWKHVHULHVKDVEHWWHUEDWWHU\OLIHD+(3$
Lecture 12—Mobile Robots at Home

¿OWHURQWKHYDFXXPDQGZLUHOHVVVFKHGXOLQJ,QWURGXFHGLQ
the 800 series has even better battery life and cleaning capacities,
LQFOXGLQJDQRSWLFDOFDPHUDWRORRNIRUÀXII\GHEULV

Scooba and Braava

z ,QL5RERWLQWURGXFHGLWVVHFRQGKRPHURERW6FRRED,WVMRELV
WRVFUXEÀRRUVDQGWKHPRGHOLQWURGXFHGLQXVHVDWKUHH
SDVVFOHDQLQJSURFHVVWKDWLQFOXGHVYDFXXPLQJDVWKH¿UVWVWHS$VLW
YDFXXPV6FRREDOD\VGRZQDWKLQ¿OPRIFOHDQLQJÀXLGWRSUHVRDN
WKHÀRRU2QWKHVHFRQGSDVVLWVFUXEVDQGVTXHHJHHVVXFNLQJXS
WKHGLUW\VROXWLRQ7KH¿QDOSDVVFRQWLQXHVWRUHPRYHWKHOLTXLG

108
z Scooba has the same basic circle-shaped design as Roomba. A
FLUFXODU GHVLJQ LV JUHDW IRU D ÀRRUFOHDQLQJ URERW LQ WKH KRPH
because a circle lacks edges to ding and get caught. The short
stature, too, is great for getting under furniture. In other words, the
robot body is built to get the robot under and around.

z An exception to the circular body design is found in iRobot’s

Braava, a robot that does either dry or damp mopping. While
behavior-based control architectures are simple and robust, how
WKH\FOHDQLVLQHI¿FLHQW6RPHDUHDVJHWPRUHFRYHUDJHWKDQRWKHUV
To make sure that most areas are covered, Roomba and Scooba are
programmed to make multiple passes. That takes more time and
battery power. Thus, the trade-off for behavior-based cleaning is
VLPSOLFLW\DQGUREXVWQHVVYHUVXVHI¿FLHQF\

z 7R FUHDWH DQ HI¿FLHQW RQHSDVV FOHDQLQJ URERW L5RERW QHHGHG WR
create a robot that could navigate using a map. GPS isn’t a good
option inside for two reasons. First, the signals from satellites can
be scattered, distorted, and blocked by walls and metal structures.
6HFRQG HYHQ LI \RX FDQ JHW D JRRG *36 ¿[ WKH DFFXUDF\ RI WKH
WUDGLWLRQDO FRQVXPHU ¿[ HYHQ LQ WKH EHVW RXWGRRU FRQGLWLRQV KDV
EHHQ PRUH WKDQ  IHHW ZKLFK LV QRW VXI¿FLHQW DFFXUDF\ WR JXLGH
indoor robots.

z (YROXWLRQ 5RERWLFV EXLOW D QDYLJDWLQJ URERW FDOOHG 0LQW :KHQ

L5RERW DFTXLUHG (YROXWLRQ 5RERWLFV LQ  0LQW ZDV UHQDPHG
Braava. Braava works using an infrared navigation system called
1RUWK6WDU ZKLFK XVHV D EHDFRQ WKDW VLWV RQ D WDEOH DQG SURMHFWV
infrared dots onto the ceiling. Braava uses three infrared sensors to
¿QGLWVGLVWDQFHIURPWKRVHGRWVDQGWULDQJXODWHLWVSRVLWLRQ

z When iRobot patented their own version of NorthStar’s infrared

WHFKQRORJ\ LQ  WKH\ FDOOHG LW ³FHOHVWLDO QDYLJDWLRQ´ %UDDYD
uses infrared navigation common to NorthStar and celestial
navigation to create its own map of the room.

109
 z Knowing its position on the map, Braava can plot a course that
PRYHVLWEDFNDQGIRUWKDFURVV\RXUÀRRULQEHDXWLIXOWLJKWURZV
This gives complete coverage in one pass when Braava is dry
mopping. In damp mopping mode, Braava still navigates, but it
moves with side-to-side and back-and-forth scrubbing motions as it
moves along the rows.

z Creating a map and knowing where you are on the map at the same
time is hugely important task in autonomous robotic navigation. It is
called simultaneous localization and navigation. This is a model-
based controller architecture that is computationally intensive.

z Now we can understand the square design. If you know where

you are and where other things are on your map, you can plot your
FRXUVHWRDYRLGREMHFWV,I\RXNQRZZKDWREVWDFOHVOLHDKHDGWKHQ
a square design is less likely to get you stuck in a corner. You avoid
the corner, and you can also aim deliberately into the corner. Also,
Braava can put the cleaning pad out front, which makes it wider
than the body. A wider cleaning pad means fewer passes—even
EHWWHUHI¿FLHQF\

z However, there are a few trade-offs with Braava. First, because it

has a wide mop out front, it doesn’t head straight into a docking
station. Second, in spite of having a map, Braava does sometimes
Lecture 12—Mobile Robots at Home

get stuck in small areas where it can’t maneuver itself out. Third,
Braava has to have a good signal from the NorthStar infrared
navigation system in order to clean an entire space. Once the
cleaning cycle has started, you also have to take care to not move
the NorthStar beacons; the navigation system has to be tuned to
make the system work optimally.

z 7KHELJWUDGHRIIIRU%UDDYDLVEHWZHHQHI¿FLHQF\DQGUREXVWQHVV
When it works optimally, Braava can navigate and clean very
HI¿FLHQWO\%XWEHFDXVHLWLVGHSHQGHQWRQPDSEDVHGQDYLJDWLRQLW
is not as robust as Scooba. Scooba and Braava clean differently and
have different trade-offs.

110
Mirra
z Another popular type of home robot works outside, in your pool.
Mirra 530 is an autonomous pool-cleaning robot built by iRobot.
Mirra has wheels for locomotion and a rotating scrub brush. It also
¿OWHUV ZDWHU ZKLFK LW VXFNV LQ WKURXJK DQ RSHQLQJ XS IURQW VR LW
removes both small particles and large items, such as leaves. Most
impressively, Mirra can move up vertical walls and navigate steps.
,WVZKHHOVKHOSJLYHLWWUDFWLRQDQGWKHVXFWLRQLWFUHDWHVIRU¿OWHULQJ
water helps it hold its ground.

z Unlike Roomba, Scooba, or Braava, Mirra has a tether. The cable

is a low-voltage energy source from a topside transformer that is
converting alternating current from the household energy supply to
the local direct current that Mirra uses. Low-voltage direct current
is very safe to use around water and people.

z :K\ QRW MXVW XVH D EDWWHU\ IRU 0LUUD" :KDW¶V WKH WUDGHRII ZKHQ
we use a battery? On the positive side, being untethered means that
the robot won’t get tangled if it goes under and through the legs
of a chair, and it won’t have stay in close proximity to its external
energy supply. On the negative side, batteries have a limited time
for which they can supply electric power. To extend the robot’s
mission time, a bigger battery can be used, but bigger batteries
increase the weight that the robot has to lug around.

z 7KH GHVLJQHUV DW L5RERW KDG D FKRLFH *LYH 0LUUD D ELJ HQRXJK
battery to handle the high power needs of its three motors over
HQRXJKWLPHWRFRPSOHWHWKUHHMREV²VFUXEORFRPRWHDQG¿OWHU²
or power Mirra through a tether with a topside energy supply.

z A big battery, being full of dense metals, would be great for keeping
Mirra on the bottom of the pool. The problem would be climbing
the walls, when the weight of the battery would tend to peel Mirra
off. The way to counteract this would be to create more suction. But
PRUHVXFWLRQPDNHVLWPRUHGLI¿FXOWWRPRYH<RXFRXOGLQFUHDVH
power to the motors driving the treads, but then that’s more power
you need. This is a vicious circle.

111
 z By giving Mirra a tether for its energy supply, the designers were
able to give Mirra the power to climb walls and even scrub right
at the water line, where the robot has to support part of its weight
out of the water. More importantly, Mirra’s effective time on station
was increased, with time limited only by having to stop and clean
WKH¿OWHUV

z To keep the tether from becoming tangled, Mirra has a gyroscopic

sensor on board. Gyros measure accelerations that come from
gravity and turning. This allows Mirra to keep track of how it
is oriented and how many times and in what direction it has
turned. After a certain number of turns in one direction, Mirra is
programmed to turn back. This removes coils in the tether and
prevents tangling.

Other Types of Service Robots

z A gutter is a great example of a structured workplace. Gutters vary
DOLWWOHLQVKDSHDQGZLGWKEXWWKHRYHUDOOMRELVSUHGLFWDEOH:KDWLV
not predictable is what kind of clutter you have to clean out. But if
\RXUMRELVWRJHWVWXIIRXWRIWKHJXWWHUWKDWLVDFRPPRQIXQFWLRQ
across all types of debris.

z $URERWWKDWGRHVWKLVMRELV/RRMWKHJXWWHUFOHDQLQJURERW,WVSLQV
an auger and then uses tank track actuators to push that actuator
Lecture 12—Mobile Robots at Home

along. An auger is usually a type of end effector for drilling, but

here the idea is to take the same rotary action and use it to move
material to the sides rather than to the rear.

z A much less structured outdoor workplace is a lawn. There are

stable but irregular obstacles, such as plants and sheds, and there is
terrain of different types, with rocks, hills, steps, pools, and ponds.
+RZHYHUODZQPRZLQJLVDMRESURYHQWREHZHOOVXLWHGWRURERWV
In fact, after robotic vacuum cleaners, robotic mowers are the
second largest group of home robots in use.

112
 z $QRWKHU KRXVHKROG MRE WKDW
robots can do is clean the
windows. Introduced in 2011,
:LQERW FUHDWHG E\ (FRYDFV
has an actuator system that
creates suction that keeps it
on the glass. It has infrared
proximity sensors that allow
it to calculate the size of

‹6HDQ*DOOXS*HWW\,PDJHV1HZV7KLQNVWRFN
the window, and then, like
Roomba, it calculates a plan to
clean the whole area.

z (IIHFWLYH DQG VXFFHVVIXO KRPH

robots thus far have been small,
mobile systems that specialize.
But as larger robots appear that In the near future, we might see
can move safely with us in our larger, more humanlike robots that
homes and other spaces, size perform a wide variety of tasks.
alone will open up new kinds of
MREVIRUWKHP$VKRPHURERWVDQGRWKHUVHUYLFHURERWVJDLQPRUH
skills, and perhaps even a humanoid form, they, like Baxter, will
move from narrow specialists toward becoming broader generalists.
We’ll increasingly be able to train our robots easily and quickly to
do new tasks that are unique to our needs and spaces.

Important Terms

acoustic sensor$OVRNQRZQDVDPLFURSKRQHDVHQVRUWKDWFRQYHUWVVRXQG
waves into electric signals that can be read by the controller.

simultaneous localization and mapping (SLAM) 6XE¿HOG RI QDYLJDWLRQ

in which the robot uses sensors and communication systems to know its
position as it creates a map of its surroundings.

113
 Suggested Reading

L5RERW³L5RERW©2ZQHU¶V0DQXDOVDQG4XLFN6WDUW*XLGHV´
;X4LDQDQG:XHousehold Service Robotics, chap. 3.

Other Resources

3URPRWLRQDOYLGHRRIWKHVHULHVIRU5RRPEDKWWSZZZLURERWFRPXV
OHDUQKRPHURRPEDDVS[

+DFNLQJ5RRPEDKDFNLQJURRPEDFRP:KLOHWKLVDSSURDFKLVDELWGDWHG
if you have an old Roomba, then get started here.

L5RERW¶VRI¿FLDOKDFNDEOH5RRPEDL&UHDWHKWWSZZZLURERWFRP$ERXW
L5RERW67(0&UHDWHDVS['HVLJQHGWRLQWHUIDFHZLWKFRQWUROOHUVOLNHWKH
Arduino Uno, this is iRobot’s response to people wanting to hack Roomba.

Question to Consider

1. In the lecture, we came up with this prediction of Roomba’s subsumption

EHKDYLRU,IWKHEXPSVHQVRULVRQEDFNXSDQGWXUQHOVHLIWKHOLJKW
touch sensor is on, keep moving forward. What is a good way to test
this prediction?
Lecture 12—Mobile Robots at Home

114
 Hospital Robots and Neuroprosthetics
Lecture 13

R
obots working in the hospital show an incredible diversity of types,
from mobile robots carrying materials, to robotic manipulators in
the hands of surgeons, to prosthetics using neural signals of the
KXPDQERG\5RERWVLQKRVSLWDOVZRUNLQDWOHDVWWKUHHGLIIHUHQWZRUNSODFHV
the busy, social world of bustling humans in the entrance hall, waiting areas,
and patient wards and rooms; the carefully controlled world of the surgical
theater; and the intimate world of prosthetics and exoskeletons used by
an individual.

Courier Robots
z In January of 1991, a HelpMate robotic courier was installed in
the Danbury Hospital in Connecticut by Transitions Research
&RUSRUDWLRQ D FRPSDQ\ IRXQGHG E\ -RVHSK (QJOHEHUJHU DIWHU KH
VROG 8QLPDWH WR :HVWLQJKRXVH LQ  +HOS0DWH ZDV WKH ¿UVW
hospital courier robot, designed to transport items and information.

z With funding from NASA, Transitions Research Corporation

wanted to tackle the problem of delivering material in the relatively
structured world of a hospital. So, in addition to using a model-
based navigation system for hallways and elevators, they needed
+HOS0DWHWREHDEOHWRGHWHFWREMHFWVDQGDYRLGWKHP

z To navigate, HelpMate used a map-based control system and

D YDULHW\ RI VHQVRUV WR DYRLG SHRSOH DQG RWKHU PRYLQJ REMHFWV
+HOS0DWHJRWDURXQGWKHGLI¿FXOWSUREOHPRIXVLQJDPDQLSXODWRU
to open doors by sending a radio signal to each door to open
LWVHOI(OHYDWRUVZHUHDOVRPRGL¿HGWRUHVSRQGWRUDGLRFRQWUROE\
the robot.

115
 z As a courier, HelpMate could deliver late meals and special dietary
foods. It could replenish supplies to nursing stations or return
samples collected by the nurses to the laboratory for analysis.
2XW¿WWHGZLWKDORFNHGFDELQHW+HOS0DWHFRXOGVHFXUHO\PRYHDQG
deliver medications from the pharmacy, along with patient records
stored on its computer. Up to 200 pounds of mail and packages
could be delivered during a single trip.

z HelpMate used lights to signal its intent to humans, move through

the door, and then signal to turn. In addition, HelpMate pioneered
the use of three types of sensors to safely interact with people and
WKLQJVVRQDUYLGHRDQGEXPSVHQVRUV

z Sensing with sonar was accomplished with 28 transducers on the

robot. These transducers send out an ultrasonic signal and measure
the time for that sound to return. A short WLPHRIÀLJKW signals that
DQREMHFWLVFORVHE\

z 5RERW YLGHR KDV KLVWRULFDOO\ EHHQ YHU\ GLI¿FXOW WR LPSOHPHQW LQ
Lecture 13—Hospital Robots and Neuroprosthetics

part because there is so much information to process. The team

at Transitions Research Corporation working on HelpMate in the
1980s employed two very creative engineering solutions.

z First, they drastically reduced the amount of information that they

were trying to get from the camera. Instead of trying to detect and
FODVVLI\ DQ\ DQG DOO REMHFWV WKH\ RQO\ VRXJKW WR GHWHFW WKH RQHV
right in front of the robot. With detection alone as the primary goal,
WKH\ FRXOG VHFRQGDULO\ FDOFXODWH WKH GLVWDQFH WR WKH REMHFW DQG LWV
VL]H%XPSVHQVRUVSURYLGHGDWKLUGZD\WRGHWHFWREMHFWVWKHQRZ
classic way, pioneered in a service setting by HelpMate, of letting
FRQWDFWZLWKDQREMHFWGHSUHVVDVLPSOHHOHFWULFVZLWFK

z Why have three different sensor systems delivering information

DERXWREMHFWV"$VLQJOHVHQVRUV\VWHPFDQJLYH\RXDQDPELJXRXV
answer. For example, if you relied on the vision system alone,
\RXZRXOGRQO\GHWHFWREMHFWVGLUHFWO\LQIURQWRI\RXDQGQRWKLQJ

116
 overhanging, such as a cantilevered countertop. In that case, the
28 sonar sensors, all sending out sound signals, would come
in handy.

z To improve your ability to get the truth about the world, to

disambiguate, you combine information from the redundant sensor
systems in a process called sensor fusion, which requires that the
URERWEHSURJUDPPHGWRJREH\RQGVLPSOHUHÀH[LYHUHDFWLRQVDQG
make a decision about when to trust one signal and when to infer the
most likely situation using multiple sources of sensory information.

z HelpMate’s solution to the problem of navigational drift—small

HUURUV WKDW DGG XS RYHU WLPH²ZDV WR JHW D QDYLJDWLRQDO ¿[
E\ SXWWLQJ UHWURUHÀHFWLYH WDSH RQ WKH FHLOLQJ DW ¿[HG LQWHUYDOV
HelpMate could sense those bands of tape, using a pair of long-
range infrared sensors that continuously emitted infrared light
upward, and then marked the peak intensities of infrared light
UHÀHFWHGRIIWKHEDQGV

z One of the challenges that this system did not initially anticipate
was the strange or irregular behavior of people. For example, people
behaved differently in different parts and times of the hospital.
Context cues were added to HelpMate’s control system, providing
UXOHVIRUKRZWRDGMXVWEHKDYLRUGHSHQGLQJRQORFDWLRQDQGWLPH

z HelpMate ceased production in 2006, but its success as an

intelligent, mobile courier has spawned a variety of robots in the
halls of hospitals, including cleaning robots, disinfection robots,
DQG URERWV WKDW FDQ FKHFN WKH VWDWXV RI GH¿EULOODWRUV DQG ¿UH
extinguishers.

z One of the trade-offs for courier robots like HelpMate is

maneuverability versus carrying capacity. A very maneuverable
courier robot is Swisslog’s RoboCourier, which can turn in place
like a Roomba, with a turning radius of zero. But it has a maximum
payload of 66 pounds.

117
 z By contrast, Aethon introduced a hospital delivery robot capable of
carrying 500 or even 1,000 pounds of supplies. These are called
TUG robots. The slim robot up front pulls along a large storage
cabinet behind. And instead of needing tape on the ceilings, or other
markers, Aethon chose to create preplanned travel routes for the
robots, either using computer-aided design drawings of the hospital
uploaded into the robot or by using lasers to manually guide the
robot through preplanned travel routes in the facility.

z 7KH78*URERWVDYRLGGULIWE\WDNLQJWKHLUQDYLJDWLRQDO¿[IURPWKH
charging stations. The company developed software that is said to
open any public elevator in the world. But because the TUG is less
PDQHXYHUDEOH LW PD\ JHW VWXFN ,Q WKDW FDVH RQH WKLQJ WKH 78*
URERWFDQGRLVDVNQHDUE\KXPDQVWRWDNHVSHFL¿FVWHSVWRKHOS

z Moreover, thanks to increasing computing power, the company

can link the TUG robots with an automated tracking system for
materials moving within the hospital. In fact, the TUG robots
are also monitored by automatic algorithms from the company’s
Lecture 13—Hospital Robots and Neuroprosthetics

Cloud Command Center, and if a problem is detected, a human

is contacted who can connect to the robot remotely and even take
remote control if necessary.

Telepresence Robots
z As good communication connections continue to improve, there
are increasing opportunities for telepresence robots. iRobot, the
company that made Roomba, teamed up with Cisco to create Ava, a
platform for a mobile service robot capable of wireless telepresence.
7KLVPDNHVSRVVLEOHDURERWWKDWFDQPDSDQGWUDYHODÀRRUEXWDOVR
engage in desktop-quality teleconferencing.

z Another company, InTouch Health, has used the Ava platform as

the foundation for the robot called RP-VITA, which can navigate
to a new place on its own, having a complete world model of the
KRVSLWDO DQG WKH ORZHUOHYHO UHÀH[LYH EHKDYLRUV WR DYRLG KLWWLQJ
people and things that aren’t on the maps.

118
z 2QH RI WKH WKLQJV WKDW9,7$ FDQ GR LV ¿QG D SDWLHQW 6RPHWLPHV
especially when a hospital is overrun with victims of a large
accident, patients can be temporarily misplaced. To identify a
patient, VITA uses a combination of face recognition and radio
IUHTXHQF\ LGHQWL¿FDWLRQ WDJV ,Q D ÀXLG VLWXDWLRQ \RX FRXOG HYHQ
have a bunch of VITAs keeping track of which patients were where,
what their status was, and what was to happen next and then help
schedule and guide the hospital staff.

z As we move from the mobile robots working hallways, wards,

and rooms and into the surgical theater, we see telepresence in a
completely different kind of robot. For surgeries and patients that
qualify for minimally invasive surgery, Intuitive Surgical created
the da Vinci Surgical System, a teleoperated robot that is a variation
on the classic robot arm, also called a robotic manipulator. The
difference is that da Vinci moves robotic arms inside the human body.

z 'D9LQFL KDV IRXU URERWLF DUPV RXWVLGH WKH SDWLHQW (DFK RI WKHVH
arms has extensions with sensors and actuators that go inside the
body. Surgical robots like da Vinci keep surgeons in the loop.
The surgeon functions as the controller in a remotely controlled
robotic system. Thus, the surgeon senses the world inside the
patient through three-dimensional cameras and issues orders to the
actuators inside the body.

Robotic Strides in Medicine

2000 The da Vinci Surgical System, a robot by Intuitive

Surgical allowing surgeons to be remotely present and
operate inside patients, is approved by the FDA.

  KH'(.$SURVWKHWLFGHYLFHFRQWUROOHGE\HOHFWURP\RJUDP
7
(0*VLJQDOVIURPPXVFOHVLVDSSURYHGIRUDPSXWHHVE\
the FDA.

119
 Capsule Robots
z Capsule robots carry sensors and actuators into the body, and some
are self-propelled. Capsule robots got their start from a technique
called capsule endoscopy, in which a capsule can be swallowed,
and it takes with it into your body a tiny video camera. A camera
ZRUNLQJ LQ WKH GDUN QHHGV OLJKW VR URERWV EULQJ VPDOO /('V WKDW
light up the scene so that pictures can be taken of a person’s insides
to diagnose problems.

z These capsules are passive and not really robotic in the sense that
they don’t have actuators that help them move. But in 1994, a
robotic endoscope was patented and built to be able to actively
wiggle its way through a patient’s gut. The front end of the snake
carries a camera, other sensors, and a small knifelike device to cut
tissue samples. The ability to snake around opens up new areas
of exploration.

z The problem with the robotic snake endoscope is that it is big,

and one danger is that it can perforate the walls of organs. So,
Lecture 13—Hospital Robots and Neuroprosthetics

a number of biomedical engineers have been trying to combine

the small size of a camera capsule with the self-propulsion of the
robotic endoscope.

z A group at the Scuola Superiore Sant’Anna in Italy has come up

with self-propelled robotic capsules. These capsules can have four
propellers on the back end. They are like miniature submarines,
controlled through a wireless communication system.

Assistive Robots
z Medical robots are also working with the human body in a more
ongoing way, collaborating with us to overcome medical challenges
on an everyday basis. There are robots that help in physical and
RFFXSDWLRQDO WKHUDS\ VRFDOOHG WKHUDS\ URERWV .QRZQ PRUH
generally as assistive robots, these are similar to what the press
likes to call an exoskeleton. Assistive robots are strapped onto your
body to help supplement or retrain your existing abilities.

120
z The Walking Assist,
created by Honda, is
a spin-off technology
from their work on
the humanoid robot
ASIMO. The Walking
$VVLVW LV D MRLQWHG
powered robotic limb, in
principle and practice,
working with the human

‹0DWFL6WRFN7KLQNVWRFN
WRKHOSVWUHQJWKHQMRLQWV
and power each step.
This is meant for people
with certain kinds of
PXVFOHRUMRLQWSUREOHPV
The Walking Assist, an offshoot of
who can still walk but Honda’s ASIMOV technology, aids people
can use the help. ZKRKDYHGLI¿FXOWLHVZLWKPRELOLW\

z In addition, Honda has developed an exoskeleton meant for workers.

The robotic legs have their own sensors, controller, actuators, and
battery. The exoskeleton helps support body weight for humans
when they have to stand for long periods or bend, crouch, or kneel
repeatedly. As you crouch, all you have to do is balance; the robotic
exoskeleton does the work of supporting your weight.

z The trick is to have the Honda leg cooperate with the human. This
ZRUNVEHFDXVHVHQVRUVLQWKHMRLQWVRIWKHURERWGHWHFWLQLWLDWLRQRI
movement by the human. The legs then work to make strides even
DQG WR PDNH WKHP ORQJHU 7KH SDWLHQW KDV WKH MRE RI FRRSHUDWLQJ
with the robot that they are wearing.

z An even more impressive area for this kind of cooperation is the

¿HOG RI SURVWKHWLFV 3URVWKHWLF OLPEV FDQ EH YHU\ FRPSOLFDWHG
machines that we attach directly to the nerve endings after
DPSXWDWLRQ7KHLGHDEHKLQGWKH¿HOGRIQHXURSURVWKHWLFVLVWRWDS
directly into the human’s nervous system for control.

121
 z This is done with what amounts to a special antenna, called an
electrode, that is designed to pick up and amplify the electric
signals generated by nerves and muscles. The electrode can be
on the surface of the
skin, under the skin
touching muscle, or in
the body right next to
or even inside of the
nervous system.

z Once the electrode is in

the right place to detect
nerve signals, then you
have to train the human
how to voluntarily
make and control the

‹%UDQGL6LPRQV*HWW\,PDJHV1HZV7KLQNVWRFN
electric signals that
the electrode reads.
The human acts as
Lecture 13—Hospital Robots and Neuroprosthetics

the controller and is

communicating directly
to the robotic arm
via his or her body’s
electric signals instead
of his or her body’s Neuroprosthetics tap directly into the
muscular forces. nervous system for control.

Activity

)LQG DQG ZDWFK WKH ¿OP Fantastic Voyage. As outdated as it is, it

is a very important cultural reference, and aside from miniaturizing
humans, the idea of micro medical robots working inside our bodies
is getting closer to reality.

122
 Important Terms

sensor fusion7KHSURFHVVRIFRPELQLQJLQIRUPDWLRQIURPPXOWLSOHVHQVRUV
to create information that is not available from individual sensors alone.

telepresence 8VLQJ VHQVRUV DQG LQVWUXPHQWV RQ URERWV WR REVHUYH DQG
measure without the need for a human to be physically present.

WLPHRIÀLJKW'XUDWLRQIRUDVLJQDOWRWUDYHOIURPLWVVRXUFHWRDVHQVRU

transducer$Q\GHYLFHWKDWFRQYHUWVHQHUJ\LQRQHIRUPWRHQHUJ\LQDQRWKHU
For example, a touch sensor transduces the kinetic energy of movement into
electric signals; a motor transduces electric signals into kinetic energy.

Suggested Reading

Burdet, Franklin, and Milner, Human Robotics, chaps. 1 and 11.

&LXWL0HQFLDVVLDQG'DULR³&DSVXOH(QGRVFRS\´
Grundfest, Burdick, and Slatkin, 1994, Robotic endoscopy, U.S. Patent
5,337,723.

Other Resources

'D 9LQFL 6XUJHU\ FRPSDQ\ ZHEVLWH KWWSZZZGDYLQFLVXUJHU\FRP

The material at this site is constantly changing, and that’s part of what is
fascinating. Because the site is really set up for patients, you’ll have to dig a
ELWWR¿QGRXWDERXWWKHLUODWHVWWHFKQRORJ\

Questions to Consider

1. The da Vinci robotic surgery system was created to have a doctor in the
loop. What added capabilities would we most need if we wanted to turn
da Vinci into an autonomous surgeon?

2. ,I\RXZDQWHGWRLPSURYHWKHSHUIRUPDQFHRIWKH'(.$QHXURSURVWKHWLF
arm, how might you do so?

123
 Self-Driving Vehicles
Lecture 14

R
obots on the road are one of the most transformative emerging
technologies on the planet. Many of us already have elements of
robotic autonomy built into our cars, and we have for years. Creating
robotic cars involves incredible challenges. Chief among them is the trade-
RIIRIVSHHGYHUVXVVDIHW\7RGULYHVDIHO\\RXVORZGRZQ²EXWLI\RXVORZ
down, you take longer to reach your destination. The goal of the modern
automotive industry is to enhance both speed and safety at the same time.
Nearly every manufacturer has already embedded robotic elements into your
car to assist you.

Cruise Control and Similar Features

z Cruise control is automatic speed control, and it is a robotic system
because it involves sensor-guided movements. You let your car take
RYHU WKH JDV SHGDO DOORZLQJ LW WR DGMXVW WKH WKURWWOH WR NHHS \RXU
speed steady as you travel up and down hills.

z Cruise control is an example of a system that works by negative

feedback control. A sensor—in this case, the speedometer—is
feeding information about the real world back into the system.
The controller compares the actual speed to the desired speed. The
Lecture 14—Self-Driving Vehicles

difference between the actual and desired setting is called an error.

Any system that uses negative feedback control is self-regulating
and embodies the kind of self-control we see in autonomous robots.

z The automobile industry has been adding, one by one, different

autonomous systems to vehicles. In response to cruise control’s
lack of braking, in 1995, Mitsubishi introduced a laser-based
adaptive cruise control. The idea behind adaptive cruise control
LV FDOOHG REMHFW DYRLGDQFH LQ URERWLFV 'RQ¶W KLW WKH FDU LQ IURQW
of you.

124
z Most adaptive cruise control systems use radar as their primary
sensor. Radar is an active sensor system; an emitter sends out radio
waves at a particular frequency, and then a receiver collects the echo,
the return of that signal. Using radar, you can measure the distance
IURP\RXWRWKHQHDUHVWREMHFW<RXFDQDOVRWHOOWKHUHODWLYHVSHHGRI
WKDWREMHFWZKHWKHULW¶VKHDGLQJDZD\IURP\RXRUWRZDUG\RX

z With adaptive cruise control, you set a target speed, like old-
fashioned cruise control, but you won’t hit the vehicle in front of
\RXEHFDXVHRIWKHREMHFWDYRLGDQFHIXQFWLRQ<RXHQGXSIROORZLQJ
at a set distance behind the vehicle in front of you. That distance is
DGMXVWHGEDVHGRQWKHVSHHG

z To avoid collisions, you need to add an actuator to your toolkit.

In addition to throttle control, you also need control of the brakes.
<RX¶OORIWHQ¿QGHPHUJHQF\EUDNLQJVROGDVDIHDWXUHRQFDUVDQGLW
LVDSDUWRIWKHZKROHREMHFWDYRLGDQFHV\VWHP

z (YHQ WKRXJK DGDSWLYH FUXLVH FRQWURO LV PHDQW IRU WKH KLJKVSHHG
movements of freeway driving, some companies, such as Bosch, have
a related system called Stop & Go that is built for slow speeds. In
KHDY\WUDI¿F6WRS *RZLOOEULQJWKHFDUWRDFRPSOHWHVWDQGVWLOOLI
needed and then initiate movement when the car in front moves. This
PDNHVWUDI¿FMDPVIDUOHVVIUXVWUDWLQJLI\RXDUHEHKLQGWKHZKHHO

z One common cause of accidents is when you change lanes and

don’t see another car in your blind spot. Bosch offers Side View
Assist. Using four ultrasonic sensors, two on each side of the car,
WKH6LGH9LHZ$VVLVWVLJQDOVWKHGULYHUZKHQDQREMHFWLVGHWHFWHGLQ
the blind spot.

z In cases where the car is simply drifting out of the lane—such as when
the driver is sleepy—some companies offer Lane Departure Warning
to signal the driver and Lane Keeping Assist to actually take over the
steering. Video is a common sensor for lane keeping, because most
paved roads have clear lines that mark the edges of the road.

125
 z Park Assist can sense other cars and be used when you want the car
to park itself. The same sensors that are used for Side View Assist
can help detect the open spot, and then commands are issued to the
steering and gas to maneuver the car into position.

z Because twice the number of accidents happen at night compared

to the daytime, one of the most exciting driver assist features is
enhanced Night Vision. This system works by combining infrared
light and video analysis. Infrared emitters up front send out signals
that are read by the infrared video camera. A special screen on the
dashboard shows the scene ahead. Using software that recognizes
pedestrians, the Night Vision system can also brightly illuminate
people so that they can be seen and avoided.

z With the exception of old-fashioned cruise control, all of these

behaviors rely on vision—or a sensory capability that functions
OLNH YLVLRQ VXFK DV UDGDU RU LQIUDUHG LPDJLQLQJ (QJLQHHU (UQVW
'LFNPDQQV ZDV DPRQJ WKH ¿UVW WR EXLOG D UREXVW ZRUNLQJ YLVLRQ
system for robotic cars. In 1986, Dickmanns and his team built a fully
autonomous robot car that drove in tests on empty streets in Germany.
With participation from the automobile industry, by 1995 vehicles
equipped with his dynamic vision system were traveling safely for
1,000 miles at speeds up to 100 miles per hour on public highways.

The DARPA Grand Challenge

Lecture 14—Self-Driving Vehicles

z Combining navigation with autonomous driving in robotic cars

was the primary challenge put forward in 2004 by the Defense
$GYDQFHG 5HVHDUFK 3URMHFWV $JHQF\ '$53$ RI WKH 86
Department of Defense. They ran a robot competition called the
DARPA Grand Challenge, which was for fully autonomous robotic
ground vehicles. To succeed, a robot had to travel 150 miles, off
and on the road, between Las Vegas and Los Angeles.

z :KDW PDGH WKH '$53$ *UDQG &KDOOHQJH PRUH GLI¿FXOW WKDQ WKH
work already done by Dickmanns’s team was all of the off-road
GULYLQJ 7KHUH ZHUH IHZHU UHJXODULWLHV QR VLJQV DQG QR FDUV WR

126
 follow. Off-road driving is less structured than on-road driving.
(YHQWKRXJKURERWVVWDUWHGQRWDVLQJOHURERWLFYHKLFOH¿QLVKHG
the 150-mile course.

z Some vehicles were able to navigate to the GPS waypoints, but

WKRVHURERWVWHQGHGWRGRDEDGMRERIGHWHFWLQJREMHFWVDORQJWKH
SDWK2WKHUYHKLFOHVZHUHEHWWHUDWVHQVLQJREMHFWVEXWWKH\ZHUHQ¶W
good at navigating to the waypoints. It was clear that a short-range
system like Dickmanns’s dynamic vision needed to be combined
with a longer-range navigation system.

z DARPA decided to hold a similar challenge the next year, in 2005.

Given all that had been learned in 2004, this turned out to be a
brilliant decision. In 2005, 23 robots started the 132-mile off-road

‹'DYLG0F1HZ*HWW\,PDJHV1HZV7KLQNVWRFN

Although the DARPA Grand Challenge in 2004 yielded no winners, it did spur
LQQRYDWLRQLQWKH¿HOG¿YHYHKLFOHV¿QLVKHGWKHFRXUVHLQWKHQH[W\HDU¶V
competition.

127
 course in the desert of Nevada, and 5 autonomous robotic vehicles
¿QLVKHGWKHFRXUVH%HFDXVHRIWKLVKXJHDQGSRVLWLYHWXUQDURXQG
many consider the DARPA challenge of 2005 to be a watershed
moment in robotics.

z Perhaps the most revealing result in 2005 was that the winning
robot, Stanley D PRGL¿HG 9RONVZDJHQ 7RXDUHJ KDGQ¶W HYHQ
competed the year before. Stanley was the brainchild of the
Stanford Racing Team, which was led by Sebastian Thrun, then the
GLUHFWRURIWKH6WDQIRUG$UWL¿FLDO,QWHOOLJHQFH/DERUDWRU\

z Stanley’s team’s approach in a nutshell was to treat autonomous

navigation as a software problem. The strategy was that all of
the failures of 2004 could be solved by building a better control
architecture. The design of a three-module controller, combined
with estimating uncertainty and machine learning, were the keys
WR WKH JDPH WKDW DOORZHG 6WDQOH\ WR ¿QLVK WKH PLOH RIIURDG
course and win the race.

z Thrun and many members of the Stanford Racing Team took

what they had learned with Stanley to Google to build the Google
driverless car.

Driverless Cars
z Nearly all automobile manufacturers have worked on autonomous,
Lecture 14—Self-Driving Vehicles

driverless cars. They have been helped by components

manufacturers, such as Bosch, that are making robotic systems that
can be deployed on any vehicles. Like Google’s driverless system,
Bosch’s system uses LIDAR, video and radar, to create a dynamic
map of the world.

z Part of the technology transferred from Stanley to the Google

driverless car was its drivability map. Google’s testing engineers
would ride in the passenger seat, essentially looking at the Google
car’s version of Stanley’s drivability map.

128
 z 2QH RI WKH EHQH¿WV RI GULYLQJ RQ URDGV DV RSSRVHG WR RIIURDG
driving, is that roads offer more regularities in the world, and more
structure, from lines to signs and curbs. Google also has mapped
roads, so the robotic car doesn’t have to start from scratch the way
Stanley had to in the desert.

z What makes the world of roads more unstructured and, therefore,

more challenging is that in suburban and urban settings, roads can
EH SDFNHG ZLWK LUUHJXODU WUDI¿F²QRW RQO\ FDUV DQG WUXFNV RI DOO
sizes, stopping and starting at unexpected times and places, but also
dogs, pedestrians, skateboarders, and bicyclists. And their positions
are constantly changing.

z States in the United States began to offer driverless cars the right
to operate in 2012, and testing began in the United Kingdom,
Singapore, and other countries. One of the great things about these
vehicular robots is that this technology can be applied to trains,
trucks, and buses as well. So, the potential is to completely overhaul
our transportation networks that carry people and goods.

%HQH¿WVRI'ULYHUOHVV&DUV
z )RUDOOYHKLFOHVDQLPPHGLDWHEHQH¿WWKDWZHVHHHYHQZLWKGULYHU
assist functions in a semiautonomous vehicle, is safety. While the
number of deaths from automobile accidents continues to trend
downward as we’ve added safety features such as seat belts, air
bags, antilock brakes, and adaptive cruise control, tens of thousands
of people still die every year from vehicle-related accidents, and
PLOOLRQVPRUHDUHLQMXUHG

z The promise of driverless cars, trucks, and buses with even more
DXWRPDWLRQLVWKDWWKH\ZRXOGUHGXFHWKRVHGHDWKVDQGLQMXULHVPXFK
further. But what about cases where some sort of collision remains
unavoidable? For those cases, rules about how to have a collision
can also be programmed into the world model of the vehicle.

129
 A Brief History of Driverless Cars

1986 A fully autonomous robot car begins test-drives on empty

streets in Germany.

1994 VaMP and VITA-2, driverless cars, operate safely in

WUDI¿FIRUPRUHWKDQPLOHVLQ*HUPDQ\

1995 Adaptive cruise control for cars, using lasers, is introduced

by Mitsubishi.

2005 Stanley, a robotic car built by a team from Stanford and

Volkswagen, wins the DARPA Grand Challenge by being
RQHRI¿YHYHKLFOHVWRDXWRQRPRXVO\QDYLJDWHDPLOH
off-road course through the desert.

  RVVDPRGL¿HG&KHY\7DKRHEXLOWE\&DUQHJLH0HOORQ
%
and General Motors, wins the DARPA Urban Challenge,
QDYLJDWLQJDPLOHFRXUVHZKLOHREH\LQJDOOWUDI¿FODZV
of California, including avoiding pedestrians.

  HYDGD DQG )ORULGD EHFRPH WKH ¿UVW 86 VWDWHV WR
1
permit testing of autonomous vehicles on ordinary roads;
Michigan and California follow in 2013.

2015 The DARPA Robotics Challenge focuses on robots for

disaster response.
Lecture 14—Self-Driving Vehicles

z In addition to safety, driverless cars could increase the independence

and mobility of people who, for a variety of reasons, are unable to
drive a car themselves. For example, as we age, what often makes
assisted living imperative is reduced mobility—not being able to
get to the grocery store or a doctor’s appointment.

130
 z One of the unexpected consequences of having driverless cars is
that we will be able to put many more cars on the road. By some
HVWLPDWHV RQO\ ¿YH SHUFHQW RI D FURZGHG URDG LV RFFXSLHG E\
vehicles. Lanes are much wider than the width of the vehicles,
and following distances are kept larger than physically required to
compensate for the slow reaction times of human drivers.

z By using robotic sensors and communication to pack moving cars

WRJHWKHUPRUHHI¿FLHQWO\ZHFRXOGHYHQWXDOO\GRXEOHRUWULSOHWKH
capacity of our existing roads. As the human population expands,
this would save countries billions of dollars on unneeded road
expansion and would keep land available for farming, housing,
wildlife, and other purposes.

z %XW ZLWK URERWLF YHKLFOHV WKH RSSRUWXQLWLHV JR IDU EH\RQG WUDI¿F
OLJKWV6WRSDQGJRWUDI¿FRIDOONLQGVLVWKHZRUVWIRUPLOHDJHIRU
WZR UHDVRQV ,I \RX DUH VWRSSHG WKHQ \RX DUHQ¶W JRLQJ DQ\ZKHUH
and when you accelerate, you use more gas than you do when you
can simply cruise at constant velocity.

z If the transportation network is tracking all vehicles on the roads,

DQG SHGHVWULDQ WUDI¿F WKDW LQIRUPDWLRQ FRXOG EH XVHG WR FDOFXODWH
and send electronic signals to each vehicle about optimum speed
DQG SRVLWLRQ WR PLQLPL]H WKH WLPH WKDW WKH WUDI¿F RQ DYHUDJH LV
VWRSSHG &RQJHVWLRQ OHYHOV WKDW FXUUHQWO\ EULQJ WUDI¿F WR D VWRS
FRXOGEHPDQDJHGPRUHHI¿FLHQWO\VRWKDWWUDI¿FNHHSVPRYLQJ

Important Terms

machine learning&RPSXWHUSURJUDPVZULWWHQWRPDNHDGMXVWPHQWVWRWKHLU
code, with or without direct feedback from a human, in order to improve
performance of the code itself or the robot that the code controls.

Stanley $ IXOO\ DXWRQRPRXV FDU WKDW ZRQ WKH  '$53$ *UDQG
Challenge, created by Stanford Racing Team.

131
 Suggested Reading

'LFNPDQQ$SSHQGURW DQG %UHQN ³+RZ :H *DYH 6LJKW WR WKH 0HUFHGHV
5RERWLF&DU´
Thrun, Burgard, and Fox, Probabilistic Robotics, chap. 1.
7KUXQHWDO³6WDQOH\´

Other Resources

³&DUV 7KDW 7KLQN´ KWWSVSHFWUXPLHHHRUJEORJFDUVWKDWWKLQN 7KLV LV

the best blog on robotic cars, created and curated by IEEE Spectrum, an
engineering magazine.

The Great Robot Race. This NOVA program, available to watch online
KWWSZZZSEVRUJZJEKQRYDGDUSDSURJUDPKWPO JHWV \RX EHKLQG WKH
scenes of the teams and under the hoods of the robots at the Grand Challenge
of 2005.

Questions to Consider

1. If you want to build the best robotic car possible, which kind of
DUFKLWHFWXUH ZRXOG EH EHWWHU IRU WKH FRQWUROOHU EHKDYLRUEDVHG RU
model-based?
Lecture 14—Self-Driving Vehicles

2. In 2014, a robotic car competition sponsored by Hyundai in South

Korea found that the cars did quite well on a dry, sunny day. But when
the road was wet and the weather was partly cloudy, the robotic cars had
far more problems, including several missed turns and other maneuvers
that they had handled easily during the nice weather of the previous
day. What kind of sensors would you suggest be added or improved to
address the problems caused by bad weather conditions?

132
 Flying Robots: From Autopilots to Drones
Lecture 15

A
nytime that we build robots to work outside, the challenges mount.
But the challenges are redoubled for aerial robots. The physical
GHPDQGV RI À\LQJ LPSDFW HYHU\ V\VWHP RQ WKH URERW 6HQVRUV
and actuators have to support full mobility in three dimensions; energy
requirements are high, while weight has to be kept as low as possible, and
the entire control loop has to function much faster. However, the challenges
of aerial mobility are what made aviation an early and consistent adopter of
robotic technologies.

Aerial Robots
z Piloted aircraft often include robotic assist technologies that
operate the aircraft autonomously when the pilot turns them on.
The best example is the autopilot found on commercial aircraft.
:KLOHZHGRQ¶WQRUPDOO\WKLQNRIDSODQHÀRZQE\DXWRSLORWDVDQ
autonomous robot, it is, at least when the pilots aren’t at the controls.
An autopilot is an embedded robotic system, with autonomous self-
control using sensors to guide movements.

z All the self-control technology assisting human pilots is equally

available for unmanned aerial vehicles, too. And unlike aircraft that
carry human pilots, unmanned aerial robots can be almost any size,
from vehicles as big as commercial passenger aircraft to micro-
vehicles the size of insects. And they can even execute maneuvers
too stressful for a human body, opening up new design possibilities.

z $HULDO URERWV JR E\ D ZKROH KRVW RI QDPHV GURQHV XQPDQQHG
aircraft, unmanned aerial vehicles, unpiloted aircraft, unpiloted
aerial vehicle, unpiloted air system, and remotely piloted aircraft.

z 8QPDQQHG GURQHV FRPH LQ WZR W\SHV IXOO\ DXWRQRPRXV DQG

remotely controlled. Autopilots operate unmanned aircraft, such as
many commercially available drones.

133
 z The Walkera TALI H500 is a hexacopter drone. It has six propellers
that provide the thrust for lift, to keep it aloft, and the thrust for
moving forward and maneuvering. You control the TALI using a
radio control console that transmits and receives signals, including
video from the drone.

z The drone controls itself to remain stable. To keep the robot still
takes all sorts of control. When some external force attempts to
tilt the drone, it senses an acceleration and tilt using its onboard
three-axis inertial sensor. Once movement is detected, the onboard
controller immediately reacts by having the three actuators on
the side pushed downward to increase their thrust to counter the
SHUWXUEDWLRQ$IWHUWKLV¿UVWFRUUHFWLYHDFWLRQWKHQWKH7$/,+
has to modulate the thrust on all six thrusters to damp any wobble
that might have been set up.

z Drones can use active ping sensors—located on the bottom of the

craft and pointed toward the ground—to measure altitude. If the
Lecture 15—Flying Robots: From Autopilots to Drones

distance that the ping sensor measures is not one meter, then the
WKUXVWLVDGMXVWHGWRHLWKHUUDLVHRUORZHUWKHFUDIW

z Maintaining that height is an example of negative feedback control.

7KHDFWXDOKHLJKWLVFRQVWDQWO\PHDVXUHGDQGDGMXVWHGWRUHGXFHWKH
difference between that actual height and the desired height. The
feedback is the information that you get from the sensors about the
actual height.

z When the TALI H500 is hovering, it is working on autopilot. It is

doing the work of coordinating all six of its motors for the person
controlling it. When the person steps into the control loop, he or she
can control the altitude, heading, and speed.

z One of the most important autonomous features for any drone to

have is landing. You want to avoid a crash landing. So, drones like
the TALI H500 have built-in landing autopilot functions that help
you preserve your investment.

134
 ‹PDUHNXOLDV]L6WRFN(GLWRULDO7KLQNVWRFN
Aerial robots are useful in many situations, enabling us to go where it might not
be safe or feasible for humans to go.

z Because the TALI H500 has autopilot functions but also allows
a person to control it, it’s a remote control drone with some
autonomous functions. Most drones are this mix of remote control
by a human and autonomous control by the robot. What varies
among drones is the degree to which they offer autonomous
autopilot functions.

z The needs to not lose the drone and to avoid destroying it are the
UHDVRQVWKDWWKHVHDXWRQRPRXVÀLJKWIHDWXUHVFRPHDVWKHVWDQGDUG
defaults. One of the great safety features built into the TALI
+ LV WKDW LI \RX À\ LW RXW RI FRPPXQLFDWLRQV FRQWDFW LW ZLOO
automatically return to its staring position. It uses GPS navigation
to accomplish this.

Designing and Building Aerial Robots

z Designing and building an aerial robot all begins with the basics of
ÀLJKWJHWWLQJDLUERUQHVWD\LQJDLUERUQHDQGODQGLQJJHQWO\

135
 z The bodies of robotic aircraft need to be lightweight to minimize
the energy needed to keep the drone aloft. Bodies need to be
aerodynamically shaped to help the actuators generate lift and
reduce drag. If engine power is lost, bodies also need to be able to
glide, parachute, or otherwise survive an emergency landing.

z The actuators of drones need to be lightweight. If the actuators are

actively generating lift as well as forward propulsion, then they
need to have a high power-to-weight ratio.

z One of the ways to get a motor to be small and powerful is to trade

off torque for power. A drone’s motors are fast with low torque.
The motors of the Parrot AR.Drone 2.0 spin extremely fast but have
virtually no torque at low speeds. They are built to operate at very
high velocities.

z The sensors of drones are of two main types. Proprioceptive

sensors monitor the state of the robot, and they must do so with
Lecture 15—Flying Robots: From Autopilots to Drones

high speeds of reaction and with good accuracy. Navigational

sensors must detect current position on a map, compass heading,
DLUVSHHGDQGDOWLWXGH<RXFDQÀ\PDQ\GURQHVXVLQJDQRQERDUG
GPS navigation system.

z (QHUJ\ VXSSO\ QHHGV WR EH RI KLJK GHQVLW\ EHFDXVH WKH FRVW RI
À\LQJLVVRKLJK7KHSULPDU\FRVWLVWRVWD\DORIWWRJHQHUDWHOLIW
If this can be done for free, as in a balloon or dirigible, the energy
FRVW RI ÀLJKW LV IRU IRUZDUG SURSXOVLRQ RQO\ %XW OLJKWHUWKDQDLU
craft suffer from needing a large size to carry gases. The large size
makes them slow.

z Any rotary winged aircraft, such as a helicopter, quadcopter, or

hexicopter, is in big trouble. The Parrot AR.Drone has a pretty
lightweight battery. But the trade-off is between the weight of the
URERWDQGWKHHQHUJ\FDSDFLW\RIWKHEDWWHU\,WRQO\FDUULHVHQRXJK
HQHUJ\WRSRZHUWKHGURQHIRUDERXWPLQXWHVRIÀLJKWWLPH7KH
heavier Walkera TALI H500 carries a larger battery for 25 minutes
RIÀLJKWWLPH

136
z <RXFDQDOVRJHWUHSODFHPHQWEDWWHULHVWKDWGRXEOHWKHÀLJKWWLPH
But there are always trade-offs. A battery with more capacity
usually means a heavier battery. A heavier drone means that the
motors have to spin faster in order to keep the quadcopter hovering.
Add a big enough battery to spin the motor for a couple of hours
and the drone can’t even take off.

z Control of aerial robots includes behavior-based systems that

RIIHU UDSLG UHÀH[LYH FRUUHFWLRQV RI ÀLJKW V\VWHPV XVLQJ QHJDWLYH
IHHGEDFN FRQWURO WR PDLQWDLQ SURSHU ÀLJKW DWWLWXGH RU WR HQJDJH
maneuvers. Rapid control is particularly important in inherently
unstable aircraft, such as a rotary-wing vehicle or an air vehicle
traveling at high speed.

z Also, certain environments produce wind turbulence that is very

challenging for microdrones in particular. Most microdrones have
operational limits on wind speed. Model-based control algorithms
are needed for navigation to compute heading to the next waypoint,
speed over ground, and altitude above ground.

z 7KHFKDOOHQJHVRIFRQWUROOLQJDURERWLQWKHDLUDUHH[HPSOL¿HGE\D
VLPSOHEHKDYLRUREMHFWDYRLGDQFH2EMHFWDYRLGDQFHLVYHU\HDV\WR
program in land-based robots. For example, Roomba uses bumpers
with infrared proximity detectors, which are active sensors that
emit electromagnetic energy in the infrared range. The sensor also
KDV D GHWHFWRU WKDW LV UHDGLQJ WKH LQWHQVLW\ RI WKH UHÀHFWLRQ RI WKH
LQIUDUHGHQHUJ\RIIDQREMHFW

z 2EMHFW DYRLGDQFH ZLWK PLFURGURQHV OLNH WKH 3DUURW$5'URQH 

won’t work with infrared sensors. You have to be close to have
them work, and then you have to be moving slowly enough to be
able to maneuver out of the way. Drones need long-range sensors.

z Large drones can carry radar, but microdrones cannot. Two

GHYHORSPHQWV PD\ VRRQ KHOS PLFURGURQHV WKH FUHDWLRQ RI VPDOO
DQG OLJKWZHLJKW UDGDU IRU FDUV DQG WKH GHYHORSPHQW RI REMHFW
detection algorithms (from robots like the Google driverless car)

137
 that use video. Because many microdrones have video, this would
EHJUHDW7KHSUREOHPLVWKDWWRGHWHFWREMHFWVTXLFNO\ZLWKYLGHR
you need a fast computer, and fast computers tend to weigh more
than slow ones.

z One of the fundamental challenges of robotics—navigation—is

easier outside and in the air than it is underwater. That’s because
GPS can be used outside, and at heights above the tree line or the
highest buildings, few obstacles exist.

z Aerial robots face different challenges depending on their size.

Aircraft-sized drones are large enough to carry sophisticated sensors
DQGKLJKGHQVLW\HQHUJ\VXSSOLHVVXI¿FLHQWIRUORQJYR\DJHV6RPH
drones can stay aloft for nearly 24 hours. Because large drones
RSHUDWHSULPDULO\LQXQUHVWULFWHGDQGKLJKDOWLWXGHDLUVSDFHVREMHFW
avoidance involves primarily other aircraft, and radar systems on
ERDUGFDQEHXVHGLQFRQMXQFWLRQZLWKODQGEDVHGDLUWUDI¿FFRQWURO
Lecture 15—Flying Robots: From Autopilots to Drones

The Future of Aerial Robots

z 2QH RI WKH ORQJVWDQGLQJ FKDOOHQJHV LQ ÀLJKW V\VWHPV LV FDUU\LQJ
VXI¿FLHQWHQHUJ\WRSRZHU\RXUÀLJKW)L[HGZLQJDLUFUDIWKDYHDQ
advantage over rotary-winged aircraft in that they generate lift by
propelling themselves forward. Once you get airborne, you can take
DGYDQWDJHRI¿[HGZLQJVE\JHWWLQJDVHQHUJ\HI¿FLHQWDV\RXFDQ
EHJOLGLQJ

z Aerial robots can be built as gliders. Gliders move forward by

sinking. The challenge is to catch an updraft and gain height. The
EHVW JOLGHUV LQ WKH DQLPDO NLQJGRP GR MXVW WKLV XVLQJ WKH ULVLQJ
air off of a warming mountain face to give them the height to
glide over.

z Birds can also use a technique called dynamic soaring, which

involves gliding quickly through regions of increasing wind velocity.
Dynamic soaring allows birds to travel for hundreds of miles without
ÀDSSLQJDQGDWVSHHGVDSSURDFKLQJPLOHVSHUKRXU

138
z Radio-controlled gliders have been able to use this technique to go
HYHQIDVWHU²PXFKIDVWHU5DGLRFRQWUROOHGJOLGHUVÀ\LQJLQDORRS
over mountainous terrain have been steadily breaking one another’s
records, with a new record of more than 500 miles per hour reached
using a Kinetic 130DP glider in 2014.

z With such results in mind, a researcher at Woods Hole

Oceanographic Institution, Philip Richardson, has calculated that
glider unmanned aerial vehicles may be able to hit top speeds close
to ten times the speed of the winds they are riding, at least for small
unmanned aerial vehicles in slower winds.

z +\EULGV WKDW XVH ERWK ¿[HG ZLQJV DQG URWDU\ ZLQJV DUH DOVR
possible. This approach has been demonstrated in the Makani
3RZHUNLWHZKLFKZDVWKH¿UVWIXOO\DXWRQRPRXVZLQGKDUYHVWLQJ
robot, demonstrated in 2013 and bought that same year by Google.

z 6WLOO WKH ELJJHVW IURQWLHU LV WKH VPDOO À\LQJ URERWV &RPSDUHG WR
larger drones, microdrones operate near the ground, in complex
urban environments, and even inside buildings. Being so near the
ground, a microdrone encounters a host of obstacles, including
buildings, wires, and trees.

z :DONHUD PDNHV D PLFURGURQH FDOOHG WKH 6FRXW ; WKDW SXVKHV

the limit of real-time navigation and localization. Using your
VPDUWSKRQH \RX SDLU ZLUHOHVVO\ ZLWK WKH 6FRXW ; DQG WKHQ \RX
FDQJHWLWWRIROORZ\RXDQG¿OP\RX

z 0LFURGURQHV OLNH WKH 6FRXW ; UHSUHVHQW WKH QHZ ZDYH RI DHULDO
robots. They are small and affordable, and because of their built-
in autonomous functions, they are very easy to pilot and navigate.
Drones with cameras on board are exciting for a host of potential
uses, including search and rescue, hazardous environment
exploration, high-resolution weather and climate mapping, and
WUDI¿FPRQLWRULQJ

139
 Important Term

dynamic soaring$ W\SH RI JOLGLQJ ÀLJKW LQ ZKLFK WKH DQLPDO RU YHKLFOH
gains velocity and height by harvesting energy from steep wind gradients
located near surfaces.

Suggested Reading

-DUQRW³+LVWRU\´
Sperry, 1931, Wireless-controlled aerial torpedo, U.S. Patent 1,792, 937.
Zaloga, Unmanned Aerial Vehicles.

Other Resources

)HGHUDO$YLDWLRQ$GPLQLVWUDWLRQ8QPDQQHG$LUFUDIW6\VWHPVKWWSVZZZ
IDDJRYXDV $V UXOHV DQG UHJXODWLRQV IRU XQPDQQHG DLUFUDIW RI DOO W\SHV
evolve rapidly, this is the epicenter of change.
Lecture 15—Flying Robots: From Autopilots to Drones

Question to Consider

1. When we remotely control a robot, that robot is said to have an open

loop; when it is autonomous, it is said to have a closed loop.
Actuators Actuators

Body Body Controller

Sensors

Open Loop Closed Loop

&RQVLGHUWKHDXWRPDWLFSLORWLQYHQWHGE\(OPHU6SHUU\,I\RXDUHÀ\LQJ
a plane and it has an autopilot, is that plane a closed-loop robot?

140
 Underwater Robots That Hover and Glide
Lecture 16

N
owhere are the challenges of working with robots in the wild more
GLI¿FXOWWKDQRQDQGLQWKHZDWHU7RZRUNLQDULYHUODNHRURFHDQ
every aspect of the robot has to be redesigned. The challenges are
especially severe if you aim to have your mobile robot go underwater, where
we call it an autonomous underwater vehicle (AUV). The body, energy
supply, actuators, sensors, controller, and communications systems of an
AUV all operate differently than they do on land or in the air.

Water and Electronics

z Water and electronics do not mix. That’s because water conducts
current, like a wire, creating pathways between components that
were not meant to have direct connections. The result is what we
call short-circuiting.

z $OZD\V GRXEOHFKHFN \RXU URERW IRU OHDNV DW WKH ODXQFK VLWH MXVW
SULRU WR SXWWLQJ LW LQ WKH ZDWHU (YHQ ZKHQ ZH EXLOG URERWV IRU
swimming on the surface of the water, we need to worry about leaks.

Sensors in Water
z The sensors we use in the air and on land often work poorly or not
at all in water. While radio waves work well in the air, they do not
work well underwater. The radio frequencies we use in air have
wavelengths from about 10 centimeters to 10 meters—a size that
is not absorbed by the atmosphere. That’s why we use them. But
water quickly attenuates those same radio waves underwater.

z Radio waves are part of the electromagnetic spectrum, which also

includes microwaves, infrared waves, visible light waves, each of
which are shorter than radio waves. Visible light also has problems
penetrating water. Oceanographers refer to the region where visible
OLJKW SHQHWUDWHV VXI¿FLHQW IRU SKRWRV\QWKHVLV DV WKH SKRWLF ]RQH
ZLWKDFXWRIISRLQWGH¿QHGDVZKHUHOHVVWKDQRQHSHUFHQWRIYLVLEOH

141
 light from the surface can penetrate. The distance where more than
99 percent does not get through occurs on average at distances of
only 200 meters (660 feet).

z In order to propagate long distances in the water, electromagnetic

energy needs a wavelength of more than 100 kilometers. These long
waves oscillate at extremely low frequencies, and they have been
used for communications with submarines. But the low frequencies
mean that messages have to be short. And they can only be one-
way, from station to submarine, because a huge antenna is needed
to create the giant waves, and even a submarine is too small to carry
one that is large enough.

z A different kind of wave that does travel well underwater is a

pressure, or acoustic, wave. Submarines can communicate while
underwater using acoustic transmission systems with a speaker and
a hydrophone. But while these pinger systems work well over short
distances, such as a few kilometers, they suffer from noise created
Lecture 16—Underwater Robots That Hover and Glide

E\UHÀHFWLRQVDQGDWWHQXDWLRQFUHDWHGE\VFDWWHURIIRISDUWLFOHVLQWKH
water. Also, the amount of information that you can transmit with a
pinger is very small, because all that you can vary to represent that
information is the pattern of the pulse of sound—the rhythm.

z An underwater acoustic modem system is the most-often-used

form of communication in underwater marine robotics. The most
sophisticated systems offer short-range communication over the
range of a kilometer in shallow-water situations. The rate of data
communication is up to 30 kilobits per second, which is slow
compared to wired or wireless transmission rates on land.

z Although acoustic modems are the standard way to communicate

wirelessly underwater, using a wire, a communications tether, also
is still very common. As long as you don’t have to have your robot
swim very far away, a communications tether can give you great
control over your robot. Underwater robots with a tether are much
PRUHOLNHO\WRKDYHUHPRWHFRQWURODQGRUDQHQHUJ\VXSSO\FRPLQJ
through the tether as well.

142
z Local sensors, operating over short distances, offer a creative way
around the limitations of radio waves, acoustic signals, or tethers.
$VROXWLRQKDVEHHQSURWRW\SHGE\(XURSHDQURERWLFLVWVLQDSURMHFW
called Collective Cognitive Robotics (CoCoRo). The idea is to use
EOXH OLJKW WR ÀDVK FRPPXQLFDWLRQV XQGHUZDWHU 7KH GLVWDQFH RI
FRPPXQLFDWLRQ LV DFKLHYHG E\ D EXFNHW EULJDGH DSSURDFK 8VH D
swarm of robots arranged in a linear chain. A signal at one end of
the chain propagates along the chain.

z (YHQ WKRXJK WKH &R&R5R UHVHDUFKHUV XVHG EOXH OLJKW IRU

communication, the signal from light degrades very quickly,
MXVW OLNH VRXQG ZDYHV RU UDGLR ZDYHV 6R LQ IDFW WKHLU UHDO
innovation is to show that a swarm of robots arranged in a line
can extend communications to new lengths. This swarm-based
collective communication system is still in the earliest stages
of development.

z A common replacement for vision underwater is sonar. This is

another use of sound underwater. With active sonar systems, the
sensor sends out a sound pulse or train of pulses and then listens
for the return, the echo. Depending on how long it takes the pulse to
return and how strong the signal is, the robot can use sonar to detect
REMHFWVXQGHUZDWHUDQGWRPDSWHUUDLQ

z )RU H[DPSOH WKH 3$,9 URERW MRLQWO\ GHYHORSHG E\ 6XEVHD  DQG
SeeByte, has a three-dimensional sonar that looks forward and a
'SUR¿OLQJVRQDUWKDWORRNVGRZQ3$,9KDVDQDYLJDWLRQV\VWHP
that includes a compass and inertial guidance sensors. PAIV also
has an acoustic communication system and a camera and lights to
gather pictures.

z The key to autonomous movement is to link information from

sensors to the control of actuators. Because sensor-guided
movement is at the heart of autonomous behavior, the choice of
sensors is a very important part of design.

143
 z Sensors underwater face a number of challenges because of the
physical properties of water and operating in the wild. Video
systems, for example, only work if the water is clear and light is
available. Go into murky waters of a delta or into the deep and you
are in trouble if you are relying on vision.

z The mission concept for PAIV is to operate as an autonomous

inspection system. For example, we humans have sunk all kinds of
electric and communications cables in the oceans and lakes. These
cables need constant maintenance that can be directly assessed by
close-up inspection.

z PAIV locates the cable riser using sonar. Then, it moves along the
FDEOHGHOLYHULQJYLGHRLPDJHVRIWKHFDEOHVDQGÀRDWVIRUDWRSVLGH
human to review. Sonar is clearly the primary sensor system for
sensor-guided movements.

z PAIV can navigate the bottom to inspect pipelines. One of the

Lecture 16—Underwater Robots That Hover and Glide

virtues of autonomous robots like PAIV is that because they lack

tethers, they can move in and around complex three-dimensional
structures without fear of tangling.

Nautical Robot Milestones

  KH WRUSHGR WKH ¿UVW DXWRQRPRXV XQGHUZDWHU YHKLFOH

7
(AUV), is invented by Robert Whitehead.

  8 DOVR NQRZQ DV 6FDUOHW .QLJKW EHFRPHV WKH ¿UVW
5
$89 WR FURVV WKH$WODQWLF 2FHDQ XQGHUZDWHU D MRXUQH\
that took 221 days.

  KH :DYH *OLGHU URERW %HQMDPLQ IURP /LTXLG 5RERWLFV

7
sets a world record of 7,939 nautical miles, the longest
MRXUQH\E\DXWRQRPRXVVXUIDFHYHVVHO

144
Bodies of Aquatic Robots
z 7KH ERGLHV RI DTXDWLF URERWV DUH RIWHQ YHU\ GLVWLQFWLYH 7KH ¿UVW
autonomous underwater robot was a torpedo. But while torpedoes
DUHZHDSRQVWKHSK\VLFVWKDWPDNHVWKHPVRHI¿FLHQWKDVLQVSLUHG
a whole body type among modern autonomous underwater vehicles
(AUVs), many used for other purposes.

z %OXH¿Q5RERWLFVKDVDWRUSHGRVKDSHG$89FDOOHGWKH%OXH¿Q
The tail doesn’t have the protruding control surfaces of a standard
:KLWHKHDG WRUSHGR QR UXGGHU DQG QR HOHYDWRU 7KDW VOHHN GHVLJQ
reduces the chances of the tail getting hung up on cables, seaweed,
or the like.

z The torpedo shape tells us a lot about the physical challenges of

moving in water. A torpedo is an example of a streamlined body.
A streamline is a visualization technique to see how smoothly the
water moves around a body as the body moves through the water.

‹,85,,.2'(17&(9L6WRFN7KLQNVWRFN

7KHSK\VLFVWKDWPDNHWRUSHGRHVHI¿FLHQWXQGHUZDWHUPDNHWKHERG\VKDSH
ideal for other autonomous underwater vehicles.

145
 Shapes like torpedoes that are tapered in the front and the back
make for very smooth traveling through the water. They are said to
EHVWUHDPOLQHGZKLFKPHDQVPRUHHQHUJ\HI¿FLHQWWRPRYH

z In contrast, an unstreamlined shape would have blunt ends and

be boxy in shape. This means that you can tell right away when
you look at an underwater robot if it is built to cruise or to hover.
Cruisers are streamlined, while hoverers are boxlike.

z )RULWVWRUSHGRVKDSHGURERWV%OXH¿QKDVGHVLJQHGWKHWDLOFRQHDV
D³ducted thruster´WKDWLVDUWLFXODWHG7KHFRQQHFWLRQRIWKHWDLO
FRQHWRWKHKXOOLVOLNHDEDOODQGVRFNHWMRLQW$VZLYHOLQJMRLQWOLNH
WKLVJLYHVWZRGHJUHHVRIIUHHGRPIRUWKHDGMXVWPHQWVRIWKHWKUXVW
The tail cone can pitch, yaw, or move in some combination of the
two. The AUV steers by rotating the tail cone to direct its thrust in
different directions.

Energy Supply and Actuators of Aquatic Robots

Lecture 16—Underwater Robots That Hover and Glide

z The energy supply and actuators of an AUV can be very different as

ZHOO%OXH¿Q¶VWRUSHGR$89ZRUNVXVLQJEDWWHULHVDQGWKH\KDYH
a mission life of about 24 hours. That’s great for mapping near the
shore, but there are times when you want the AUV to have missions
that last longer. For example, extensive monitoring of sea surface
WHPSHUDWXUHVZKLFKLVYLWDOIRUSUHGLFWLQJ(O1LxRHYHQWVUHTXLUHV
time on station that we would measure in weeks or months.

z One way to extend mission time is to reduce the amount of energy

WKDW\RXVSHQGRQSURSXOVLRQ<RXPLJKWHYHQGHFLGHWRMXVWÀRDW
along with currents or tides, but then you aren’t an AUV because
\RXDUHQ¶WVHOISURSHOOHG<RX¶GEHDVHQVRUEXR\WKDWÀRDWHGDORQJ
and collected data.

z A very clever way to be self-propelled but to do so with minimal

cost is to glide. There is a whole class of AUVs called gliders, and
WKH\ ORRN OLNH WRUSHGRHV ZLWK ZLQJV OLNH %OXH¿Q¶V 6SUD\ *OLGHU
*OLGHUV DUH VR HI¿FLHQW²WDNH VXFK OLWWOH HQHUJ\ WR RSHUDWH²WKDW
they can run on a single mission for more than six months.

146
 z Gliders work by changing their density relative to the water. An
REMHFW GHQVHU WKDQ ZDWHU ZKDW ZH FDOO QHJDWLYHO\ EXR\DQW ZLOO
VLQN DQ REMHFW OHVV GHQVH WKDQ ZDWHU ZKDW ZH FDOO SRVLWLYHO\
EXR\DQWZLOOÀRDW$JOLGHUDWWKHVXUIDFHRIWKHZDWHUPDNHVLWVHOI
negatively buoyant so that it sinks and uses that sinking motion to
allow it to move forward, thanks to its wings.

z Most of the energy of propulsion in a glider comes from the

gravitational effect called buoyancy, so we talk about these AUVs
having a buoyancy engine. You can think of a buoyancy engine as
VLQNLQJDVKLSE\ÀRRGLQJRQHRILWVFRPSDUWPHQWVDQGWKHQÀRDWLQJ
a ship by bailing out that same compartment.

z Gliders are the long-distance, low-power champions on the opposite

end of the AUV spectrum from the hovering AUVs like the PAIV
robot. Hovering AUVs are vehicles designed to not go very far and
WRUXQIRUMXVWDIHZKRXUV

z Hovering AUVs perform close-up inspection of hulls of ships,

submerged piers, or rocky bottoms. To hover and hold station, they
PD[LPL]H VRPHWKLQJ WKDW JOLGHUV DQG WRUSHGRHV DUH YHU\ EDG DW
maneuverability.

z The champion of long-distance and long-duration robots is Wave

Glider, which was built by Liquid Robotics and looks like a
surfboard with a few antennae. Wave Glider is an example of how
to solve two of the big problems that the aquatic environment
presents to robotics. First, its wave-based power supply is endless,
and because of this, so is its effective time on station conducting its
PLVVLRQ(QHUJ\LVVLPSO\QRWDSUREOHP6HFRQGEHFDXVHSDUWRI
Wave Glider is always at the surface, it can always communicate
using radio waves and navigate using GPS.

Important Term

ducted thruster $ SURSHOOHUEDVHG DFWXDWRU XVHG LQ WRUSHGRVKDSHG

underwater robots to both propel and steer the vehicle.

147
 Suggested Reading

Bohm and Jensen, Build Your Own Underwater Robot, and Other Wet
Projects.
Bureau of Ordnance, Department of Navy, The Whitehead Torpedo.
/RQJ³%LRPLPHWLFV´

Other Resources

*HWVWDUWHGRQRSHQVRXUFHXQGHUZDWHUURERWVKWWSRSHQURYFRP

Questions to Consider

1. One of the challenges of working with robots underwater is how to

control motion in three dimensions. How can we simplify the control of
our robot underwater?
Lecture 16—Underwater Robots That Hover and Glide

yaw

roll

pitch

148
 a. Use a three-axis accelerometer, compass, and inclinometer to
determine and control the rotation of the robot in three rotational
dimensions.

b. Use wings and other stabilizers to prevent or reduce rotation in

the roll dimension while adding control surfaces to allow for pitch
and yaw.

c. Let the robot freely roll, pitch, and yaw as forces dictate.

2. Water is an extremely challenging environment for robots. What makes

OLIHGLI¿FXOWIRUDPDULQHURERWDQGLWVDWWHQGLQJURERWLFLVWV"

a. For humans working either on the shore or in a surface vessel, any

URERWEHORZWKHVXUIDFHLVGLI¿FXOWWRNHHSWUDFNRI

b. Any leak in the robot’s hull will short-circuit the onboard electronics.

c. The enormous pressure of water at depth can easily crush most

robots.

149
 Space Robots in Orbit and on Other Worlds
Lecture 17

S
SDFH LV WKH ¿QDO IURQWLHU DQG URERWV DUH WKH YDQJXDUG RI RXU KXPDQ
explorations of the universe. With space missions, we push the
extreme of working remotely with and through our robots. Robots in
space include rockets, spacecraft, shuttles, satellites, probes, landers, and
URYHUV2QERDUGVKXWWOHVRUVSDFHVWDWLRQVZHDOVR¿QGPDQLSXODWRUURERWV
and humanoids. Much of the work in space robotics has focused on low-
(DUWKRUELWDQGWKHVXUIDFHVRIRWKHUSODQHWV

Orbital Robots
z 7KH PRVW H[FLWLQJ SURMHFW LQ RUELWDO URERWLFV LV WKH ,QWHUQDWLRQDO
Space Station, where humanoid robots are being developed to help
humans work inside the space station. Robonaut 2 (R2) arrived on
Lecture 17—Space Robots in Orbit and on Other Worlds

the International Space Station in 2011. Like Baxter, R2 is built to

interact with humans, safely. It consists of a head and a torso, with
two arms that function independently.

z R2 can be controlled by a human operator in the space station

wearing a glove and headset or remotely by ground control. R2
can also be programmed to do some tasks autonomously. Many
of its low-level operations, such as grasping control, are done
DXWRQRPRXVO\ HYHQ LI IRU H[DPSOH WKH REMHFW WR EH JUDVSHG LV
determined by its human operator.

z How can R2 help humans in the challenging work conditions of

microgravity? Inside the space station, R2 could set up an area for
the manipulations that only humans can do. During an operation,
R2 could assist by providing tools or holding something in place.
And then it could clean up afterward.

150
z NASA plans for R2 to work on the outside of the station, too,
ZKHUHFRQGLWLRQVDUHH[WUHPHO\KDUVK$WWKH(DUWK¶VGLVWDQFHIURP
the Sun, such as the International Space Station, temperatures in
sunlight can rise to more than 120 degrees Celsius or 248 degrees
Fahrenheit, well past boiling (for water). But on the side of the craft
not facing the Sun, heat radiates away so quickly that temperatures
RQWKHGDUNVLGHFDQVLQNWRíGHJUHHV&HOVLXVRUíGHJUHHV
Fahrenheit, well below freezing.

z Because it takes about 90 minutes for the space station to orbit the
(DUWKLI5ZHUHZRUNLQJRXWVLGHIRUDIHZKRXUVLWFRXOGHDVLO\
encounter those extreme temperatures. Without thermal control, R2
would likely suffer a number of problems, including camera and
battery malfunctions.

z 5ZLOOQHHGWREHRXW¿WWHGZLWKDQDUUD\RISDVVLYHWKHUPDOFRQWURO
(such as insulation and external coatings) and active thermal control
systems (such as electric-powered heaters and coolers).

z Once thermally hardened for work on space walks, R2 will be able

to replace humans on these dangerous external tasks. One of the
reasons to design R2 as a humanoid is so that it can use the same
tools, handholds, hatchways, and equipment as humans.

z Part of R2’s great ability to reach and grasp and interact safely
comes from the touch sensors that it has on its hands. These sensors
provide R2 with feedback so that it can hold onto very fragile
REMHFWVRUDSSO\VWURQJHUIRUFHZKHQQHHGHG

z R2 is also strong. It can lift a 20-pound weight, and it can do things

that humans can’t, such as hold that 20-pound weight with arm
outstretched for long periods of time.

151
 z R2 has been designed to be compliant. It knows when it has
HQFRXQWHUHGDQXQH[SHFWHGREMHFWDQGZLOOVKXWGRZQLPPHGLDWHO\
This makes R2 safe to work with. R2 also has a vision system. It
can pick up an envelope and examine it. This function involves
UXGLPHQWDU\REMHFWLGHQWL¿FDWLRQ

z (YHQWKRXJK5RQO\ZRUNVLQVLGHWKHVSDFHVWDWLRQDWWKHPRPHQW
other robots are at work outside. They are all versions of armlike
robots that are similar to the pick-and-place robots that work in
IDFWRULHV7KH ¿UVW RI WKHVH RXWVLGH RXWHU VSDFH URERWLF DUPV ZDV
the so-called Canadarm. Built for the space shuttles by the Canadian
Space Agency, the remote manipulator system was teleoperated by
crew members through a computer interface.

z The remote manipulator system was used to unpack the cargo bay
RI WKH VKXWWOH RU WR JUDE DQG PDQLSXODWH REMHFWV DOUHDG\ LQ RUELW
7KH MRLQWV ZHUH NH\ WR WKH IXQFWLRQLQJ RI WKH UHPRWH PDQLSXODWRU
V\VWHP,WKDGDZULVWHOERZDQGVKRXOGHUHDFKMRLQWKDYLQJWKUHH
Lecture 17—Space Robots in Orbit and on Other Worlds

one, and two degrees of freedom, respectively.

z The remote manipulator system turned out to be a workhorse for

WKH ZKROH VKXWWOH SURJUDP 1$6$ TXLFNO\ ¿JXUHG RXW WKDW WKH\
could use the remote manipulator system to move astronauts
around on space walks. The remote manipulator system proved to
be incredibly useful, and it spawned the creation of the Canadarm2
for use on the International Space Station.

z +DYLQJEXLOWWKH,QWHUQDWLRQDO6SDFH6WDWLRQRQHRIWKHPDLQMREV
of Canadarm2 is to grab, dock, and unload unmanned spacecraft
that deliver vital supplies. Canadarm2 has the ability to perform
¿QHUVFDOHPDQLSXODWLRQVE\DGGLQJRQWRLWVHQGDWZRDUPHGURERW
called Dextre, which is loaded with many degrees of freedom,
allowing it to swivel, pitch, yaw, and grasp.

152
 Extraterrestrial Robots

1970 Lunokhod 1, a Soviet remote-controlled lunar rover,

EHFRPHVWKH¿UVWXQPDQQHGYHKLFOHWRH[SORUHWKHVXUIDFH
of an extraterrestrial planetary body.

  RMRXUQHU WKH VHPLDXWRQRPRXV URYHU RI 1$6$¶V

6
3DWK¿QGHU PLVVLRQ EHFRPHV WKH ¿UVW URERW WR ODQG DQG
work on Mars.

2004 Spirit and Opportunity, semiautonomous NASA rovers

and geological explorers, land on Mars.

2012 The Curiosity rover, a robotic geological explorer, lands

on Mars.

z Canadarm2, which is a robotic manipulator, can use, as its end

effector, another robotic manipulator, Dextre. Hanging onto and
moving around another larger robot is made possible because of the
near weightlessness of being in orbit.

z Dextre is used, along with Canadarm2, to change batteries and

UHSODFH FDPHUDV MREV IRUPHUO\ GRQH E\ DVWURQDXWV LQ ORQJ WLULQJ
DQGGDQJHURXVVSDFHZDONV'H[WUHLVEXLOWVSHFL¿FDOO\WRZRUNRQ
the International Space Station. Dextre has a toolkit from which it
can select different end effectors. Different graspers are built for
GLIIHUHQWMREV

z Robonaut2, Canadarm, Canadarm2, and Dextre are all designed

to help humans grasp and manipulate the world. Whether located
on a shuttle, crawling around the outside of the space station, or
hanging out inside, they all operate as nearby extensions of human
arms and hands.

153
 Interplanetary Robots
z ,QDGGLWLRQWRRUELWDOURERWLFVWKHRWKHUPDMRU¿HOGRIVSDFHURERWLFV
is that of interplanetary robotics, whose stars are the planetary
URYHUV 8QOLNH URERWV LQ RUELW DURXQG WKH (DUWK URERWV RQ RWKHU
planets are separated from us by both long distances and long lag
times in communications.

z %XW WKH URERWLFV FKDOOHQJH LV QRW MXVW DERXW KRZ WR FRPPXQLFDWH
with your robots but how to control them over long distances,
for purposes of virtual presence and virtual agency. Presence and
agency are essential for scientists and explorers and for being
DEOH WR DGMXVW WKH RULJLQDO PLVVLRQ DV QHZ LQIRUPDWLRQ UHYHDOV
faulty assumptions, as the robots have problems, or as increased
understanding of the planet provides opportunity for new tasks and
goals to emerge.

z Twin robot rovers, Spirit and Opportunity, were built to study the
Lecture 17—Space Robots in Orbit and on Other Worlds

complex geology of Mars. Their primary goal was to understand

KRZ WKH DFWLYLW\ RI ZDWHU RQ 0DUV LQÀXHQFHG WKH SODQHW¶V
HQYLURQPHQWRYHUWLPH7KXVWKH0DUV([SORUDWLRQ5RYHUVDUHSDUW
RI1$6$¶VORQJWHUPPLVVLRQWRDFKLHYHIRXUVFLHQWL¿FJRDOVWKDW
H[WHQGDFURVVPXOWLSOHPLVVLRQVDQGLQWRWKHIXWXUH
ż determine if life ever arose on Mars,

ż characterize the climate on Mars,

ż characterize the geology of Mars, and

ż prepare for humans to travel to and explore Mars.

z The incredible success of this mission has set a new standard in

space robotics for planetary science and exploration.

z %HFDXVH RI WKHLU VFLHQWL¿F WDVNV 6SLULW DQG 2SSRUWXQLW\ DUH RIWHQ
FKDUDFWHUL]HG DV URERW JHRORJLVWV )RXU WHDPV RQ (DUWK JXLGH WKH
JHRORJLFDO ZRUN ZKLFK IDOOV LQWR IRXU FDWHJRULHV PLQHUDORJ\ DQG
geochemistry, soils and rocks, geology, and atmosphere.

154
 z (DUO\LQWKHPLVVLRQWKHVHIRXUVFLHQWL¿FWHDPVPHWGDLO\WRGLVFXVV
the latest information from the robots and then created a new plan
for the following day. Then, those plans were uploaded, via radio,
to each rover. The rovers would then autonomously enact the plans
and report back.

z What’s neat about this process is that scientists at mission control

were reprogramming the autonomous robotic rovers every day.
Contrast this with real-time remote control, where, instead of a plan
enacted in computer code, scientists have the plan in their minds,
and they directly control direction, speed, and any instruments by
inputting commands manually.

z Because of the communication delays that result from real-time

remote control, NASA decided early on in its planetary exploration
programs to avoid any attempts at real-time remote control. For
control of these robots so distant in time and space, that leaves
autonomy, with the rover operating on its own according to
programming on board its computers.

z Autonomy for planetary robotics comes in two types. One type

LV ¿[HG DXWRQRP\ ZKHUH WKH SURJUDPPLQJ WKDW FRQWUROV WKH
LQWHUDFWLRQV RI WKH VHQVRUV DQG WKH DFWXDWRUV LV ¿[HG 7KH RWKHU
type NASA calls batch processing, but it is better described as
reprogrammable autonomy.
‹6WRFNWUHN,PDJHV7KLQNVWRFN

5RERWJHRORJLVWV6SLULWDQG2SSRUWXQLW\DUHQRWUHPRWHO\FRQWUROOHGLQUHDOWLPH
the robots autonomously enact the plans uploaded by NASA.

155
 z Reprogrammable autonomy is the type employed in Spirit,
2SSRUWXQLW\ DQG &XULRVLW\ URYHUV RQ 0DUV (YHU\ GD\ D QHZ
program, a new plan, is uploaded to each rover.

z The trade-off with reprogrammable autonomy is one of adaptability

versus time. If you can reprogram your robot every day, then
every day it can do something new; it can adapt to the information
it discovers and any unforeseen changes in how it is operating.
However, it takes time to reprogram, to evaluate what to do next,
and to communicate those plans. So, robots with reprogrammable
DXWRQRP\FRPSDUHGWRWKRVHZLWK¿[HGDXWRQRP\PD\QHHGPRUH
time to accomplish a given task or mission.

z Interplanetary communications, which happen in both directions, to

and from Mars, to and from the robotic rovers, are delayed, of short
duration, and of small size.

z
Lecture 17—Space Robots in Orbit and on Other Worlds

The robot’s energy supply powers the radio transmitter and receiver.
All of the sensors and instruments on the rovers also require energy,
so their use has to be carefully planned and scheduled. With their
cameras, the rovers have created an incredible catalogue of images of
0DUV7KUHHVFLHQWL¿FVHQVRUV\VWHPVDOORIZKLFKDUHspectrometer
instruments, look for water, iron, and chemical composition.

z 0DNLQJ LW SRVVLEOH IRU WKHVH LQVWUXPHQWV WR GR WKHLU MREV LV WKH
arm, or what NASA calls the instrument deployment device. The
URYHU¶VDUPKDV¿YHGHJUHHVRIIUHHGRP7KHMRLQWVRIWKHVKRXOGHU
HOERZDQGZULVWDUHDOOUHYROXWHMRLQWVWKDWURWDWH7KHVKRXOGHUKDV
WZR MRLQWV RULHQWHG SHUSHQGLFXODU WR HDFK RWKHU 7KH HOERZ KDV D
VLQJOHMRLQW7KHZULVWKDVWZRMRLQWV$VWKHZULVWURWDWHVLWEULQJV
different instruments on the hand to bear on the soil.

z 7KH 0DUV ([SORUDWLRQ 5RYHUV KDYH PHW DQG H[FHHGHG 1$6$¶V

expectations. Originally designed for a mission of 90 days, Spirit
sent its last communication in 2010, 6 years after it began operations
and traveled nearly 5 miles. Opportunity operated for more than 10
years and logged more than 25 miles.

156
 z 6FLHQWL¿FDFKLHYHPHQWVRIWKHPLVVLRQKDYHEHHQPDQ\6SLULWWRRN
movies of swirling dust devils, found circumstantial evidence of
water by detecting sulfate minerals, and found silica deposits that
indicate a much wetter past. Opportunity found evidence for water
over a wider range of terrain, indicating a past with large bodies of
standing water.

Activity

&DQDGDUPVLPXODWRU

ZZZDVFFVDJFFDHQJPXOWLPHGLDJDPHVFDQDGDUP

0DGH DYDLODEOH E\ WKH &DQDGLDQ 6SDFH$JHQF\ WKLV LV WKH RI¿FLDO
simulator game of the Canadarm2. One of the challenges is to use the
three different cameras to create a three-dimensional understanding
RIWKHVSDFHVWDWLRQLQ\RXUKHDG<RXUPLVVLRQLVWZRIROGWRLQVWDOO
DPRGXOHDQGWKHQWRWUDQVSRUWDQDVWURQDXW*RRGOXFN

Important Term

spectrometer$Q\RIDODUJHFODVVRILQVWUXPHQWVGHVLJQHGWRPHDVXUHKRZ
the intensity of a physical property varies across a range of frequencies,
energies, or masses.

Suggested Reading

Brooks, Flesh and Machines, chap. 3.

Clancey, Working on Mars, chaps. 4 and 6.

157
 Other Resources

7KH RI¿FLDO ZHEVLWH IRU WKH URERWV 6SLULW DQG 2SSRUWXQLW\ KWWSPDUVMSO
QDVDJRYPHUKRPH

7KHRI¿FLDOZHEVLWHIRUWKHURERW&XULRVLW\KWWSPDUVMSOQDVDJRYPVO

Question to Consider

1. When we decide that a planetary rover, such as the Soviet Union’s

Lunokhod, is going to be remotely controlled, we are opting for the
same kind of human-in-the-loop architecture that we still see in our
partially robotic planes and automobiles. What are the real engineering
advantages to these open-loop systems?

a. The controller, the human, is more complex and therefore more

capable than any set of algorithms that we can code on a computer.
Lecture 17—Space Robots in Orbit and on Other Worlds

b. If the vehicle contains sensors, then humans can be remotely

SUHVHQWZKDWVRPHSHRSOHFDOO³WHOHSUHVHQW´LQWKHZRUOGLQZKLFK
the robot sits.

c. A human in the control loop allows for careful checks on the status
of a valuable robot.

158
 Why Military Robots Are Different
Lecture 18

M
ilitary robots lead the pack in terms of innovations in robotics,
because they have to be customized to operate in the air, on and
under the sea, and even in space. Many military robots have been
self-guided weapons; they were designed to destroy and kill. Therefore,
there are tensions between semiautonomous military robots and the need
for humans to oversee safety and control. There are trade-offs to consider
EHWZHHQ VSHHG RI DFWLRQ DQG FRPSOH[LW\ RI SODQ DQG WKH LGHQWL¿FDWLRQ RI
friend or foe becomes far more important in military situations than in any
other area of robotics.

Safety and Control

z One way to maintain control over a robot—to make the robot obey
you—is to maintain communications with it and give it limited
autonomy to act on its own. While partial autonomy keeps control
LQWKHKDQGVRIWKHKXPDQVLWSUHVHQWVDVHFXULW\SUREOHP5DGLR
communications can be hacked.

z One robotics solution to the changing situations, such as being

KDFNHG LV FDOOHG DGMXVWDEOH DXWRQRP\ )RU H[DPSOH LI D UDGLR
OLQNLVFXWPDQ\GURQHVDXWRPDWLFDOO\À\WKHPVHOYHVKRPH7KH\
PRQLWRU FRPPXQLFDWLRQV DQG WKHQ DGMXVW WR WKH FKDQJHG VLWXDWLRQ
and become fully autonomous.

z A second concern about maintaining safety and control resides

in how decisions are made. In robotics, there are two main kinds
RI GHFLVLRQPDNLQJ V\VWHPV EHKDYLRUEDVHG FRQWURO DQG PRGHO
based control.

z %HKDYLRUEDVHGFRQWUROOHUVDUHIDVWUHÀH[LYHV\VWHPVVXFKDVWKH
escape response on a Roomba vacuum cleaner. A sensor triggers a
UHVSRQVH:KHQLWGHWHFWVDQREMHFWRUDFKDQJHLQHQYLURQPHQWDO
conditions, that information triggers the programmed software

159
 controller to send instructions to the motors of the wheels to initiate
a series of movements called an escape. If multiple sensors are
WULJJHUHGWKHFRQWUROOHUDOORZVRQHUHÀH[WRRYHUUXOHWKHRWKHUVD
decision that it makes based on a set of priorities programmed into
its code by humans.

z Model-based systems are relatively slow, deliberative systems that

can weigh information from multiple sensors, calculate higher-level
inferences, update maps, and make plans for actions in the short
and long term. At the heart of this kind of controller is a model of
the world that includes the current state of the robot in that world.
As both change, the model has to be updated in order to remain
accurate and useful. With a model-based system, decisions are
made based on the current state of the robot and the desired future
states of the robot that are required in order to achieve the explicit
goals programmed into its code by humans.

z The tension between these two systems can be characterized as

a trade-off between speed of action and complexity of plan. In a
military situation, we often need both. Many robots combine fast
and slow autonomous control architectures.
Lecture 18—Why Military Robots Are Different

z Both types of control architectures rely on the quality and accuracy

of the information provided by sensors and by the models of the
ZRUOG %XW QRWKLQJ LV HYHU FHUWDLQ DQG WKH ¿HOG RI probabilistic
robotics recognizes that explicitly.

z Probabilistic robotics grapples with uncertainty in mathematical

terms, building and updating models of the world and the robot
that create robust behavioral choices in the face of that uncertainty.
The autonomous robot makes decisions based on what it knows and
what it doesn’t know, and its degree of uncertainty about both.

z Probabilistic robotics has helped build the very successful and safe
autonomous robots known as self-driving, or driverless, cars. In this
area of robotics, in addition to making autonomous robots operate

160
 VDIHO\ LQ G\QDPLF UHDOZRUOG VLWXDWLRQV ZH FDQ ZLWK VXI¿FLHQW
investment, make them operate much more safely than humans
working under similar conditions.

z In warfare, the equivalent of the rules of the road are international

treaties, such as the Geneva Conventions. Protocol I is an
amendment from 1977 outlawing, among other things, attacks on
places of worship. This rule could be written into the programming
RIDQXQPDQQHGDHULDOYHKLFOH$Q\SODFHRIZRUVKLSLVLQGLFDWHGRQ
the robot’s model of the world, its map, as an off-limits area.

z (YHQ ZLWK WKLV SURJUDPPLQJ DQG D SUREDELOLVWLF FRQWUROOHU WKDW

accounts for uncertainty, we could never expect a robot to operate
perfectly. The robot could accidentally bomb a place of worship
if the information in its world model were inaccurate. We would
DOVRFRQVLGHULWDQDFFLGHQWLIDPLVVLOH¿UHGE\WKHURERWLQWHQGHG
for another target malfunctioned, for reasons that had nothing
to do with the programming of the robot, and hit the place of
worship. Also, there’s nothing to keep an unauthorized human from
reprogramming the unmanned aerial vehicle.

z One thing that we wouldn’t have to worry about anytime soon is

a robot with emergent self-consciousness, free will, and a sudden
GHVLUH WR NLOO KXPDQV 1R RQH ZRUNLQJ LQ URERWLFV DUWL¿FLDO
intelligence, psychology, or neuroscience has ever created a
computer program or machine that is conscious and self-aware in
the way that most humans think of those experiences. While we can
build robots with behavioral autonomy, their autonomy is not of the
same type that we humans possess.

z And that difference in how robots and humans work is a great thing,
practically speaking. That’s why robots are better drivers than
humans. They don’t get tired, or distracted, or emotional. And that’s
why—if programmed correctly—robots can be better soldiers. But
we have a long way to go.

161
 Automation and Navigation
z One area in which robots are not yet capable of being better soldiers
WKDQKXPDQVLVLQWKHLGHQWL¿FDWLRQRIIULHQGRUIRHLQQRQWUDGLWLRQDO
battle situations. This is a huge problem for humans, too; an
occupying army faces this problem continuously.

z 7KHJHQHUDOSUREOHPLVFDOOHGFRPEDWLGHQWL¿FDWLRQDQGWKHPLOLWDU\
has been working on a technological solution since World War II
transponders with coded signals on vehicles or individuals can be used
to positively identify a friend, as can careful tracking from known
starting positions. But confusion exists if a transponder is broken or
tracking is interrupted. In addition, the activation of a transponder
can also alert the foe to the presence of your own soldiers.

z 7KH GLI¿FXOW\ RI DXWRPDWLQJ FRPEDW LGHQWL¿FDWLRQ DQG JHQHUDO

concerns about maintaining safety and control are two of the
reasons that the most widely deployed military robot—the Predator
unmanned aerial vehicle and its descendants—is operated by a
human pilot and given only partial autonomy.

z In robotics, navigation is one of the most fundamental problems

Lecture 18—Why Military Robots Are Different

that any mobile robot has to solve. You have to know where you are
to get where you want to go. GPS works as a system of navigational

‹LFKRODNRYL6WRFN7KLQNVWRFN

The Predator drone and its descendants are given only partial autonomy to
PLQLPL]HHUURUVLQFRPEDWLGHQWL¿FDWLRQDQGPDLQWDLQVDIHW\DQGFRQWURO

162
 beacons that can be used to calculate your position. The current
*36V\VWHPLVDQHWZRUNRI(DUWKRUELWLQJVDWHOOLWHV7KHURERW
needs the ranges and locations of four satellites to calculate its
position using a process called trilateration.

z What makes accurate navigation particularly important to unmanned

military robots like the Predator is that they can remain on remote
station for more than 24 hours. If the unmanned aerial vehicle were
trying to navigate by dead reckoning, errors in estimating position
would increase with time and distance.

z Navigation does not present a problem for other types of military

URERWV DV ORQJ DV WKH\ RSHUDWH LQ VLJKW SHUIRUPLQJ MREV LQ FORVH
proximity to the soldier in charge of them. PackBot, created and
produced by iRobot, is a class of small, mobile remote-controlled
robots with partial autonomy that have been customized for a
variety of different tasks in close proximity to the operator.

z They are built to dispose of bombs, inspect for improvised

H[SORVLYHV SURYLGH VLWXDWLRQDO DZDUHQHVV WR ¿UVW UHVSRQGHUV
aid search and rescue, collect video and air samples in hazard
responses, and detect and localize snipers.

z In each of these tasks, PackBot

increases the standoff distance
between a human soldier and
‹-DQ3LWPDQ*HWW\,PDJHV1HZV7KLQNVWRFN

a potential threat. As a stand-

in for the soldier, PackBot
allows operators to have
virtual presence and virtual
agency. Virtual presence lets
the operator perceive the
world from the robot’s point
of view. Virtual agency lets
Militaries use robots to dispose of
the operator act in the world bombs so that humans aren’t put
as the robot. in danger.

163
 Bioinspired Robots
z The legged squad support system (LS3) is basically a robotic
SDFNKRUVHEXLOWLQFRQMXQFWLRQZLWK'$53$E\%RVWRQ'\QDPLFV
This four-legged robot, which is the size of a horse, can carry 400
pounds and, most importantly, can do so over rough terrain. Because
the design of LS3 is inspired by animals, it has a biomorphic body
plan. LS3 is partially autonomous.

z LS3 is the descendent of BigDog, which was a big breakthrough in

legged robots. Until BigDog, most legged robots were either upright
humanoids with two legs or sprawling six-legged insectoids. With
four legs, BigDog is more stable than a humanoid and able to carry
heavier loads. With an upright mammalian posture, BigDog has
better clearance than an insect body form. The problem with using
a dog, or any mammal, for a robot design is that they are unstable.

z To solve this problem, the engineers, led by Martin Buehler, had

to create a robot with dynamic stability. The trick is that BigDog is
almost always moving its legs, even if it is standing still. If you are
always moving, then you are actively keeping your center of mass
balanced between your legs. But dynamic stability is costly in two
Lecture 18—Why Military Robots Are Different

ZD\V ,W WDNHV HQHUJ\ WR DOZD\V EH PRYLQJ \RXU OHJV DQG LW DOVR
takes a lot of sensing to do the balancing act.

z In many ways, BigDog and LS3 epitomize bioinspired design in

robotics. Working from the anatomy and physiology of trotting
PDPPDOVHQJLQHHUVZRUNHG EDFNZDUG IURP QDWXUHLGHQWL¿HGWKH
key functional principles, and then built machines that worked in
the same way.

Missiles and Close-In Weapons

z 7KH¿UVWORQJUDQJHEDOOLVWLFPLVVLOHZDVWKH9URFNHWZKLFKZDV
created by the Germans in World War II and was used against the
allies. The V-2 is a robot in the sense that it operates on the principle
of sensor-guided movements.

164
 Robots in the Military

1944 V-1 bombers, unmanned aircraft (UA), are deployed by

WKH*HUPDQDLUIRUFHDJDLQVWWDUJHWVLQ(QJODQG

   URFNHWV WKH ¿UVW ORQJUDQJH EDOOLVWLF PLVVLOHV DUH

9
deployed by the German military against urban targets
LQ(QJODQG

  KH 54 3UHGDWRU DQ XQPDQQHG DHULDO YHKLFOH 8$9

7
built by General Atomics Aeronautical Systems, is
GHSOR\HG E\ D 86 1DY\$UP\ WHDP LQ WKH IRUPHU
Yugoslavia for reconnaissance.

  UHGDWRU 8$9V ¿UH PLVVLOHV IRU WKH ¿UVW WLPH LQ 86
3
campaigns, in Iraq and Afghanistan.

z The German engineers built an automated controller for guidance

based on a gyroscope and a pendulum, the Pendulous Integrating
Gyroscope Accelerometer (PIGA). The PIGA senses both the
magnitude of the acceleration and speed of the V-2 rocket in that
direction. If you know your acceleration, speed, how long you
are moving at that speed, and your angle of takeoff, then you can
calculate where you are going to land using the simple physics of
WKURZQEDOOVDQGRWKHUREMHFWVFDOOHGEDOOLVWLFV+HQFHDEDOOLVWLF
missile throws itself to hit a target.

z From a range of 100 miles or more, the V-2 could hit within 600
meters of its target. That’s a measure of its functional accuracy.
After WWII, the United States and the Soviet Union captured as
PXFK RI WKLV9 WHFKQRORJ\ DV WKH\ FRXOG7KH9 MXPSVWDUWHG
their own ballistic missile programs and their space programs.
Because of their accuracy, PIGA sensors allowed the V-2 rocket to
be independent of human control once the rocket was launched.

165
 z Fully autonomous robotic systems have also been developed for
defensive actions. The Phalanx, a close-in weapons system, is a
very important unmanned system that has been used by the U.S.
Navy since 1980.

z The task of Phalanx is more complex than that of a V-2. It’s not
MXVWWRODXQFKDQGODQGDQH[SORVLYHGHYLFH3KDODQ[KDVWRGHWHFW
evaluate, track, and engage targets that are rapidly moving and
attempting to destroy the Phalanx’s ship. The task of the Phalanx
is to be the last line of defense against anti-ship missiles and high-
speed aircraft. This is a do-or-die situation in which both speed and
accuracy are critical.

z The environment of Phalanx is likewise more complex than that of

D9,W¶VQRWMXVWWKHDLUEXWREMHFWVLQWKHDLUDQGREMHFWVRQWKH
ZDWHU$QGWKHUHDUHRWKHUVKLSVLQ\RXURZQÀHHWDURXQGWRRVRWKH
environment is mixed, with friends and foe sometimes intermingled.

z Because of the need for speed, the Phalanx needs a high degree of
DXWRQRP\ IURP KXPDQV 8QGHU PRVW EDWWOH FRQGLWLRQV WKH MRE RI
WKHKXPDQRSHUDWRULVWRKLWD³KROG¿UH´EXWWRQLIVRRUGHUHGE\WKH
Lecture 18—Why Military Robots Are Different

¿UHFRQWUROFRPPDQG(YHQWKRXJK3KDODQ[DXWRPDWLFDOO\GHWHFWV
tracks, and engages targets, it does not decide who is friend and
who is foe.

z All detected targets approaching the ship are treated as foe. This
LV RQH ZD\ DURXQG WKH GLI¿FXOW SUREOHP RI FRPEDW LGHQWL¿FDWLRQ
trying to categorize your targets. Admittedly, this is a very
dangerous way to program a robot.

Important Term

gyroscope 0HFKDQLFDO RU HOHFWURQLF URWDWLRQ VHQVRU WKDW LV XVHG IRU
maintaining or measuring orientation; often linked with an accelerometer for
more precise location in three-dimensional space.

166
probabilistic robotics$IUDPHZRUNIRUWKHGHVLJQRIURERWVWKDWH[SOLFLWO\
models the uncertainty inherent in the signals provided by sensors, the
models created by controllers, and the movement instantiated by actuators.

trilateration 0HWKRG IRU GHWHUPLQLQJ ORFDWLRQ XVLQJ WKH PDWKHPDWLFV RI

circles, spheres, and triangles.

Suggested Reading

Huang, Messina, and Albus, Autonomy Levels for Unmanned Systems

(ALFUS) Framework.
Singer, Wired for War, chaps. 1–5.
U.S. Department of Defense, Unmanned Systems Integrated Roadmap
FY2013–2038.

Question to Consider

1. The Aegis system is arguably one of the most complex multi-agent,

human-robot systems. If you were trying to design the next generation
of naval systems, what might you learn from the Aegis about
its vulnerabilities?

a. Detection is poor of small, slow weapons or robots of any type—

air, surface, or underwater.

b. Use of highly modularized subsystems, such as the close-in

targeting system, doesn’t allow full integration of modular systems
across the whole platform.

c. Humans in the loop slow down the response time.

167
 Extreme Robots
Lecture 19

R
obots with legs offer some of the most extreme physical challenges
in all of robotics. The physics of dynamic walking, whether with two
or four legs, is a balancing act similar to an inverted pendulum—and
LVMXVWDVXQVWDEOH(YHQPRUHH[WUHPHWKHSK\VLFVRIUXQQLQJLVFRPSDUDEOH
WRDSRJRVWLFNZKHUHEDODQFLQJLVLQWHUVSHUVHGZLWKVKRUWSHULRGVRIÀLJKW
Military packhorse robots already confront the challenges of dynamic
stability. The challenges become even greater in some of the largest, fastest,
and smallest of all legged or limbed robots, making them some of the most
extreme robots in existence.

Size versus Speed

z The laws of physics create a trade-off between size and speed. The
study of how size matters is called scaling. The most fundamental
scaling law is purely geometric. It’s the surface-area-to-volume
ratio, also known as the square-cube law, which impacts form and
function in biology and engineering.

z Isometric scaling refers to scaling something in equal measure;

proportional relationships are preserved as size changes. The warning
that the square-cube law broadcasts is that isometric enlargement
JURZVYROXPHIDVWHUWKDQVXUIDFHDUHD7KHLQFUHDVHLQVXUIDFHDUHDLV
QRWHQRXJKWR³NHHSXS´ZLWKWKHLQFUHDVHLQYROXPH$QLPSRUWDQW
Lecture 19—Extreme Robots

sequel of the square-cube law is that the mass and weight of an

animal or robot usually increases in direct proportion to volume.

z Isometric scaling of the area-to-volume ratio impacts the movement

of animals or robots at extremes of size. At very small sizes and
speeds, movement through air, and especially water, is dominated
E\WKHHIIHFWVRIYLVFRVLW\DPHDVXUHRIWKHVWLFNLQHVVRIDÀXLG$W
WKHRWKHUHQGRIWKHVFDOHIRUWKHELJDQGWKHIDVWÀXLGIRUFHVDUH
GRPLQDWHGE\WKHHIIHFWVRILQHUWLDWKHWHQGHQF\IRUREMHFWVWRNHHS
moving once in motion.

168
z For terrestrial animals and robots, those that don’t move about in
ÀXLG WKH ¿UVW FKDOOHQJH LV WR KROG XS WKHLU ZHLJKW DW OHDVW ZKHQ
standing still. Weight is the force generated by gravitational
acceleration acting on their body’s mass.

z Mass matters for movement. Newton’s second law tells us that

if you keep the force constant, then doubling the mass halves the
acceleration. In the case of muscles, the force that they can generate
is proportional to the cross-sectional area of those muscles. As
animals get bigger, the square-cube law tells us that the animal needs
to get bigger muscles or they will be able to accelerate only slowly.

z In robotics, one way to postpone the trade-off between size

and speed is to build walkers with very light skeletons. A totally
different way to solve this problem is to build more forceful motors.
The challenge is to increase force and power without increasing
weight, friction, or resistance.

z A team led by Dan Granett, a former engineer with NASA, is

working to create a new kind of power system for large robots
called hydraulic accumulators. Powered by a diesel engine, the
hydraulic system is pressurized, and then that potential energy can
EHTXLFNO\UHOHDVHGWRUDSLGO\PRYHMRLQWV

z The dinosaur-sized Tradinno robot, the world’s largest four-legged

walking robot, also uses a hydraulic system. It has a turbo diesel
HQJLQHWKDWSUHVVXUL]HVJDOORQVRIK\GUDXOLFÀXLGIHGWKURXJKD
VHULHVRIWXEHVDQGLQWRWKHOHJDFWXDWRUV7KDWK\GUDXOLFÀXLGPRYHV
WKHSLVWRQVWKDWLQWXUQFUHDWHWKHWRUTXHWKDWSRZHUVWKHMRLQWVRI
the legs.

z Making Tradinno walk was a serious challenge. The legs have to

VXSSRUWWKHZHLJKWRIWKHURERWDQGPRYHLQMXVWWKHULJKWZD\D
pattern of footfalls that we call a gait when we are thinking about
animals. This involves careful coordination. The engineers had to
carefully examine animal gaits and movement patterns in order to
solve this problem in Tradinno.

169
 z Hydraulic systems are useful because they allow the motor to be
centralized. The alternative, which we tend to see in smaller robots,
VXFKDV+H[\IURP$UFERWLFVLVWRGLUHFWO\GULYHHDFKMRLQWZLWKLWV
own motor. A great thing about individual motors is that you can
easily program different behaviors by controlling each motor.

z With hydraulics, you keep the motor or motors inside the body
DQGXVHÀXLGWRWUDQVPLWWKHIRUFHRXWWRWKHMRLQWV)RUYHU\ODUJH
robots, a hydraulic system is a clever way around a physical limit.

z Another kind of extreme is large exoskeleton robots, which are

powered exoskeletons that increase the capacity of human beings,
including their strength, speed, and ability to use and wield
weapons. The largest of these exoskeletons are not mobile because
it would take additional power to move legs around and to control
posture and balance.

z At the other end of the scale, at the really small end, there are
challenges of a different kind. There are mini robots based on the
ERGLHVRIÀLHVDQGEHHV7REXLOGDPLFURPDFKLQHWKDWÀLHVDQGFDQ
be powered and controlled, with the lateral maneuvers, has been a
ELJFKDOOHQJH)RUVPDOOURERWVDPDMRUSUREOHPLVKDYLQJHQRXJK
fuel, or energy storage, on board.

z Another problem related to power is that electromagnetic motors,

OLNH WKH RQHV XVHG WR SRZHU WKH ZKHHOV RU ÀDS WKH WDLOV LQ URERWV
GRQ¶WVFDOHZHOOWRYHU\VPDOOVL]HV7KHPDJQHWLFDQGHOHFWULF¿HOGV
Lecture 19—Extreme Robots

DWYHU\VPDOOVL]HVMXVWGRQ¶WSURGXFHPXFKIRUFH

z When you try to make microbots, the trade-off is that the longer that
you want your microbot to operate on its own, without a tether to an
off-board energy supply, the bigger you have to make the battery.
Bigger batteries mean bigger robots.

170
Nanobots versus Large Robots
z One intriguing way around the problem of putting energy and
control on board a tiny robot is to build a world outside the robot
that contains both. If you can’t modify the robot, modify the
environment. This approach is allowing nanobots to be created.
With nanobots, the idea is to make tiny robots that operate on the
level of molecules.

z The scale for nanotechnology is typically from 100 nanometers

down to one nanometer, which is one billionth of a meter. At this
scale is where single molecules, such as DNA, live. When we talk
about building nanomachines, we are talking about being able to
manipulate single molecules.

z Amazingly, a team at Tufts University has built an electric motor

from a single molecule. While these nanoscale motors aren’t yet
ready to start powering nanobots, there is work elsewhere on
nanoscale walkers. Kyle Lund and his colleagues at Arizona
State University built what they called molecular robots that can
PRYH DXWRQRPRXVO\ DQG LQ D SUHGLFWDEOH ZD\ 7KH\ DUH MXVW IRXU
nanometers long. Lund and his team built an environment out of
what is called DNA origami that contained information for the
nanobots to start, follow, turn, and stop.

z One trade-off that nanobots encounter is that they move slowly.

Lund’s molecular robots, when they are really moving quickly, can
move about 90 nanometers in 30 minutes, or 3 nanometers (about
the length of its body) per minute. As a comparison, in terms of
absolute speed, Tradinno moves much faster, but its relative speed
is almost two body lengths per minute. By this measure, the largest
and smallest robotic walkers move about the same relative speed,
about one to two body lengths per minute.

z OutRunner, built by Robotics Unlimited, is almost 20 times faster

than Tradinno. OutRunner moves about 900 body lengths per
minute, a speed in terms of body length that is hundreds of times
faster than Tradinno or Lund’s molecular robots.

171
 z Robots at this intermediate size can be built to be the fastest-
legged creatures. OutRunner is biologically inspired. The legs are
key, acting like individual pogo sticks in much the same way that
your legs work when you are running. As the leg hits the ground, it
compresses, storing energy in internal springs. That spring energy
can be reused to help propel OutRunner forward, pushing it forward
as the legs rotate backward.

z 2XW5XQQHULVDYHU\HI¿FLHQWUXQQHUSHUVWHSRUSHUERG\OHQJWK
very little energy is used. The battery on board will last for two full
hours—outstanding mission time for something moving so quickly.
$QRWKHUDVSHFWRI2XW5XQQHU¶VGHVLJQWKDWPDNHVLWHI¿FLHQWLVLWV
springlike legs, which store and return energy for propulsion. Also,
those legs are lightweight and don’t have to be accelerated back and
forth, like human legs do.

Robot Legs
z Speed is the product of stride length and stride frequency. What
limits your speed as you walk is, in part, that your stride length
is limited by the length of your legs. When you run, you actually
launch yourself into the air. You are leaping from one foot to the
other, and that aerial suspension increases your stride length.

z Based on stride length alone, we would expect the largest animals

to be the fastest. A horse is a great example of a big and very fast
animal that takes advantage of its long legs.
Lecture 19—Extreme Robots

z If we were to redesign OutRunner to make it faster, we couldn’t

make the legs longer or have the legs spin faster, because that would
break the legs. They would start snapping off. Could we make the
legs stronger by adding material, making the legs thicker? This
adds mass, and additional mass also acts to increase the energy of
impact of the foot with the ground, so the legs would still break off.
:KDWDERXWDGLIIHUHQWNLQGRIPDWHULDOVRPHWKLQJOLNHFDUERQ¿EHU
WKDW LV VWURQJHU ZLWKRXW DGGLQJ PRUH PDVV" 7KHUH¶V QR LQ¿QLWHO\
strong material, so legs will break if they are long enough or spin
fast enough.

172
 ‹UHJROXLVL6WRFN7KLQNVWRFN
Horses are fast due to a combination of size and stride length.

z 6FDOLQJ ZRUNV IRU DQLPDOV DQG URERWV <RX QHHG WR VFDOH \RXU
skeleton, your support structure, allometrically—which is a
variation from isometric scaling—instead of linearly to compensate
for volume and mass increasing faster than the bone’s cross-
VHFWLRQDODUHD7KLVLVRQHRIWKHPDMRUWUDGHRIIVLQOHJJHGURERWV
Longer legs need to be stronger.

z Speed and size are linked to each other by the laws of physics.
The ability to generate more force per unit mass is key for the
largest-limbed robots. To reduce mass as we push the upper limits
of size, we can use strong materials and lightweight structural
arrangements to overcome problems that come along with simple
isometric scaling. The ability to have energy and motors on board
are challenges for the smallest robots.

z One way around this whole problem with legs is to turn them into
wheels. Wheels can spin quickly, carry heavy loads, and are less
likely to break. But legs can be picked and placed, lifted up and over
REVWDFOHVRUVWHSVDQGWKDWLVYHU\GLI¿FXOWWRGRZLWKZKHHOV

173
 z Robots of extreme size or speed pose extreme challenges for
robotics, and using legs increases those challenges even more. But
for moving through extreme environments, robots with legs are
extremely useful.

Activity

You can build your own extreme robot, using biomechanical

simulation tools, which are perfect for creating crazy, new, and
extreme robot designs.

$QLPDW/DELVDIUHHVLPXODWLRQV\VWHPKWWSDQLPDWODEFRPWKDW

allows you to build whole agents (animals or robots), limbs with
DFWXDWRUV DQG VHQVRUV ZLWK VHQVRU\ ¿HOGV 1HXUDO FRQWURO V\VWHPV
are an explicit part of each system, connecting and coordinating
sensors and actuators.

7U\WKHKH[DSRGZDONLQJURERWWXWRULDO

KWWSDQLPDWODEFRP+HOS7XWRULDOV([DPSOHV+H[DSRG5RERW

Important Terms

hydraulics Using pressurized liquid to create mechanical motion.

Lecture 19—Extreme Robots

square-cube law 0DWKHPDWLFDO SULQFLSOH WKDW GHVFULEHV WKH UDWLR RI WKH
VXUIDFHDUHDWRWKHYROXPHRIDFODVVRIVLPLODUO\VKDSHGREMHFWVWKDWGLIIHU
in size alone.

174
 Suggested Reading

3LQKHLUR+DQ6KLKDQG<DQ³&KDOOHQJHVDQG2SSRUWXQLWLHVIRU6WUXFWXUDO
'1$1DQRWHFKQRORJ\´
Vogel, Comparative Biomechanics, appendix 3.
:RRG 1DJSDO DQG:HL ³7KH 5REREHH 3URMHFW ,V %XLOGLQJ )O\LQJ 5RERWV
WKH6L]HRI,QVHFWV´

Questions to Consider

1. One of the problems associated with becoming larger is that it is easier

to fall down and hurt oneself. Small animals, like mice, can fall long
distances and suffer little damage; large animals, like horses, break
bones when they fall. Part of the reason for this difference boils down
to differences in kinetic energy, the energy of motion. Kinetic energy is
half the product of the mass of the animal or the robot and the square
of the velocity of its motion. Given that formula, what is the difference
LQ WKH NLQHWLF HQHUJ\ RI D NLORJUDP PLFURERW À\HU FUDVK ODQGLQJ
at 1 meter per second and a 100-kilogram robotic cheetah running into
something at 10 meters per second?

2. *LYHQZKDW\RX¶YHMXVWOHDUQHGDERXWNLQHWLFHQHUJ\ZKDWZRXOG\RX
say is the functional trade-off for pushing the extremes of speed?

175
 Swarm Robots
Lecture 20

O
ne of the grand challenges in robotics is to get robots to work
together as a group to do things that they couldn’t do alone. The
IXQFWLRQDOEHQH¿WRIZRUNLQJLQDJURXSLVWKHSUHPLVHRIVZDUP
robotics, which deals primarily with groups of simple, similar mobile
robots. Swarm robotics proves that sets of behaviors and tasks, such as
coordinating groups and working together to build structures, can be
DFFRPSOLVKHG E\ URERWV 6ZDUP URERWLFV LV D ¿HOG WKDW LV EDUHO\ LQ LWV
infancy, but it has vast potential.

Swarm Robotics
z Swarm roboticsLVUHODWHGWRWKH¿HOGRIFRPSXWDWLRQDOPRGHOLQJ
known as agent-based modeling. Physicists are also interested in
swarms and have created self-propelled particle theory, where a
particle is an autonomous agent, to formalize the behavior of the
agents and the swarm. The agent is any autonomous actor, such as a
physical robot or an animal.

z The core principle of agent-based models and self-propelled particle

theory is that simple rules enacted by each agent independently
generate complex behavior of the group as a whole. When we apply
this principle in swarm robotics, we frame it as a matter of creation
DQGFRQWURO+RZGRZHFUHDWHDQGWKHQFRQWURODVZDUP"
Lecture 20—Swarm Robots

z In terms of controlling the behavior of the robot, its controller, a

FRPSXWHU ORDGHG ZLWK FRGH GRHV RQO\ SDUW RI WKH MRE :KHQ ZH
FKDQJH WKH FRQ¿JXUDWLRQ RI WKH ERG\ WKH EHKDYLRU RI WKH URERW
changes; the program on the computer controller is unchanged. The
robot interacts with the world through its body, and as the body or
WKH ZRUOG FKDQJH VR GRHV WKH VSHFL¿F SDWWHUQ RI WKH LQWHUDFWLRQ
That interaction of the robot with the world is called behavior.

176
 z Ironically, and importantly, behavior is only partially controlled by
what we call the controller. Control is distributed throughout the
URERWZRUOGV\VWHP(DFKSDUWKDVDUROHWRSOD\7KLVSULQFLSOHFDQ
be referred to as distributed control of behavior. This principle can
be applied to a single agent interacting in the world, but we can also
see it in action in swarms.

Ant Robots and Construction Robots

z Algorithms of swarm navigation have been implemented in robotic
ants. Simon Garnier and his colleagues in the SwarmLab at the
New Jersey Institute of Technology have built small behavior-based
URERWV WKDW IROORZ VLPSOH UXOHV (DFK URERW KDV LQIUDUHG VHQVRUV
RULHQWHGODWHUDOWRDYRLGKLWWLQJREMHFWVDQGWKHQWZROLJKWVHQVRUV
pointed upward.

z In a maze, a robot heads out from the starting point and blunders its
ZD\DURXQGHYHQWXDOO\¿QGLQJDSDWK,WOHDYHVQRWDFKHPLFDOWUDLO
but a trail of light on the ground, thanks to a special set of overhead
spotlights. The next robot that comes along uses its light sensors to
GHWHFWWKHSDWKODLGGRZQE\WKH¿UVWURERW,WDOVROHDYHVDWUDLORI
light that makes for any even brighter path for the next robot, and
that makes for faster navigation.

z Garnier’s ant robots are individual agents that use simple rules to
enhance the behavior—in this case, the navigation—of the group
as a whole.

z A diametrically opposed approach to the control of swarms is to

KDYHWKHDJHQWVEHFHQWUDOO\FRQWUROOHGE\DVLQJOHV\VWHP$QQMRH
Wong-Foy at Stanford Research Institute takes this approach with
tiny construction robots. These microbots, which can vary in size
from one to 10 millimeters in length, are magnetic and are propelled
E\FKDQJHVLQWKHHOHFWURPDJQHWLF¿HOGRQWKHSULQWHGFLUFXLWERDUG
over which they hover.

177
 z %HFDXVHWKH\XVHPDJQHWLFIRUFHVDQGFDQPDQLSXODWHREMHFWVWKHVH
PLFURERWV KDYH EHHQ FDOOHG GLDPDJQHWLF PLFURPDQLSXODWRUV (DFK
microbot can carry a different set of arms so that specialized worker
FDVWHVOLNHDQWVFDQSRSXODWHWKHVZDUP2QHFDVWHFDQFDUU\¿EHUV
and another superglue. Programmed to work together, they can
construct truss-like systems.

z These microbots are very interesting not only because they can be
programmed to build things at a tiny scale, but also because they do
so by violating the core principle of agent-based models and swarm
URERWLFV6LPSOHUXOHVHQDFWHGE\HDFKDJHQWLQGHSHQGHQWO\JHQHUDWH
complex behavior of the group as a whole. Instead, the rules, while
they may be simple, are enacted by the central control program that
PRYHV WKH PDJQHWV DURXQG RQ WKH FLUFXLW ERDUG (DFK PLFURERW LV
not an autonomous agent; they don’t act on their own.

z The core principle was discarded with microbots for the simple,
practical reason that they are too small to carry an energy supply, a
computer controller, or sensors. All of those parts and processes are
off-loaded to a smart and centrally controlled environment. Once
you do this, the limit to the size of the microbots is only limited
E\\RXUDELOLW\WRFRQWUROWKHPDJQHWLF¿HOGWKDWGULYHVWKHP7KLV
LV D WHUUL¿F LQQRYDWLRQ LQ VZDUP URERWLFV WKDW KDV WKH SRWHQWLDO WR
change micromanufacturing.

Control in an Unstructured World

z We can’t always create the highly structured world required for
preprogrammed and centralized control of a group of agents.
Lecture 20—Swarm Robots

Almost anything we might want a swarm of robots to do outside

will be working in an unstructured world that is rapidly changing.

z Leandro Marcolino at the University of Southern California has

been conducting experiments on small mobile robots using a group
of six identical e-puck robots attracted to a goal—namely, to move
from one side of an arena to the other. There are two groups moving
LQWKHRSSRVLWHGLUHFWLRQVRWKH\KDYHWUDI¿FWRQDYLJDWH

178
z In addition to being programmed with the goal to reach the other
VLGHHDFKURERWDOVRKDVDQDYRLGDQFHUHÀH[EXLOWLQ:KHQDURERW
hits or approaches another robot closely, it will stop, back up, and
then set off in a slightly different direction toward its goal.

z 0DUFROLQRFDOOVWKLV¿UVWSURFHVVXQFRRUGLQDWHGLQWKHVHQVHWKDW
each agent is acting independently. And both groups are able to
make it past the other. However, the time to navigate past the
other group is reduced by 50 percent with coordination among
the individuals.

z 7KH FRRUGLQDWLRQ DOJRULWKP LV VLPSOH 7KH ¿UVW URERW WR GHWHFW
congestion up ahead signals a warning to the rest of its team.
The whole group then switches to the state of deviating from the
FRQJHVWLRQ DW D SUHVHW DQJOH DV LQGLFDWHG E\ /('V LOOXPLQDWLQJ
7KH UHVXOW LV D EHDXWLIXO GLVSOD\ RI ZHOOEHKDYHG URERWLF WUDI¿F
0DUFROLQR¶V WUDI¿F H[SHULPHQWV SRLQW RXW WKH XVHIXOQHVV RI
communication among agents in a swarm.

z 8VLQJDVLPLODUSURFHVVRIQRQYHUEDOFRPPXQLFDWLRQ¿VKVZLPPLQJ
in a school may avoid obstacles or threatening circumstances. The
VFKRROLQJEHKDYLRURIUD\VRUWKHÀRFNLQJRIELUGVLVQRWSUHSODQQHG
and there is not some kind of group leader. That is, even though one
individual may be in the front of the formation, different individuals
WDNHXSWKDWSRVLWLRQDOOWKHWLPH7KHIDFWWKDWVFKRROVRI¿VKDQG
ÀRFNV RI ELUGV FDQ FRRUGLQDWH PRYHPHQWV ZLWK QRQYHUEDO VLJQDOV
suggests the possibility of simple algorithms.

z In 1987, a computer scientist named Craig Reynolds was looking

for a way to realistically simulate how individuals move in a
group. Agent-based modeling has its origins in computer-based
DUWL¿FLDO LQWHOOLJHQFH VLPXODWLRQV DQG 5H\QROGV¶V ZRUN ZDV D
typical example.

z 5H\QROGVSURSRVHGWKUHHYHU\VLPSOHUXOHVIRUVZDUPLQJZKDWFDQ
be called attraction, avoidance, and alignment. These guide the
movement of each individual in the group. Attraction is the force

179
 that makes you want to move toward other individuals. Avoidance
is the force that repels you from others, keeps you from running
into them. Alignment is the force that turns you to move in the same
direction as others.

z If all three rules are working at the same time, this swarm algorithm
creates a chain reaction, coordinating the motions of the whole
group. Reynolds’s three simple rules allow anyone to make realistic
computer simulations.

Neighbor Awareness
z Swarming requires that you be aware of what your neighbors are
doing. We humans might naturally think of neighbor awareness as
VRPHWKLQJYLVXDOEXWLWGRHVQ¶WKDYHWREH:KHQ¿VKVZLPZLWKRXW
any available light, they can still school, using a sensory system
called the lateral line, an earlike series of external hairs that detect
FKDQJHV LQ KRZ WKH ZDWHU LV ÀRZLQJ DURXQG WKH ERG\7KH FORVHU
\RXVZLPWRDQRWKHU¿VKWKHPRUH\RXUÀRZZLOOEHGLVWXUEHGE\
their wake.

z But vision-based swarming works, too, as seen in nature and in

robot swarms. A Lily robot, created by CoCoRo, moves underwater
DQGXVHVWKHEOXHOLJKWVRQLWVERG\WRVLJQDORWKHU/LO\URERWV(DFK
Lily also has photodetectors to pick up those signals.

z Lily robots are detecting what their nearest neighbors are doing.
If a neighbor’s blue lights are on and blinking, then the Lily that
detected that signal starts blinking its blue light. Then, the next
Lecture 20—Swarm Robots

neighbor detects the blinking lights and starts blinking, and so on.

z In the case of this group of Lily robots, the chain reaction of

blinking allows information to be sent along the physical chain of
robots. In this situation, the message is that the robots have detected
a robot with blinking blue lights.

180
 ‹5LFKDUG%URRNVL6WRFN7KLQNVWRFN
Fish swimming without light are still able to school using a sensory system
FDOOHGWKHODWHUDOOLQHWKHFORVHURQH¿VKVZLPVWRDQRWKHUWKHPRUHWKHÀRZ
will be disturbed by its wake.

z 7R NHHS WKH /LO\ URERWV RUJDQL]HG WKH &R&R5R SURMHFW XVHG D
VZDUP DOJRULWKP EDVHG RQ WKH EHKDYLRU RI KRQH\EHHV FOXVWHULQJ
7KH\FDOOHGLW%((&/8675RERWVVWDUWE\PRYLQJUDQGRPO\,ID
robot detects an obstacle ahead, it stops and listens for a signal that
LQGLFDWHVWKDWWKHREMHFWLVDQRWKHUURERW,IDQRWKHUURERWLVGHWHFWHG
the focal robot measures the intensity of light. The higher the light
intensity, the longer it waits before returning to random movement.

z %((&/867LVDYHU\VLPSOHDOJRULWKPIRUVZDUPLQJDQGLWZRUNV
in the water, with Lily robots, and on land with Jasmine robots.
7KH&R&R5RSURMHFWHQYLVLRQVXVLQJVZDUPVRIXQGHUZDWHUURERWV
WRVHDUFKPRUHHI¿FLHQWO\IRUWKLQJVOLNHVXQNHQWUHDVXUHGRZQHG
airliners, or leaks in underwater oil pipes. These search-and-detect
tasks are exactly the right kind for swarms of robots. You simply
can’t search far and wide very quickly with a single robot.

181
 z At least in some cases, you don’t need to sense other agents in order
to swim in a coordinated manner; sharing a goal and colliding with
each other are all you need to create teamwork. Sharing a goal is a
key principle in swarm robotics. But goals can be shared in a swarm
without all of the robots having to act in the same way at the same
time. This allows more complicated tasks to be accomplished, such
as mapping a novel environment and then navigating through it.

Activity

7U\WKH%RLGVVLPXODWRU

ZZZUXQWKHPRGHOFRPPRGHOV

This simulator runs on JavaScript, so you can operate it from

your web browser and think about the rules of engagement for
robotic agents.

Important Term

swarm robotics The coordination and control of multiple robots of similar

type, operating at the same time and place.

Suggested Reading

Floreano and Mattiussi, %LRLQVSLUHG$UWL¿FLDO,QWHOOLJHQFH, chap. 7.

Lecture 20—Swarm Robots

.HUQEDFK³,QWURGXFWLRQWR&ROOHFWLYH5RERWLFV5HOLDELOLW\)OH[LELOLW\DQG
6FDODELOLW\´
5H\QROGV³)ORFNV+HUGVDQG6FKRROV´
:HUIHO3HWHUVRQDQG5DGKLND³'HVLJQLQJ&ROOHFWLYH%HKDYLRULQD7HUPLWH
,QVSLUHG5RERW&RQVWUXFWLRQ7HDP´

182
 Other Resources

9LGHR RI WKH RULJLQDO %RLGV VLPXODWLRQ IURP  KWWSVZZZ\RXWXEH

FRPZDWFK"Y L4L9,$

Question to Consider

1. Do any of the following tasks strike you as a situation in which robots

PLJKWEHQH¿WIURPGHVLJQLQJVWLJPHUJ\LQWRWKHV\VWHP"

a. The Mars explorer robot Curiosity analyzes the chemical

composition of a rock outcropping.

E $
  ÀHHW RI VHOIGULYLQJ FDUV DQG WUXFNV WUDQVSRUWV SHRSOH DQG WKHLU
belongings to a new urban center across the country.

c. A group of search-and-rescue robots looks for a group of hikers that

might have been engulfed in an avalanche.

183
 Living Robots?
Lecture 21

T
hroughout this course, we’ve been paying close attention to the bodies
and behaviors of biologically inspired robots. But beyond body
shapes, movement patterns, or sensory processing, even core features
of life are beginning to become possible for robots. First, robots can eat,
obtaining and harvesting energy on their own. This is metabolism. Second,
robots can build themselves. This is growth. Third, robots can evolve. And,
¿QDOO\URERWVFDQUHSOLFDWHWKHPVHOYHVDQGEXLOGRWKHUURERWV

Harvesting Energy
z How close to living can we make a robot? Being alive means
that you can harvest energy. And when you can’t, you cease to
function—you die. Robots need to harvest and use energy, too. The
energy supply for robots is often a battery. It’s not fair to say that
the robot harvests energy if we humans put new batteries in.

z But Roomba goes back to its charger. Other robots can do even
PRUH7KH35WHVWEHGURERWIURP:LOORZ*DUDJHFDQRSHQGRRUV
IRULWVHOIPRYHRQLIDGRRULVORFNHG¿QGDVWDQGDUGZDOORXWOHW
DQG SOXJ LWVHOI LQWR DQ\ RXWOHW LW ¿QGV 2WKHU URERWV FXW WKH FRUG
Wave Glider uses wave power to swim and solar power to charge
batteries that power its sensors and communications systems.
Lecture 21—Living Robots?

z But none of these methods of energy harvesting is similar to what

animals do. We animals eat food. Our stomach, intestines, and
digestive organs work to chemically breakdown the organic matter
that we ingest and then extract—at a molecular level—chemical
potential energy for later use by our cells.

184
z ($75(QHUJHWLFDOO\$XWRQRPRXV7DFWLFDO5RERWLVDURERWFDSDEOH
of ingesting organic matter for food, or fuel. It’s a wheeled ground
robot under development by Robotic Technology Incorporated,
LQSDUWQHUVKLSZLWK'$53$7KHEDVLFLGHDRI($75LVVLPSOHLQ
FRQFHSW6HDUFKWKHHQYLURQPHQWIRURUJDQLFPDWHULDOFROOHFWLWDQG
then convert it into a usable energy source.

z The guts of the operation is an external combustion engine. Water

in one chamber is heated by the combustion of organic matter,
ELRPDVV LQ DQ DGMDFHQW EXW VHSDUDWH FKDPEHU 2QFH LQJHVWHG WKH
biomass, or food, is burned to provide heat. The heat from the food
boils the water. That heat is used in a heat exchanger to generate
HOHFWULFSRZHU%XW($75LVQ¶WGLJHVWLQJLWVIRRGFKHPLFDOO\

z However, even real chemical digestion and energy creation are

SRVVLEOH$WWKH8QLYHUVLW\RIWKH:HVWRI(QJODQG&KULV0HOKXLVK
and his colleagues have been working on a biologically based fuel
cell that uses bacteria—microbes—to convert things like rotting
fruit into electric energy. Fuel cells convert chemical energy into
electric energy, and they are popular as energy alternatives to
internal combustion engines in vehicles.

z Melhuish has put the microbial fuel cells into a robot called
(FR%RW,,$VWKHIXHOFHOOVZRUNWKHHOHFWULFFXUUHQWWKH\SURGXFH
charges capacitors, which are electronic devices built to store
up, and then quickly release, electric charge. The stored charge
LQ (FR%RW ,,¶V FDSDFLWRUV LV XVHG WR SRZHU HLWKHU LWV ZKHHOV RU
FRPPXQLFDWLRQV\VWHP(FR%RW,,KDVDFRQWUROOHUWKDWZRUNVZLWK
photodetectors to head the robot toward a light source, powered
by rotten apples.

z (FR%RW,,LVDW\SHRIJDVWURERWDURERWWKDWJHWVDOORILWVHQHUJ\
IURPWKHGLJHVWLRQRIIRRG(FR%RW,,LVRQWKHURDGWRWUXHHQHUJ\
DXWRQRP\DQGWKDWLVWKHJRDORI0HOKXLVK¶VUHVHDUFKSURMHFW

185
 z ,I ZH FRXOG FRPELQH (FR%RW¶V FKHPLFDO GLJHVWLRQ WDOHQW ZLWK
($75¶V DELOLW\ WR VHDUFK IRU DQG ¿QG DSSURSULDWH IRRG LQ WKH
environment, then we would be moving robots toward total energy
autonomy and operating like life-forms.

Development
z Another thing that life-forms do, that robots don’t, is build
themselves. When engineering robots or other devices, we normally
assemble the whole thing. One exception is the International Space
Station, which has had to function while we’ve been building it.

z How would you build a robot that builds itself? Sam Felton begins
with a sheet of composite material that includes paper, copper,
polyimide, and prestretched polystyrene. When you heat this
composite sheet of material, the polystyrene shrinks. If you arrange
the paper and polystyrene in the right way, the heating will cause
the layer to fold. Add heat, the sheet folds, and the sheet builds
itself. This is like a simple kind of origami, but instead of a human
doing the folding, the paper is folding itself.

z Felton and his team in Robert Wood’s lab at Harvard have

³SURJUDPPHG´WKLVVKHHWWRIROGXSLQDSDUWLFXODUSDWWHUQWRPDNH
legs, engage a motor, and walk away. This machine builds itself.
7KH³LQVWUXFWLRQV´IRUEXLOGLQJLWVHOIZHUHQRWZULWWHQLQFRPSXWHU
code or DNA. Instead, the instructions were built into the laminate
sheet. Felton and his team call this programming matter.
Lecture 21—Living Robots?

z A different approach to life cycles in robots has been taken by

Josh Bongard at the University of Vermont. He and his team have
focused on metamorphosis, the transition from one life cycle stage
to another. Working in digital simulation and with embodied robots,
they discovered that individuals programmed to change their bodies
over time found new, better behaviors more rapidly than robots that
held their bodies constant.

186
 z 7KH URERW VWDUWV E\ OD\LQJ ÀDW RQ WKH JURXQG ZLWK D VSUDZOHG
SRVWXUHOHJVRXW,WKDVDFRQWUROOHUSURJUDPZKRVHMRELVWR¿JXUH
out how to move those legs in order to move. Using an evolutionary
program, many different and randomly created controller programs
can be tried out. The best one is the one that moves the robot the
fastest across the ground.

z Once a good controller is evolved that works well for the sprawled
posture, it can be used as the starting point for when the robot
brings its legs underneath its body. If the body changes in one fell
swoop, then it takes longer to evolve a controller that works well
compared to allowing the change in the body’s posture to be part
of the robot’s development. During development, the legs slowly
change their posture. That helps bridge the original controller to the
¿QDOFRQWUROOHU7KHUHVXOWLVWKDWGHYHORSPHQWVSHHGVXSHYROXWLRQ

z 7KLV PHWDPRUSKRVLV VKRZV WKH EHQH¿W RI KDYLQJ D ERG\ WKDW FDQ
FKDQJHDQGWKHEHQH¿WRIKDYLQJDSURFHVVWKDWFDQVHDUFKIRUQHZ
controllers that can the make best use of the ever-changing body.
That search process within the life of an individual can be thought
of as the compound processes of learning and development. This
takes place by trial and error.

Evolution
z A search process that is related to development is evolution. One
of the differences between development and evolution is that while
development describes the changes taking place to an individual,
evolution describes the changes that are taking place from one
generation to the next, from one group of individuals to another
group of related individuals—from parents to offspring.

z Like development and learning, evolution works by trial and error,

EXWRQDGLIIHUHQWVFDOH(YROXWLRQLVFKDUDFWHULVWLFRIOLIHDQGXVLQJ
the process of evolution, we can evolve robots.

187
 z For roboticists, evolution is one way to design robots. Standard
engineering and hacking both involve humans making decisions
about the design of the robots, but this is not so with evolution.
We take the humans out of the design loop. This is like taking
humans out of the control loop, allowing robots to perform tasks
autonomously. With evolution, we are allowing robots to design
themselves autonomously.

z We can boil life down to autonomy. All life-forms eat, behave,

develop, and evolve on their own. As we work from nature to create
robots that have more properties of life, we are extending the kinds
RI DXWRQRP\ WKDW WKH\ KDYH (DWLQJ EHFRPHV HQHUJ\ KDUYHVWLQJ
Behaving becomes performing tasks. Development becomes self-
assembly. And evolution becomes automatic redesigning.

z PreyRo, a kind of Tadro robot, is a model of the 400-million-year-

ROGH[WLQFWIRVVLO¿VKFDOOHG'UHSDQDVSLV$WHDPDW9DVVDU&ROOHJH
is trying to understand what kinds of selection pressures may have
FDXVHGDQDQFLHQW¿VKOLNH'UHSDQDVSLVWRHYROYHWKHERQHVLQWKHLU
back that we humans have. Those bones are called vertebrae. The
answer lets us understand how vertebrates got their start.

z The team’s guess was that predators might be selecting for faster
¿VKRYHUVORZHURQHV%DVHGRQELRPHFKDQLFDOWHVWVZLWK¿VKDQG
VZLPPLQJURERWVWKH\FRUUHFWO\SUHGLFWHGWKDWIDVWHU¿VKZRXOGEH
the ones with more vertebrae in their back. To test the evolutionary
idea about the importance of vertebrae, they put different individuals
Lecture 21—Living Robots?

of PreyRo, each with a backbone with different numbers of bones,

into a tank with a robot predator called Tadiator.

z In the evolutionary trials, the team started with two behaviorally

autonomous robots, each with different goals. PreyRo was one
of six individuals in its population, and it was competing against
the other PreyRos to see who is selected—(indirectly) by the
SUHGDWRU²WREHSDUHQWVIRUWKHQH[WJHQHUDWLRQ(DFK3UH\5RZDV

188
 cloned so that there were three identical triplets to test against the
predator. With six different genotypes, that made for 18 PreyRo
tails that were tested every generation.

z $OOWKDWDQ\LQGLYLGXDO3UH\5RZDQWVWRGRLV³HDW´OLJKWVRLWKHDGV
toward the light. The light is a metaphor for food. One PreyRo was
tested at a time in the tank, and while it was trying to eat, Tadiator is
RQWKHORRVH$OOWKDW7DGLDWRUZDQWVWRGRLV³HDW´RWKHUURERWVVRLW
heads toward PreyRo, which is food for Tadiator.

z The team ran trials on each individual in each generation multiple

times. They scored how well the individuals did, and the top three
PreyRos in each generation got to pass on their genes to the next
generation to make offspring.

z When we evolve robots, we don’t know how events will turn out.
We have a guess, such as more vertebrae will make for better
IHHGLQJDQGÀHHLQJEXWZHDUHQ¶WVXUH:HKDYHWROHWHYROXWLRQUXQ
its course. What determines the outcome are random factors, such
as genetic mutations that make sure that individuals are different,
and selection—in this case, the predator.

z (YROXWLRQ LV DQ DXWRPDWLF GHVLJQ SURFHVV ZH KXPDQV GRQ¶W

predetermine the outcome, but we do play a role. For the population
of PreyRos, they depend on us humans to conduct a mating
algorithm on the computer. We also use a genetic algorithm to add
in small amounts of mutation to the reproductive cycle. When we
produce the genomes for the next generation, those are actually
plans for us to construct, by hand, the offspring of the parents.

Reproduction
z The idea of self-replicating machines has been a serious scholarly
topic at least since mathematician John Von Neumann introduced the
thought experiment of a self-reproducing automaton in 1948. Much
more recently, NASA has funded research proposals into how robots
PLJKWUHSOLFDWHWKHPVHOYHVEHFDXVHWKLVFRXOGEHDYHU\HI¿FLHQWZD\
to begin mining, or other large-scale operations, on other planets.

189
 z A partial proof of concept came from Hod Lipson at Cornell
University. His team created a modular robot, built of cubes, that
FRXOG UHSOLFDWH LWVHOI LI JLYHQ RWKHU FXEHV (DFK PRGXOH FDOOHG D
molecube, is its own autonomous robot, and they link with other
FXEHV YLD UHYHUVLEOH PDJQHWLF LQWHUIDFHV (DFK PROHFXEH KDV WKH
plans for replication.

z Most importantly, each cube has an actuator that allows it to rotate

its faces about a single revolute axis. This allows the cube to grab
another cube and then rotate itself to lift that new cube to a position
above it; in essence, it stacks itself. With a column of multiple
molecubes, the modular robot can bend over to construct itself.

z A completely different approach has been taken by Mark Kim, who

runs the modular robotics lab at the University of Pennsylvania. He
and his colleagues have built FoamBot, which is able to build other
robots of various designs. FoamBot sprays foam onto actuators.
The foam hardens, and then the new FoamBot can walk away. The
NH\WRWKLVLGHDLVWKDWWKHUHLVRQHURERWZKRVHMRELVWRUHSURGXFH
or make, other robots.

z Both molecubes and FoamBot show us that the idea of reproducing

robots is feasible. For robot reproduction to work, what we still need
is the robot, or the group of specialized robots, to be able to harvest
HQHUJ\DQGUDZPDWHULDOVXVHZKDWWKH\KDUYHVWWRFUHDWH¿QLVKHG
PDWHULDOVDQGFRPSRQHQWVDQG¿QDOO\DVVHPEOHWKHPVHOYHV
Lecture 21—Living Robots?

Activity

/XGRERW(GXFDWLRQLQ(YROXWLRQDU\5RERWLFV

KWWSZZZXYPHGXaOXGRERWVLQGH[SKS3OD\3OD\

This is a great place to see robots evolving. Depending on your

interest level, you can follow the tutorials and learn how to create
digital simulations of evolving robots.

190
 Suggested Reading

Long, Darwin’s Devices.

6WDPS³$%ULHI+LVWRU\RI5RERW%LUGV´
Wood, Edison’s Eve, introduction and chap. 1.

Other Resources

Living Machines is an international conference for biomimetic technology

DQG ELRK\EULG URERWLF V\VWHPV WKDW FRPELQH PDFKLQHV ZLWK OLYLQJ WLVVXH
KWWSFVQHWZRUNHXOLYLQJPDFKLQHVFRQIDERXW

Questions to Consider

1. :KDWIHDWXUHVDQGRUEHKDYLRUVZRXOGDURERWQHHGWRKDYHLQRUGHUIRU
you to consider it to be alive?

2. If we are going to evolve robots to let them autonomously redesign

themselves, then it is critical to understand how evolution works.
Which one of the following best describes evolution as a process
involving robots?

D $
 QLQGLYLGXDOURERWWKURXJKGHYHORSPHQWDQGRUOHDUQLQJDGMXVWV
LWVERG\DQGRUFRQWUROOHUWREHFRPHEHWWHUDWZKDWHYHUEHKDYLRULV
being optimized.

E $
 JURXSRIUHODWHGEXWQRQLGHQWLFDOURERWVDUHMXGJHGLQGLYLGXDOO\
on how well they perform a given task. The designs that perform the
best are combined and altered to create a new group of nonidentical
robots to be tested.

c. A group of robots works together to solve a task, and by virtue of

ZRUNLQJWRJHWKHUWKH\DUHDEOHWRGRDEHWWHUMREWKDQDQ\RIWKHP
could do alone. They are able to encode how they accomplished
this so that the instructions can be passed down to other robots.

191
 Social Robots
Lecture 22

T
he complex behavior that we call social and emotional communication
LVD¿HOGULSHIRUURERWLFV6SRNHQZRUGVH[SUHVVLRQVWRQHRIYRLFH
and body language can all be used to create the user interface between
a human and a robot. Social signals, simple but powerful signals, can be
communicated very quickly. And thanks to their sensors and controllers,
robots can produce and recognize these standard social and emotional signals
to interact more effectively with humans.

Social Robotics
z $VRFLDOURERWLVDQDXWRQRPRXVURERWZLWKDYHU\VSHFL¿FIXQFWLRQ
to interact and communicate with humans and other autonomous
SK\VLFDODJHQWV%XW¿QGLQJVIURPVRFLDOURERWLFVFDQGLIIXVHPXFK
more widely into other kinds of robots as well.

z 6RFLDOURERWLFVLVD¿HOGWKDWFRPELQHV¿QGLQJVIURPHQJLQHHULQJ
biology, and human psychology. Of course, psychology continues
WREHDGDXQWLQJO\FRPSOH[¿HOG%XWZHNQRZHQRXJKDERXWKRZ
humans approach their world and communicate that we can state
the following principles to help us design social robots.
ż Humans tend to anthropomorphize. Most owners of a robot
vacuum cleaner give it a name. The classic example is when
we ascribe our own thoughts and emotions to our pets. We do
the same thing to each other, and we call that empathy.
Lecture 22—Social Robots

ż Humans expect a social agent to be interactive. We expect a

puppy to react to us with eye contact, sounds, or a change in
posture. With humans or social robots, if we ask a question, we
expect a response.

192
 ż Humans expect a social agent to show initiative. We expect
a puppy to have his or her own drives and desires, such as
wanting to play or chew on our shoes. We expect the same for
a social robot.

z These three principles get us started thinking about what robots need
to be socially competent. Social robots must be able to perceive
and express emotions; understand and express spoken language;
understand and express the
³ERG\ ODQJXDJH´ RI JD]H
expression, gesture, posture,
and movement; establish and
maintain a social relationship;
and have a personality.

z While all these skills feed

into social communication,
their diversity make it clear

‹&KXQJ6XQJ-XQ*HWW\,PDJHV1HZV7KLQNVWRFN
that the challenges for social
robotics is about more than
MXVWKDYLQJDVSRNHQRUZULWWHQ
conversation. That we can do
with chatterbots (or chatbots),
computer programs like Joseph
:HL]HQEDXP¶V(/,=$RI
whose text-based interactions
ODXQFKHG WKH ¿HOG RI DUWL¿FLDO The design of a social robot’s body
conversational entities. must invite human interaction.

z Social robotics focuses on communication that is embodied—

communication between two physical agents. Therefore, the design
of the body of a social robot is extremely important. In fact, if you
WDNHDEHKDYLRUEDVHGDSSURDFKWKH¿UVWWKLQJWKDWDURERW¶VERG\
has to do is to invite the human to interact.

193
 PARO
z Cuteness is an invitation to touch, and Takanori Shibata realized
that when he sought to design a therapeutic robot in the 1990s. He
chose the baby harp seal as a model, with its white fur and big black
eyes. In 2001, he revealed PARO as a therapeutic robot meant to
RIIHUWKHEHQH¿WVRIDQLPDOWKHUDS\LQHQYLURQPHQWVOLNHKRVSLWDOV
and extended-care facilities where having animals is problematic.

z When you touch PARO, it responds, thanks to tactile sensors, by

moving its head and body and making sounds. If you talk to PARO,
it has directional microphones that allow it to turn toward you, and
its controller has algorithms that allow it to learn words, such as
the name you give it, and recognize praise. PARO interacts with
language, even without producing language.

z PARO also learns what behaviors you prefer in order to initiate

interaction. Using a stroke as a sign of positive reinforcement,
PARO’s controller looks for patterns in its behavior that came
before the stroking. PARO will then try to initiate the stroking by
exhibiting those pre-stroking behaviors.

‹-XQNR.LPXUD*HWW\,PDJHV1HZV7KLQNVWRFN
Lecture 22—Social Robots

7KHWKHUDSHXWLFURERW3$52RIIHUVWKHEHQH¿WVRIDQLPDOWKHUDS\LQSODFHV
where having animals is not possible.

194
 z In addition to tactile and audio sensors, PARO has light sensors so
that it can tell the time of day and alter its behavior accordingly. It
also has posture and temperature sensors to give information about
how it’s being held and the nature of its environment.

z PARO is a great example of social communication through body

language and nonverbal auditory signals. Research has found that
the interactions of PARO robots with humans can reduce patient
stress, improve patient motivation, and improve communication
between patient and caregiver.

z Humans also use nonverbal body language in the form of facial

expressions and vocal intonation. These dynamic features of our social
communication are part and parcel to emotional communication.

Kismet and Jibo

z The most important early robot for emotional communication was
Kismet, a robot developed in the late 1990s, about the same time as
PARO. Kismet, designed by Cynthia Breazeal at MIT, made use of
the natural and nonverbal emotional communication that we have
with infants. The approach was revolutionary, and it set the stage
for humanoid social robots.

z Breazeal built and programmed Kismet to react to human emotions

in a way that signaled, to the human, Kismet’s own apparent
emotional state. Breazeal took advantage of our human tendency
to anthropomorphize.

z Humans interacting with Kismet use words and intonation. They

treat Kismet like a baby, scolding or praising. Intonation, stress, and
rhythm of the speech are all part of what linguists call prosody, and
%UHD]HDOSURJUDPPHG.LVPHWWRUHVSRQGWR¿YHW\SHVRISURVRG\
approval, prohibitions, attention, comfort, and neutral.

z Depending on the type of prosody, Kismet responds by changing

its posture, hanging or lifting its whole head, and by changing the
position of its ears and the shape of its mouth and eyes.

195
 z Kismet is communicating—through its postural changes—that it
has understood the emotional content of the human. Kismet signals
surprise by raising its eyebrows and ears and opening its eyes and
mouth. Kismet signals sadness by tilting its head down, lowering its
ears, and drooping its eyelids.

z +XPDQV LQWHUDFWLQJ ZLWK .LVPHW UHSRUWHG WKDW WKH\ HQMR\HG WKH

experience and felt some kind of emotional connection. Breazeal
quickly realized that Kismet was allowing her to study humans and
what triggers their emotions and empathy.

z (YHQ WKRXJK .LVPHW ORRNV OLNH ³MXVW´ D KHDG .LVPHW LV DFWXDOO\
an autonomous robot. It has sensors in the form of cameras and
a microphone. In addition to visual processing, Kismet has an
optional auditory system. Kismet also has the ability to vocalize—
not words, but coos that sound like those of a young child. The voice
synthesizer has parameters that can be altered to add emotional
TXDOLWLHVDQGDGMXVWWKHSHUFHLYHGSHUVRQDOLW\RIWKHYRLFH

z The sensory input, motor output, and heavy processing demands are
KDQGOHGE\GLIIHUHQWFRPSXWHUV(DFKPRWRUPXVWEHFRQWUROOHG
and each sensory input must be processed, but in addition, high-
level perceptions, motivations, and behaviors have to be modeled
and coordinated.

z The motivation system contains Kismet’s six basic emotions

GHULYHGIURPKXPDQSV\FKRORJ\DQJHUGLVJXVWIHDUMR\VRUURZ
and surprise. Kismet communicates its emotions through its facial
Lecture 22—Social Robots

expressions and nonverbal vocalizations, and these different

emotional states can be varied continuously in order to create
smooth transitions and intermediate neutral states.

z Kismet presents these emotional states in response to the person it

is interacting with. Kismet can also initiate a particular string of
interactions because it possesses what Breazeal calls drives that

196
 signal Kismet’s own agenda. These longer-term drives, along with
the immediate emotions, create the motivation system, all of which
is located in the computers networked to the Kismet head.

z In the short term, emotional interaction begins with Kismet

assessing whether a stimulus is positive or negative. One example
is personal distance. If you move in too close, Kismet backs away.
If you move too far away, Kismet makes cooing-type noises to draw
you in. All of this interaction—this social communication based on
emotional signaling—only works because both you and Kismet can
exercise attention. Breazeal built into Kismet an attention system
directly linked to the control of the eye motors.

z One of Breazeal’s later robotic creations seems to take lessons

from Kismet. Breazeal founded a company aimed at offering the
¿UVWLQH[SHQVLYHIDPLO\URERW-LERZKLFKKDVDERG\WKDWLVVPDOO
DQGVLPSOH,WVKHDGLVDURXQGVFUHHQWKDWFDQSURMHFWDQH\HDQG
swivel around.

z The ideas embodied in Jibo’s design don’t rely on facial expression

to engage humans. What has been carried forward from Kismet is
voice recognition, speech production, and attention seeking and
giving in the form of the movements that Jibo makes.

Pepper
z Launched in 2014, Pepper is built on the mobile humanoid design
but with a wheeled base instead of legs. Pepper is built to learn,
XQGHUVWDQGDQGDGMXVWLWVEHKDYLRUWRKXPDQHPRWLRQV3HSSHUZDV
WKHMRLQWSURMHFWRIWKH)UHQFKURERWLFVFRPSDQ\$OGHEDUDQDQGWKH
Japanese company SoftBank Telecom.

z Built with the purpose of being a social companion, Pepper can

sense and respond to some emotions in humans. Pepper has what
SURJUDPPHUV DUH FDOOLQJ DQ ³HPRWLRQDO HQJLQH´ DQG LW FDQ OHDUQ
about emotions through its interactions with you and from other

197
 Peppers interacting with other people. Its aim is to interact with you
in a way that is engaging and positive for you, whether you’re at
home or in a commercial setting.

z Pepper has four microphones, a pair of color video cameras in the

eyes and mouth, three-dimensional depth sensor behind the eyes,
touch sensors in the head and hands, bump sensors, lasers, sonar,
and a gyroscope in the torso. Pepper can sense the position of its
MRLQWV XVLQJ +DOO HIIHFW VHQVRUV ZKLFK XVH PDJQHWV WR PHDVXUH
motion. Pepper’s base has two sonars, six lasers, three bump
sensors, and another gyroscope.

z Pepper has three modules. Starting at the bottom, Pepper has a

transportation module. It can roll around at speeds of up to three
miles per hour and uses bump sensors, sonar, laser, and a gyroscope
WR QDYLJDWH DQG DYRLG REMHFWV ,Q WKH PLGGOH LV WKH PDQLSXODWLRQ
module, which has robotic arms that touch, reach, and gesture. On
the top is the interaction module, where vision, hearing, sounds,
and signals are created to facilitate communication.

z For a very sophisticated robot like Pepper, all three of these sets
of functions—interaction, manipulation, and transportation—are
likely to coexist and cooperate on the robot. We can think about
HDFKRIWKHVHPRGXOHVDVLWVRZQDXWRQRPRXVURERW(DFKPRGXOH
is independent from the other in terms of how it creates functional
ORRSVRIVHQVRUVDQGDFWXDWRUVWRGRLWVEDVLFVHWRIMREV+RZHYHU
HDFKPRGXOHLVGHSHQGHQWRQWKHRWKHUVWRJHWWKHRYHUDOOMRERIWKH
robot done.
Lecture 22—Social Robots

z Softbank expects Pepper robots to learn from their interactions

and enhance their learning by having them exchange information
with one another about what they learn. This may work especially
well for Pepper robots stationed in stores, where the same kinds
of interactions may occur many times. Pepper can also record
and summarize how people respond, making it a likely tool for
market research.

198
 z The most obvious change in context for a social robot is the person
with whom they are interacting. What a social robot learns about a
given person becomes their model of that person. But it might not
apply to other people.

Cobots
z Another change in context for social robots is when they ask for
help—and what they do if they don’t get it. Dr. Manuela Veloso
and her colleagues have invented robots that aren’t afraid to ask for
help, called cobots, which is short for collaborating robots. Cobots
KDYH D VSHFL¿F VHW RI WDVNV WKDW WKH\ GR 7KH\ FDQ QDYLJDWH WR
VSHFL¿FORFDWLRQVLQDEXLOGLQJGHOLYHUPHVVDJHVJUHHWDQGHVFRUW
YLVLWRUV WUDQVSRUW REMHFWV SURYLGH WHOHSUHVHQFH IRU FRQIHUHQFHV
and offer companionship.

z The only problem is that cobots don’t have arms, so they can’t pick
XSREMHFWVDQGWKHUHIRUHKDYHWRUHO\RQKXPDQVWRSODFHREMHFWV
in their basket. They have to ask humans to help. When a cobot gets
stuck, it waits for a human to come by. But cobots do get impatient.
:KHQ D FRERW VLPSO\ FDQ¶W ¿QG D KXPDQ WR KHOS LW HPDLOV 'U
Veloso. Then, she or one of her students has to go help out.

z Cobots are not only robots that can navigate around a building
using low-level behavior-based systems and high-level world-
model systems, but they are also robots that understand what
they can and can’t do. This is not very far from self-awareness,
at least in a very functional sense, and it shows the importance of
understanding context.

Important Terms

Kismet $XWRQRPRXV URERWLF KHDG FDSDEOH RI VRFLDO DQG HPRWLRQDO

interactions with humans that was built by Cynthia Breazeal in the late 1990s.

prosody 3URSHUWLHV RI VSHHFK RWKHU WKDQ WKH ZRUGV RIWHQ FRQYH\LQJ
emotional content.

199
 Suggested Reading

%UHD]HDO³&RJQLWLYH0RGHOLQJIRU%LRPLPHWLF5RERWV´
Vernon, $UWL¿FLDO&RJQLWLYH6\VWHPV, chap. 9.

Other Resources

7KH ,QWHUQDWLRQDO &RQIHUHQFH RQ 6RFLDO 5RERWLFV KWWSLFVRURRUJ 7KHLU

WDJOLQHLV³%UHDWKLQJOLIHLQWRPDFKLQHV´

Questions to Consider

1. What is a key feature in a robot’s communication with a human? That is,

where should a robot’s attention be directed in order to know whether a
human is attending to it?

a. Direction of the human’s gaze.

b. Response time following a verbal question.

c. Body language of the human.

2. The use of robots as caregivers opens new possibilities for improved

medical therapy. For example, what features might you want to build
LQWRDURERWZKRVHMRELWZDVWRPDNHVXUHWKDWDSDWLHQWDWKRPHNHSW
to his or her schedule of medications?
Lecture 22—Social Robots

200
 Humanoid Robots: Just like Us?
Lecture 23

T
UDGLWLRQDOO\WKHUHKDVEHHQDKXJHODJEHWZHHQDUWL¿FLDOLQWHOOLJHQFH
and humanoid robotics. But we have made considerable progress—if
not in closing the gap, at least in advancing what robots can do in the
physical world. Most of the advances in robotics have come from building
DERG\IRUDVSHFL¿FVHWRIWDVNVLQDJLYHQZRUNSODFH2IWHQWKHZRUNSODFH
LV PRGL¿HG WR KHOS WKH URERW 7KH VFLHQFH ¿FWLRQ DSSURDFK WR KXPDQRLGV
SUHVXPHV WKDW ZH DUH WU\LQJ WR EXLOG URERWV WKDW DUH MXVW OLNH XV ZLWKRXW
WKLQNLQJRIVSHFL¿FWDVNVRUZRUNSODFHV%\FRQWUDVWVRPHSURMHFWVFRQVLGHU
H[SOLFLWO\GH¿QHGWDVNVLQDKXPDQHQYLURQPHQWWKDWKDVQRWEHHQPRGL¿HG
to help the robot.

Humanoid Hands and Arms

z (DFK ERG\ SDUW RI D KXPDQRLG SURYLGHV LWV RZQ HQJLQHHULQJ
challenge. Hands turn out to be very complex to engineer, in part
because each one has at least 15 degrees of freedom, three per
¿QJHU (DFK GHJUHH RI IUHHGRP QHHGV WR EH XQGHU LQGHSHQGHQW
FRQWURO\HWLW¶VGLI¿FXOWWR¿WDOOWKHHOHFWULFPRWRUVRUSQHXPDWLF
PXVFOHVLQWRWKHFRQ¿QHGVSDFHRIDKXPDQVL]HGKDQG

z 7KH5$3+D(/5RERWLF$LU3RZHUHG+DQGZLWK(ODVWLF/LJDPHQWV
robotic hand is an air-powered robot hand, introduced in 2009 at
Virginia Tech. The use of compressed gas is what provides the
HQHUJ\DQGSUHVVXUHWRPRYHWKH¿QJHUVDOORZLQJWKHKDQGWRKROG
RQWROLJKWEXOEVRUKHDYLHUREMHFWVZLWKFDUHIXOO\FRQWUROOHGOHYHOV
of force.

z Dexterity can also be achieved with miniature direct current

motors embedded in the hand itself. In 2012, the German company
6&+81. UHOHDVHG D FRPPHUFLDO ¿YH¿QJHUHG KDQG (DFK ¿QJHU
has three degrees of freedom. The motions are driven by small,
geared direct current motors embedded in the palm of the hand and

201
 ‹PXUDWVHQHOL6WRFN7KLQNVWRFN
Hands are very complex to engineer, in part because each hand has at least 15
degrees of freedom.

LQ WKH ¿QJHU LWVHOI &DUHIXO FRQWURO RI WKHVH PRWRUV DQG IHHGEDFN
Lecture 23—Humanoid Robots: Just like Us?

from pressure sensors on the surface of the skin, allows the

SCHUNK hand to hold a key or a pin. It can also shake hands safely
with a human.

z Shadow Robot Company sells a Dexterous Hand with 129 sensors.

This is a hand that has 20 actuated degrees of freedom, plus four
DGGLWLRQDO MRLQWV WKDW DUH LQGLUHFWO\ DFWXDWHG 7KH WKXPE KDV ¿YH
extra degrees of freedom, and the pinky has four. In combination,
the hand has greater dexterity. The Dexterous Hand can operate as
an autonomous robot and can be attached to other robots.

z -XVW DV WKH KDQG FDQ EHFRPH D PRGXOH WR DQ DUP ¿QJHUWLSV FDQ
EHFRPH PRGXOHV RI WKH KDQG 0RGXODU ¿QJHUWLSV KDYH EHHQ
designed by SynTouch, a company started at the University of
6RXWKHUQ&DOLIRUQLDWRKDYHWKHVHQVLWLYLW\RIDKXPDQ¿QJHU

202
 z But there are always trade-offs. For many tasks, the only important
aspect of the human hand is the opposable grip of the thumb and
RWKHU¿QJHUVIRUJUDVSLQJ%D[WHU¶VHQGHIIHFWRURQO\KDVRQHGHJUHH
of freedom, but it might be all you need for picking up different
W\SHVRIREMHFWV

z Moreover, a great thing about robots is that we don’t need to

limit their workings to whatever humans happen to have. A great
example of breaking out of the human box is the vacuum cup
gripper, which uses a vacuum pump to generate a negative pressure
relative to ambient. Factory managers have long known that a
vacuum cup gripper can grasp things that are sometimes really
GLI¿FXOWIRU¿QJHUVWRPDQDJHÀDWVXUIDFHVRUVPRRWKVKDSHV%XW
we haven’t been putting grippers like that on humanoid robots until
much more recently.

Humanoid Legs
z Robotics legs that mimic the gait of a human walk with incredible
biorealistic functioning were introduced in 2012 by researchers at
the University of Arizona. Humans have a neural network in the
lumbar region of the spinal cord that controls and produces rhythmic
neural signals to the muscles responsible for routine walking. The
researchers created a simple robotic version of this system, where
feedback from load sensors on the legs provided information to a
central pattern generator, which varied the frequency of its rhythm
in response.

z 7KH HDUOLHVW KXPDQRLG URERWV UHOLHG RQ ³VWDWLF ZDONLQJ´ ZKLFK

ZDV PRUH RI D VKXIÀH WKDQ D ZDON $ JRRG H[DPSOH RI VWDWLF
walking is Robosapien, one of the most famous humanoid robot
toys. Robosapien, made by the WowWee company, is controlled
remotely and is famous for its autonomous mode, where it
FDQ PRYH DURXQG WKH KRXVH RQ LWV RZQ DYRLGLQJ REMHFWV 7KH
Robosapien class of robots was invented by Mark Tilden, known
for developing robots with solid-state electronics that do not need
an embedded computer.

203
 z The problem with static walking is that it involves a lot of
rocking from side to side. In larger humanoids, this movement
sideways takes up energy better spent moving forward. Rocking
LV HQHUJHWLFDOO\LQHI¿FLHQW$OVR URFNLQJFUHDWHVLQVWDELOLW\WKDWLV
liable to make your robot fall over.

z The key to stable walking is to keep your body’s center of mass in

between the support provided by your feet. To mimic human feet,
we can put pressure sensors on the feet of our humanoids to let
them know when they’re tipping over.

z When humans are actually walking, we use our legs as inverted

SHQGXOXPV)URPWKHWLPHRXUKHHOVWULNHVWRWKHWLPHRXUWRHÀLSV
off the ground, we’re using our leg to pivot our mass up and over
our foot. The key to the dynamics of this motion is to coordinate the
transfer of momentum from one leg to the other, always keeping

The Development of Humanoid Robots

  $%27 WKH ¿UVW IXOOVFDOH KXPDQRLG URERW ZLWK

:
Lecture 23—Humanoid Robots: Just like Us?

systems for vision, conversation, and walking, is

developed at Waseda University in Japan.

  KH %($0 ELRORJ\ HOHFWURQLFV DHVWKHWLFV PHFKDQLFV

7
approach to robotics is started by Mark Tilden.

1997 RoboCup, a soccer match for autonomous robots, has its

¿UVWWRXUQDPHQW

2000 ASIMO, a humanoid robot, is introduced by Honda Motor

Corporation.

2005 Ballbot, a mobile robot that balances on a sphere, is

created by Ralph Hollis.

204
 WKHFHQWHURISUHVVXUHLQVLGHWKHVXSSRUWVRIWKHIHHW(YHQWKRXJK
we think about walking as something we do with our legs, walking
involves the whole body.

z 7KH URERW RIWHQ FRQVLGHUHG WR EH WKH ZRUOG¶V ¿UVW VHOISURSHOOHG
bipedal humanoid is WABOT-1, who stepped into robotic stardom
in 1973. Researchers led by Ichiro Kato at Waseda University
in Tokyo built a series of WABOTs over the years, and by 1984,
:$%27 PRGHO :/5' EHFDPH WKH ¿UVW URERW WR DFKLHYH
dynamic walking.

z Dynamic walking is a balancing act. The robot Atlas, introduced

by Boston Dynamics in 2013, is a great balancer. Atlas is able to
maintain its pose as a stable inverted pendulum by virtue of making
UDSLGDGMXVWPHQWVWRLWVFHQWHURIPDVVDQGLQWXUQWKDWFHQWHURI
pressure on the foot.

z For humans to balance standing up on one leg, or even two legs,

is an active process. We’re constantly using our muscles, sensing
WKH SRVLWLRQ RI RXU MRLQWV DQG WKH WHQVLRQ LQ RXU PXVFOHV ,I ZH
ZHUHWRVWRSGRLQJDOORIWKDWZRUNZHZRXOGMXVWFUXPSOHWRWKH
ground. Biped robots moving like a human will fall down unless
they actively balance.

z One solution for the problem of a statically unstable structure is to

put it in motion. Passive dynamic walkers adhere to this principle.
8VLQJRQO\JUDYLW\DMRLQWHGOLQNDJHV\VWHPZLOOZDONRQLWVRZQLI
LWLVDUUDQJHGMXVWULJKWRQDWLOWHGVXUIDFH3DVVLYHG\QDPLFZDONLQJ
is dynamically stable because one leg catches the falling system. The
system pivots over that leg, and then the other leg catches the system.

z Atlas, the balance master, also has impressive stability when

walking. It can sense changes in its center of pressure on its feet
DQG PDNH UDSLG DGMXVWPHQWV WR LWV FHQWHU RI PDVV$WODV FDQ NHHS
its balance while walking with obstacles put on the path that it has
WRVWHSRYHU(YHQPRUHLPSUHVVLYHLVLWVDELOLW\WRNHHSLWVEDODQFH
while walking across an uneven surface of good-sized rocks.

205
 z $WODV ZDONV LQ D PDQQHU FDOOHG ³*URXFKR ZDONLQJ´ LQ KXPDQ
biomechanics. With a Groucho walk, named after the famous
comedian, you keep your center of mass level, and you never pivot
up and over your planted foot. Your knee never straightens. The
*URXFKRZDONLVYHU\VWDEOH7KHURERWLVVTXDWWLQJDOOWKHWLPHVR
it has a lower center of mass. Also, a Groucho walk allows you to
move with precision that you don’t normally have with an inverted
pendulum walk.

Humanoid Appearance
z Another frontier in humanoid robotics focuses on how the body
looks. Looks turn out to be very important for how we react
emotionally to robots. While the promise of humanoid robotics is
WKDWURERWVFDQ¿WLPPHGLDWHO\LQWRRXUZRUNSODFHVDQGKRPHVZH
won’t want them around unless we feel comfortable with them.

z ,I \RXU DI¿QLW\ IRU D URERW LV ORZ LQGLIIHUHQW RU QHJDWLYH WKDW
eventually means no attention. And no attention means no
communication. Professor Hiroshi Ishiguro of Osaka University
LQ -DSDQ DGGUHVVHV WKLV SUREOHP RI DI¿QLW\ +H LV ZRUNLQJ ZLWK D
company called Kokoro to build highly realistic humanoids that
Lecture 23—Humanoid Robots: Just like Us?

they call Actroids.

z The Actroid model Geminoid HI-2 is built to gaze, blink, turn,

gesture, and interact verbally with people. Geminoid HI-2 is
remotely controlled by a human. When it speaks, it is the voice
of the human in the control loop. Actroids have come in many
different models, and more recent ones, such as Actroid-SIT, have
conversational autonomy.

z In 1970, robotics professor Dr. Masahiro Mori, working at the

Tokyo Institute of Technology, created a hypothesis that captured
what turns out to be a complex relationship between the level of our
DI¿QLW\IRUDURERWDQGKRZKXPDQOLNHLWLV

206
 z Mori proposed that as a robot becomes more like a human in
DSSHDUDQFHRXUDI¿QLW\IRULWJURZV²DW¿UVW+RZHYHUDVVRRQDV
the robot gets very close to human likeness, such as what you might
see in a scary mask, then the appeal of the robot quickly plummets
to creepy. Mori called this small region of disgust and creepiness
the uncanny valley.

z 0RULNQHZWKDWDI¿QLW\ZRXOGPDWWHULIZHZDQWHGKXPDQVWRDFFHSW
robots and be willing to work together with them or to purchase
them. However, some researchers now think that the location of the
uncanny valley might change over time or depend on culture.

‹.RLFKL.DPRVKLGD*HWW\,PDJHV1HZV7KLQNVWRFN

As robots look increasingly similar to humans, our reactions become increasingly

positive—until we reach the point of revulsion, known as the uncanny valley.

The Future of Humanoid Robotics

z What’s next for humanoid robotics? Smooth transitions from
one kind of movement to another (for example, from walking to
climbing stairs, or changing pace, or stopping on a dime) will be
LPSRUWDQW$QRWKHU ZLOO EH DOORZLQJ WKH DUPV WR FDUU\ REMHFWV RU
perform other tasks while the legs continue moving the entire body
in various ways.

207
 z A third area of work will be thinking fast and moving fast. Those
are two related challenges we see being addressed at the annual
RoboCup, the robotic soccer contests that are held around the world.
NAO humanoid robots, made by Aldebaran, the maker of the social
robot Pepper, are a favorite among robotic soccer teams.

z NAO comes with two cameras, an inertial measurement unit, touch

VHQVRUV IRXU GLUHFWLRQDO PLFURSKRQHV DQG WZR VRQDU UDQJH¿QGHUV
When NAO teams play soccer, they are moving autonomously.
They use cameras to detect
the ball and to recognize
their teammates versus the
opposition. Their inertial
measurement unit measures
acceleration so that they
can keep track of their
balance and position.

z Teams of humanoids are

another focus. Working

‹-HQV6FKOXHWHU*HWW\,PDJHV1HZV7KLQNVWRFN
with humanoids that look
Lecture 23—Humanoid Robots: Just like Us?

a lot like Baxter, General

Motors is working on
coordinating the reaching
and grabbing—the robotic
pick and place—of multiple
humanoids. The challenge
is that the robots have to
be able to know where they NAO robots play soccer autonomously.
are and where they need to
be, as well as where others are and are likely to be. Better sensors,
better actuators, and better controllers will all be needed and will
continue to be the foci of research and innovation.

208
 Important Terms

Actroid +XPDQRLG URERW PDQXIDFWXUHG E\ -DSDQHVH URERWLFV FRPSDQ\

.RNRURWKDWLVSDWWHUQHGDIWHUVSHFL¿FKXPDQVDQGEXLOWZLWKKLJKGHWDLOIRU
facial features and gestures.

inertial measurement unit (IMU) $Q LQWHJUDWHG VHQVRU WKDW PHDVXUHV

orientation, velocity, acceleration, and, sometimes, compass heading.

uncanny valley +\SRWKHVL]HG E\ 0DVDKLUR 0RUL D VXGGHQ ORVV RI

DI¿QLW\ WKDW KXPDQV FDQ IHHO DV PDVNV RU URERWV EHFRPH DOPRVW EXW QRW
quite, human.

Suggested Reading

Hornyak, Loving the Machine, chap. 9.

0RUL³7KH8QFDQQ\9DOOH\´
Rosheim, Robot Evolution, chap. 6.

Question to Consider

1. When you think about the tools that Leonardo da Vinci had for building
a humanoid robot, what do you think his greatest challenges would
have been?

209
 The Futures of Robotics
Lecture 24

B
eyond the many interesting robots already in existence, what makes
robotics such an exciting endeavor is that there are so many different
possible futures. This variety of possibilities stems partly from the
fact that robotics is a multidisciplinary collaboration of mechanical and
electrical engineering, computer science and neuroscience, and cognitive
science and biology. The moment we encounter new problems and challenges
will be the moment that we see the necessity to create new kinds of robots to
help us respond. Using the principles of robotics, we learn to understand and
design the robots of our many futures.

Specialists versus Generalists

z 5RERWLFVLVVXFKDGLYHUVH¿HOGWKDWGLVWLQFWURERWRULJLQVDQGURERW
futures can be found in virtually every domain of modern life and
HYHU\ ¿HOG RI VFLHQFH DQG WHFKQRORJ\ :KDW XQLWHV URERWLFV LV D
fundamental trade-off that comes up in the design of nearly any tool
RUPDFKLQHVSHFLDOLVWYHUVXVJHQHUDOLVW

z $ VSHFLDOLVW GRHV RQH RU D IHZ MREV 8VXDOO\ WKH VSHFLDOLVW GRHV
WKHWDVNVYHU\ZHOOYHU\HI¿FLHQWO\RUYHU\FKHDSO\5RRPEDLVDQ
Lecture 24—The Futures of Robotics

H[DPSOHRIDVSHFLDOLVWEXLOWWRYDFXXPÀRRUV

z In contrast, a generalist does many things but usually does no single

task as well as a specialist would. But the value of a generalist is
that a single robot can do many things. Baxter, built by Rethink
Robotics, is a great example of a generalist. Baxter is able to do any
QXPEHURIGLIIHUHQWMREVWKDWLQYROYHSLFNLQJDQGSODFLQJ

z $ JHQHUDOLVW IURP VFLHQFH ¿FWLRQ ZRXOG EH D ZDONLQJ WDONLQJ

humanoid that could work in any of the physical spaces that a
human can and do all the physical tasks that a human could.

210
z In the short term, it’s probably more likely that we’ll be in a world
where the specialist robots dominate. We’ll have robotic vacuum
cleaners, window cleaners, lawn mowers, dog walkers, and cars.
In this future of specialized robots, we won’t even see most of
WKHVH DV URERWV WKH\¶OO MXVW EH RXU DSSOLDQFHV YHKLFOHV DQG
workplace machines.

z As with any tool or

technology, we have
legitimate concerns
about how we choose
to use robots. We can
focus our concerns
on something that we
know how to create
DQG FRQWURO LQ URERWV
autonomy. For every
kind of robot, for
every task, in every
workplace, we can ask

‹DELGDOL6WRFN7KLQNVWRFN
WKLV JHQHUDO TXHVWLRQ
What type and level of
autonomy will allow the
robot to accomplish its
task without harming A humanoid robot that does a variety of
humans? tasks could be considered a generalist.

z Is energy autonomy important because the mission is long term

and refueling is impractical? Think about Wave Gliders swimming
DFURVV WKH 3DFL¿F 2FHDQ DQG KDUYHVWLQJ HQHUJ\ IURP ZDYHV WR
do so. Is behavioral autonomy important because the mission
is remote and humans can’t rapidly communicate to the robot
about how to perform a task? Think about the Curiosity robot
on Mars.

211
 ‹'DYLG3DXO0RUULV*HWW\,PDJHV1HZV7KLQNVWRFN
In the future of specialized robots, we will likely see them not as robots but as
appliances, machines, and vehicles.

Cloud Robotics
z One way to envision how the various dimensions of robot
autonomy will be managed is to focus on what is called cloud
robotics. Just like with cloud computing, where you share
Lecture 24—The Futures of Robotics

applications or storage capacity over the Internet, the idea behind

cloud robotics is to link up individual robots with shared programs
and storage capacity.

z 5RER(DUWK LV D FORXG URERWLFV SURMHFW VWDUWHG E\ WKH (XURSHDQ

Commission in 2009. The idea is to create a collective database
that robots access directly. If a robot encounters a new situation,
LWTXHULHV5RER(DUWKWRVHHLIDQRWKHUURERWKDVDOUHDG\VROYHGWKH
same problem. If a robot learns something new on its own, then it
uploads code for its newly created solution.

212
 z In a cloud robotics future, intelligence is linked to your ability to
communicate with other robots and learn from their experience
without having to learn the task directly yourself.

z 7R PDNH FORXG URERWLFV D UHDOLW\ WKH 5RER(DUWK SURMHFW FUHDWHG

Rapyuta, the working cloud platform. In addition to sharing
solutions with each other, one idea with Rapyuta is that you can
do your computationally expensive processing in the cloud. In this
PRGHO\RXDUHVSOLWWLQJXSWKHZRUNRIWKHFRQWUROOHULQWRWZRSDUWV
Part of your controller is operating in the robot, and part of your
controller is operating on a clone of your system in the cloud.

z Google was awarded a U.S. patent for cloud-based robotics in

7KHQDPHRIWKHSDWHQWLV³6KDUHGURERWNQRZOHGJHEDVHIRU
XVHZLWKFORXGFRPSXWLQJV\VWHP´ZLWK5\DQ+LFNPDQDVWKHOHDG
LQYHQWRU7KHLGHDLVVLPLODUWRWKDWRI5RER(DUWKWRFRQQHFWURERWV
to and through the cloud.

z The fact that Google has been able to patent what may be seen in
the courts as a huge part—or all—of cloud robotics gives them an
advantage over any other company that wants to use the web to
coordinate their robots, have their robots learn from the experience
of others, or share controller space.

Watson
z Cloud robotics is not the only future for increased autonomy in
robotics. Imagine professional and personal settings where you
cannot, or do not want to, expand the abilities of your robot by using
cloud computing or cloud robotics. What might be an innovative
and alternate way forward?

z IBM’s Watson system, an example of what is called cognitive

computing, is one of the most intelligent computer systems on the
market. Watson made headlines in 2011 when it defeated in the
game show Jeopardy! the two best humans to have ever played
that game.

213
 z Jeopardy! Watson was followed by a variety of more practical
applications, including Dr. Watson diagnosing patients, Chef
Watson coming up with new recipes, Finance Watson, and
Customer Service Watson helping callers and offering technical
support advice.

z All these versions of Watson were not a robot but, rather, a computer
program. If Watson had a body, what could we do with Robot
Watson? For example, imagine an emergency situation where you
need to be able to think on your feet. What if communication with
remotely located humans had degraded because you, the robot, were
underground, or electric noise was swamping your transmissions.
What do you do?

z First responders, acting on their own, are trained to assess the

VLWXDWLRQ DW KDQG :KDW LV KDSSHQLQJ ZK\ LV LW KDSSHQLQJ DQG
what, if anything, can be done and in what order should I proceed?

z First Responder Robot Watson walks in and compares the readings

on its cameras, chemical sensors, and hazard detection systems to
data patterns in its database. What is happening? Watson makes an
HGXFDWHGJXHVV1DWXUDOJDVLVEHLQJYHQWHGLQWRWKHURRP:K\LV
it happening? A pipe has broken. What can be done? The cut-off
valve can be closed.
Lecture 24—The Futures of Robotics

z %HFDXVH:DWVRQWKHURERWKDVWKHDELOLW\WRMXGJHWKHFRQ¿GHQFHRI
its answers, and make a decision based on the particular situation
at hand, it recognizes that if in fact the leak is natural gas, then the
situation is extremely volatile—any spark will cause a devastating
explosion. Thus, Watson accepts this natural-gas-leak answer with
D UHODWLYHO\ ORZ FRQ¿GHQFH WKUHVKROG ,W LV EHWWHU WR DVVXPH WKH
leak is natural gas and be wrong than to waste time seeking higher
FRQ¿GHQFH:DWVRQZDONVRYHUWRWKHVKXWRIIYDOYHDQGWXUQVRII
the leaking gas.

214
 z $QRWKHU JUHDW EHQH¿WV RI :DWVRQVW\OH FRJQLWLYH FRPSXWLQJ LV
that a robot can estimate when it does not know enough about the
situation to make a decision. Sometimes it’s better not to guess. For
humans, knowing when not to guess is often the difference between
being a novice and being an expert. A robot Watson would be good
at knowing when to call in other robots.

Modules
z Because the various approaches to robotics all share the same
goal—autonomous behavior—each approach, each discipline, has
something to offer other approaches within robotics. The result
is that we have solutions, principles, and concepts that serve as
modules we can mix and match.

z :KHQ \RX KDYH PRGXOHV \RX FDQ UHFRQ¿JXUH WKHLU SK\VLFDO

connections and reassign their tasks. One of the challenges of a
VZDUPLVWKDWVSHFL¿FMREVKDYHWREHDVVLJQHGDQGHYHQVZDSSHG
dynamically, as the situation develops, and not ahead of time. When
WKLVSUREOHPLVVROYHGWKHZKROHV\VWHPLVDEOHWRDGMXVWDVWKHWDVN
unfolds, circumstances change, or some robots fail.

z This is the power of modules. Imagine giving Roomba a module

to handle extreme conditions. Or what about a Roomba module
for swarm robotics, allowing a team of Roombas to coordinate
how they clean a very large space? Or what about a social robotics
module for the Roomba, making it more interactive, thereby easier
and more fun to take care of? Modules and modular robots expand
the limits of what we think of as a robot.

Inspired by Nature and Art

z Where do wonderfully crazy ideas like modular robots and
biohybrid robots come from? How might we think of new kinds
of bodies, sensors, controllers, and actuators—or new modes of
behavior and cognition, action and perception?

215
 z History shows that roboticists are inspired in many different
ZD\V %XW WZR SDWKV WR WKH IXWXUH VWDQG RXW 7KH ¿UVW VRXUFH RI
inspiration, from behavior-based robotics to biohybrids, is nature,
especially biology.

z The bioinspired future is a wet one. This future involves an even

more fundamental shift away from the classic approaches of
mechanical and electrical engineering to the bioinspired approaches
of tissue engineering and synthetic biology that are helping us
create biohybrid systems.

z Classically, we build robots out of dry materials, such as metal,

rubber, and plastic. But when we look at animals, they build
themselves out of wet materials, such as muscle and bone. A
biohybrid system contains materials of both types in order to get the
best of both worlds.

z 2QHRIWKH¿UVWRIELRK\EULGURERWVLV0HGXVRLGDURERWLFMHOO\¿VK
created by Janna Nawroth and her colleagues at Caltech and
Harvard. Medusoid’s body is built out of silicon rubber. On top of
that body Nawroth deposited proteins. She used those proteins as a
scaffold upon which to grow cardiac cells from the heart of a rat.

z After the cells grew and connected, they put Medusoid into a tank
Lecture 24—The Futures of Robotics

and then pulsed electric charges in the water. The cardiac cells
UHVSRQGHGE\FRQWUDFWLQJMXVWOLNHWKH\ZRXOGLQDKHDUWVWLPXODWHG
E\DQHPHUJHQF\GH¿EULOODWRU%HFDXVHRIWKHSDWWHUQLQZKLFKWKH\
had arranged the protein upon which the cells grew, the contractions
DFWXDWHGWKHERG\LQDZD\WKDWDOORZHGWKHZKROHURERWLFMHOO\¿VK
to swim.

z Medusoid is a great, simple proof of concept that we can begin to

grow parts of robots. Because cells are wet and need to be kept wet
in order to function, the key to the growth of wet robotics will be
either to work in a wet medium or to build bodies that contain that
wet medium inside of their bodies.

216
 z The promise of biohybrid systems like Medusoid is that with the
growth of tissue, such as muscle, we expand our toolbox of solutions
for robots. For example, animals are great at building extracellular
tissue, such as tendon and ligament, that have mechanical properties
that are tuned to the motion and load of the system in which they
are operating. In addition, a system that can grow itself should also
EHDEOHWRKHDOLWVHOI7KLVZRXOGEHDWHUUL¿FDELOLW\IRUDQ\URERW
on a remote mission.

z $ ¿QDO SDWK WR WKH IXWXUH DV WKH EHVW ZULWHUV DQG PRYLHV PDNH
clear, is art. As a society, we sometimes get into a kind of team
FRPSHWLWLRQVFLHQFHYHUVXVDUW%XWVRPHRIWKHJUHDWHVWVFLHQWLVWV
were on both teams, such as Leonardo da Vinci, who also was
DUJXDEO\WKH¿UVWURERWLFLVW0RUHRYHUWKHWHDPVRIDUWDQGVFLHQFH
do different and complementary things.

z Art can imagine the impossible and the highly improbable. But
art can also offer scenarios for what is possible or even likely.
&RQVLGHUVFLHQFH¿FWLRQIRUH[DPSOHDVSURWRW\SLQJWKHIXWXUHLQ
our imaginations.

Important Terms

cloud robotics$VXEGLVFLSOLQHLQZKLFKURERWVDUHGHVLJQHGWRLQWHUDFWZLWK
databases via computer networks and then with each other.

cognitive computing $ QHZ VW\OH RI SURJUDPPLQJ SLRQHHUHG E\

IBM’s Watson that aims to simulate human cognitive processes, such as
understanding natural language, evaluating context, and assessing the
reliability of its knowledge.

217
 Suggested Reading

Hickman, Kuffner, Bruce, Gbarpure, Kohler, Poursohi, Francis, and Lewis,

2014, Shared robot knowledge base for use with cloud computing system,
U.S. Patent 8,639,644 B1.
YDQGH0ROHQJUDIW³)LQDO3URMHFW5HSRUW´
Nourbakhsh, Robot Futures, chap. 6.

Other Resources

The New York Times ³6KRRWLQJ IRU D 0RRQ )LOOHG ZLWK 5RERWV´
ZZZQ\WLPHVFRPLQWHUDFWLYHWHFKQRORJ\JRRJOHQHZ
JHQHUDWLRQURERWVYLGHRVKWPO"UHI WHFKQRORJ\ 7KLV ZHEEDVHG DUWLFOH
ORRNVDWWKH¿UVWZDYHRIURERWVDFTXLUHGE\*RRJOH0DNHVXUHWRUHDGWKH
UHODWHGDUWLFOHDWZZZQ\WLPHVFRPWHFKQRORJ\JRRJOHDGGVWR
LWVPHQDJHULHRIURERWVKWPO ZKHUH \RX¶OO ¿QG RXW ZK\ WKLV ORRNHG DW WKH
WLPH OLNH D ³PRRQVKRW´ DQ DOORXW HIIRUW RI WLPH DQG UHVRXUFHV WR DFKLHYH
a goal.

Questions to Consider

1. As we’ve seen with embedded microprocessors becoming part of every

appliance, we are likely to see embedded robotic systems become much
Lecture 24—The Futures of Robotics

more common. Which technological sectors are ripe for a dramatic

increase in embedded robotic systems?

a. Transportation of goods and people.

b. Assembly and manufacturing.

c. Medical devices used on and in the bodies of humans and their pets.

d. Agriculture.

218
 e. Retail services.

f. Personal and home health care.

g. Military and defense systems.

h. First-responder services.

2. Put on your entrepreneurial hat. If you wanted to assemble a team to

design, engineer, and produce a new commercial robot, how would you
proceed to search for ideas that you could pitch to venture capitalists?

a. Look for a profession that is dominated by humans doing manual

labor (see previous question) or by humans providing social
interactions and information transfer.

b. Look for any task that is done by humans or for which humans use
a tool or machine.

F /RRNDWIRUPVRIHQWHUWDLQPHQWWKDWKXPDQVHQMR\

d. Look for any tool or appliance that does not yet possess an
embedded robotic system.

e. Look at military and defense systems.

f. Look to nature for life-forms that behave as our machines do not.

219
 Timeline

~1495 Leonardo da Vinci designs a mechanical knight using

pulleys and cable, as well as a mechanical cart capable of
serving as a programmable controller.

1739 A duck automaton by Jacques de Vaucanson draws crowds

to see fake mechanical digestion.

1788 Governor, an automatic speed control, is invented for

steam engines by James Watt and Matthew Boulton.

 7KH WRUSHGR WKH ¿UVW DXWRQRPRXV XQGHUZDWHU YHKLFOH

(AUV), is invented by Robert Whitehead.

1872 Erewhon, Samuel Butler’s novel, develops the idea that

machines might evolve consciousness.

 7KHUHPRWHFRQWUROLV¿UVWSDWHQWHGDQGGHPRQVWUDWHGRQD
model ship by Nikola Tesla.

 7LN7RND¿FWLRQDOPHFKDQLFDOPDQZLWKZLQGXSNH\VIRU
thought, movement, and speech, appears in Ozma of Oz by
L. Frank Baum.

 $QDHULDOWRUSHGRWKH¿UVWXQPDQQHGDHULDOV\VWHP8$6

LV LQYHQWHG E\ (OPHU 6SHUU\ XVLQJ D PRGL¿HG &XUWLVV
N-9 seaplane.

1920 R.U.R. (Rossum’s Universal Robots), a play published by

.DUHOýDSHNSRSXODUL]HVWKHZRUG³URERW´

 0DULD WKH ¿UVW URERW WR DSSHDU LQ D PRYLH VWDUV LQ WKH
Timeline

silent classic Metropolis.

220
1938 An industrial programmable robot, introduced by
*ULI¿WK37D\ORULQ³$Q$XWRPDWLF%ORFN6HWWLQJ&UDQH´
Meccano Magazine LV WKH ¿UVW H[DPSOH ¿WWLQJ ODWHU
FULWHULDRILQWHUQDWLRQDOVWDQGDUG,62

1941 The word roboticsLV¿UVWXVHGLQWKHVKRUWVWRU\³/LDU´E\

writer Isaac Asimov.

1942 The three laws of robotics are created by Isaac Asimov in

KLVVKRUWVWRU\³5XQDURXQG´

1944 V-1 bombers, unmanned aircraft (UA), are deployed by

WKH*HUPDQDLUIRUFHDJDLQVWWDUJHWVLQ(QJODQG

   URFNHWV WKH ¿UVW ORQJUDQJH EDOOLVWLF PLVVLOHV DUH

9
deployed by the German military against urban targets
LQ(QJODQG

±  OVLHDQG(OPHUWKH¿UVWELRORJLFDOO\LQVSLUHGDXWRQRPRXV
(
robots, are built by neuroscientist Gray Walter using
analog electronics.

1950 I, Robot, a collection of robotics short stories by Isaac

Asimov, is published.

1950 The Turing test, the so-called imitation game, is proposed

by Alan Turing as an operational test of intelligence
for computers.

1956 $UWL¿FLDOLQWHOOLJHQFH is coined and launched as a research

area by John McCarthy and others at a conference at
Dartmouth College.

  QLPDWLRQWKHZRUOG¶V¿UVWURERWFRPSDQ\LVIRXQGHGE\
8
*HRUJH'HYRODQG-RVHSK(QJHOEHUJHU

221
   XPHULFDOO\FRQWUROOHG1&PDFKLQHVDUH¿UVWSURGXFHG
1
by FANUC in Japan.

  XWRSLORWVLQFDUVEHFRPHWKH¿UVWFRPPHUFLDOO\DYDLODEOH
$
speed control on land.

 ³ 3URJUDPPHG $UWLFOH 7UDQVIHU´ 86 3DWHQW 

key intellectual property for Unimation, is issued to
George C. Devol, Jr.

  QLPDWHWKH¿UVWLQGXVWULDOURERWIURP8QLPDWLRQJRHVWR
8
work unloading die casts at the General Motors factory in
Trenton, NJ.

± 6
 KDNH\ WKH ¿UVW DXWRQRPRXV PRELOH URERW ZLWK GLJLWDO
electronic computer, is built by Stanford Research Institute.

1968–1972 Zero-moment point analysis for motion planning in bipedal

URERWVLVGHYHORSHGLQSDSHUVE\0LRPLU9XNREUDWRYLü

1970 Lunokhod 1, a Soviet remote-controlled lunar rover,

EHFRPHVWKH¿UVWXQPDQQHGYHKLFOHWRH[SORUHWKHVXUIDFH
of an extraterrestrial planetary body.

  KH 0RZERW SDWHQW IRU WKH ¿UVW URERWLF ODZQ PRZHU LV
7
issued to S. Lawrence Bellinger.

  DPXOXV WKH ¿UVW LQGXVWULDO URERW ZLWK VL[

)
electromechanically controlled axes, is introduced by
German robotic company KUKA.

  $%27 WKH ¿UVW IXOOVFDOH KXPDQRLG URERW ZLWK

:
systems for vision, conversation, and walking, is
developed at Waseda University in Japan.
Timeline

222
±  WKH¿UVWFRPPHUFLDOO\DYDLODEOHURERWDUPFRQWUROOHG
7
by a minicomputer, is designed by Richard Hohn and
released by Cincinnati Milacron.

1976 Robot arms are used on Viking 1 and 2 space probes.

1979 The Stanford Cart, designed by Hans Moravec,

autonomously navigates a room full of obstacles.

  REHUW:LOOLDPVEHFRPHVWKH¿UVWKXPDQNLOOHGE\DURERW
5
while working in a Ford Motor casting plant.

1981 SCARA industrial robots are introduced in Japan.

  ,77 D ¿FWLRQDO DXWRQRPRXV WDONLQJ FDU DSSHDUV LQ WKH

.
television series Knight Rider.

  LFWLRQDO ³UHSOLFDQWV´ DSSHDU LQ Blade Runner, a science

)
¿FWLRQ ¿OP DGDSWHG IURP 3KLOLS . 'LFN¶V Do Androids
Dream of Electric Sheep? (1968).

1984 Vehicles, a book by Otto Braitenberg lays the groundwork

for the design of the simplest-possible autonomous robots.

1985 FANUC develops robots to assemble other robots.

1986 A fully autonomous robot car begins test-drives on empty

streets in Germany.

1986 Genghis, a behaviorally autonomous hexapod robot, is

introduced by Rodney Brooks, Colin Angle, and their
team at MIT.

1987 Star Trek: The Next Generation, a television show, debuts

with android lieutenant commander Data.

223
 ± +
 HOS0DWH WKH ¿UVW ZRUNLQJ KRVSLWDO FRXULHU URERW LV
developed and debugged at Danbury Hospital by Joseph
(QJHOEHUJHU¶V VHFRQG URERWLFV FRPSDQ\ 7UDQVLWLRQV
Research Corporation (renamed HelpMate Robotics
in 1997).

1990 The iRobot Corporation is founded by Rodney Brooks,

Colin Angle, and Helen Grenier.

  KH %($0 ELRORJ\ HOHFWURQLFV DHVWKHWLFV PHFKDQLFV

7
approach to robotics is started by Mark Tilden.

  D03DQG9,7$GULYHUOHVVFDUVRSHUDWHVDIHO\LQWUDI¿F
9
for more than 600 miles in Germany.

  2720$1 RIIHUV WKH ZRUOG¶V ¿UVW FRQWUROOHU IRU

0
synchronizing two robots.

  KH543UHGDWRUDQXQPDQQHGDHULDOYHKLFOH8$9EXLOW

7
by General Atomics Aeronautical Systems, is deployed
E\ D 86 1DY\$UP\ WHDP LQ WKH IRUPHU <XJRVODYLD
for reconnaissance.

1995 Adaptive cruise control for cars, using lasers, is introduced

by Mitsubishi.

  RMRXUQHU WKH VHPLDXWRQRPRXV URYHU RI 1$6$¶V

6
3DWK¿QGHU PLVVLRQ EHFRPHV WKH ¿UVW URERW WR ODQG DQG
work on Mars.

1997 RoboCup, a soccer match for autonomous robots, has its

¿UVWWRXUQDPHQW

1997 Deep Blue, IBM’s computer, defeats world chess

Timeline

champion Gary Kasparov.

224
  KH /(*2 0LQGVWRUPV URERWLFV NLW LV LQWURGXFHG EDVHG
7
on MIT Media Lab’s programmable brick technology.

1999 AIBO, the robotic dog, introduced by Sony, sells more

than 150,000 units before ceasing production in 2006.

2000 Kismet, a robotic head built to interact socially with

humans, is created by Cynthia Breazeal at MIT.

2000 The da Vinci Surgical System, a robot by Intuitive

Surgical allowing surgeons to be remotely present and
operate inside patients, is approved by the FDA.

2000 ASIMO, a humanoid robot, is introduced by Honda Motor

Corporation.

2002 Roomba, an animal-inspired robotic vacuum cleaner, is

introduced commercially by iRobot.

2004 Spirit and Opportunity, semiautonomous NASA rovers

and geological explorers, land on Mars.

2005 Stanley, a robotic car built by a team from Stanford and

Volkswagen, wins the DARPA Grand Challenge by being
RQHRI¿YHYHKLFOHVWRDXWRQRPRXVO\QDYLJDWHDPLOH
off-road course through the desert.

2005 Ballbot, a mobile robot that balances on a sphere, is

created by Ralph Hollis.

2005 Arduino, an inexpensive open-source microcontroller, is

created by Interactive Design Institute in Ivrea, Italy.

  HFRQGJHQHUDWLRQ /(*2 0LQGVWRUPV 1;7 URERWLFV NLW

6
is released.

225
   RVV D PRGL¿HG &KHY\7DKRH EXLOW E\ &DUQHJLH 0HOORQ
%
and General Motors, wins the DARPA Urban Challenge,
QDYLJDWLQJDPLOHFRXUVHZKLOHREH\LQJDOOWUDI¿FODZV
of California, including avoiding pedestrians.

  UHGDWRU 8$9V ¿UH PLVVLOHV IRU WKH ¿UVW WLPH LQ 86
3
campaigns, in Iraq and Afghanistan.

  8 DOVR NQRZQ DV 6FDUOHW .QLJKW EHFRPHV WKH ¿UVW
5
$89 WR FURVV WKH $WODQWLF 2FHDQ XQGHUZDWHU D MRXUQH\
that took 221 days.

2012 The Curiosity rover, a robotic geological explorer, lands

on Mars.

  HYDGD DQG )ORULGD EHFRPH WKH ¿UVW 86 VWDWHV WR
1
permit testing of autonomous vehicles on ordinary roads;
Michigan and California follow in 2013.

  KH :DYH *OLGHU URERW %HQMDPLQ IURP /LTXLG 5RERWLFV

7
sets a world record of 7,939 nautical miles, the longest
MRXUQH\E\DQDXWRQRPRXVVXUIDFHYHVVHO

 7KH'(.$SURVWKHWLFGHYLFHFRQWUROOHGE\HOHFWURP\RJUDP
(0*VLJQDOVIURPPXVFOHVLVDSSURYHGIRUDPSXWHHVE\
the FDA.

2015 The DARPA Robotics Challenge focuses on robots for

disaster response.

 - LER DQG 3HSSHU WKH ¿UVW FRPPHUFLDOO\ DYDLODEOH VRFLDO

URERWV DUH RIIHUHG E\ -LER DQG 6RIW%DQN$OGHEDUDQ
Robotics, respectively.
Timeline

226
 Glossary

accelerometer $ VHQVRU WKDW FRQYHUWV FKDQJHV LQ YHORFLW\ LQWR HOHFWULF

signals that can be read by the controller.

acoustic sensor$OVRNQRZQDVDPLFURSKRQHDVHQVRUWKDWFRQYHUWVVRXQG
waves into electric signals that can be read by the controller.

Actroid +XPDQRLG URERW PDQXIDFWXUHG E\ -DSDQHVH URERWLFV FRPSDQ\

.RNRURWKDWLVSDWWHUQHGDIWHUVSHFL¿FKXPDQVDQGEXLOWZLWKKLJKGHWDLOIRU
facial features and gestures.

actuator 7KH PRYLQJ SDUWV RI D URERW WKDW DOORZ LW WR DFW DQ\ SDUW
appendage, or mechanical system that uses motors to move a robot or
manipulate the world through movement.

android 8VHGPRVWO\LQ¿FWLRQWRUHIHUWRDKXPDQOLNHURERW6HHhumanoid.

animatronics Robotic devices used to simulate animals or humans for

education and entertainment.

articulated robot$PDQLSXODWRURUDUPWKDWKDVKLQJHUHYROXWHMRLQWV

Arduino $Q RSHQVRXUFH KDUGZDUH DQG VRIWZDUH FRPSDQ\ NQRZQ IRU

microcontrollers.

ASIMO7\SHRIKXPDQRLGURERWHPSKDVL]LQJELSHGDOPRELOLW\RYHUXQHYHQ
terrain; in development since 2000 by Honda.

automaton $ PHFKDQLFDO PDFKLQH VKDSHG OLNH D KXPDQ RU DQLPDO WKDW

works automatically without feedback from sensors or direct control
by humans.

Baxter$WZRDUPHGURERWLFPDQLSXODWRULQWURGXFHGE\5HWKLQN5RERWLFVLQ
2012 for light manufacturing tasks, featuring rapid reprogramming and safe
interactions with humans.

227
 BEAM robotics &UHDWHG E\ 0DUN7LOGHQ DQ DSSURDFK WR EXLOGLQJ URERWV
that uses analog electric circuits instead of computers for controllers.

behavior-based robotics'HVLJQRIURERWVWKDWHOLPLQDWHVRUPLQLPL]HVWKH
XVHRILQWHUQDOZRUOGPRGHOVDQGPD[LPL]HVWKHXVHRIUHÀH[OLNHVHQVHDFW
modules.

biomechatronics$SSOLHGVFLHQFHWKDWFRPELQHVPHFKDQLFDODQGELRORJLFDO
parts into a single device.

biomorph$Q\URERWPRGHOHGDIWHUDOLIHIRUP

camber 7KH PHDVXUH RI WKH IURQWWREDFN FXUYDWXUH RI D ZLQJ WKDW KHOSV
create lift.

Cartesian coordinate robot$QLQGXVWULDOURERWZLWKWKUHHSHUSHQGLFXODU

translational degrees of freedom.

caster wheel 7\SH RI XQSRZHUHG ZKHHO PRXQWHG WR D URERW WR SURYLGH
stability and maneuverability.

chassis7KHSULPDU\VWUXFWXUDOVXSSRUWV\VWHPRIDURERW¶VERG\

cloud robotics$VXEGLVFLSOLQHLQZKLFKURERWVDUHGHVLJQHGWRLQWHUDFWZLWK
databases via computer networks and then with each other.

cognitive computing $ QHZ VW\OH RI SURJUDPPLQJ SLRQHHUHG E\

IBM’s Watson that aims to simulate human cognitive processes, such as
understanding natural language, evaluating context, and assessing the
reliability of its knowledge.

compliant actuators0RWRUVDQGOLQNDJHVPDGHRIÀH[LEOHVRIWPDWHULDOV

controller7KHPHFKDQLFDOHOHFWURQLFRUFRPSXWHUL]HGSDUWRIDURERWWKDW
Glossary

converts information provided by sensors into instructions—whether to

DGMXVWDFWXDWRUEHKDYLRURURYHUDOOPRWLRQXSGDWHPDSVDQGRWKHULQWHUQDO
models, or provide information for human operators.

228
CTD sensor 8VHG E\ XQGHUZDWHU URERWV WR PHDVXUH FRQGXFWLYLW\
temperature, and depth.

cybernetics7KHVWXG\RIKRZG\QDPLFV\VWHPVDUHUHJXODWHGIRFXVLQJRQ
issues of feedback, control, and communication.

cyborg6KRUWIRU³F\EHUQHWLFRUJDQLVP´WHUPXVHGPRVWO\LQ¿FWLRQWRUHIHU
to a hybrid of biological and nonbiological parts.

direct current (DC)$W\SHRIHOHFWULFFXUUHQWWKDWLVXQLGLUHFWLRQDO

DC brushed motor$Q HOHFWULF PRWRU WKDW XVHV D PHFKDQLFDO FRQQHFWLRQ

between the rotor and the stator to power and reverse the electromagnetic
¿HOGWKDWGULYHVWKHURWRU

dead reckoning $ PHWKRG RI QDYLJDWLRQ LQ ZKLFK FXUUHQW SRVLWLRQ LV
calculated based on speed, heading, and time from last known location.

degrees of freedom 7KHQXPEHURILQGHSHQGHQWPRWLRQVDYDLODEOHLQDMRLQW

structure, or robot, whether the motion is linear (prismatic, translational),
DQJXODUUHYROXWHURWDWLRQDORUVSKHULFDOOLNHDEDOOMRLQW

deliberative control$URERWLFFRQWURODUFKLWHFWXUHWKDWHPSKDVL]HVSODQQLQJ
as part of the ongoing operations of the robot.

differential drive$QDFWXDWRUGHVLJQLQZKLFKDURERWLVVWHHUHGE\FUHDWLQJ
a difference in the speeds of two independently driven wheels or propellers.

drivability map$SODQQLQJPRGHOWKDWLVFRQWLQXDOO\XSGDWHGDQGXVHGWR
plot the immediate course for a robot.

drone $Q\ XQPDQQHG DHULDO YHKLFOH 8$9 HVSHFLDOO\ RQH WKDW FDQ À\
autonomously (using GPS or other navigational data) and beyond the line of
sight needed for radio-controlled (RC) aircraft.

ducted thruster $ SURSHOOHUEDVHG DFWXDWRU XVHG LQ WRUSHGRVKDSHG

underwater robots to both propel and steer the vehicle.

229
 dynamic soaring$ W\SH RI JOLGLQJ ÀLJKW LQ ZKLFK WKH DQLPDO RU YHKLFOH
gains velocity and height by harvesting energy from steep wind gradients
located near surfaces.

echolocation$FWLYHVHQVLQJLQZKLFKDQDQLPDORUURERWEURDGFDVWVDVRXQG
senses the returned echo, and uses the difference between the signal and its
echo to calculate the distance and composition of the target. See sonar.

end effector$WRRORURWKHUGLVWDOHOHPHQWRQDURERWLFPDQLSXODWRUWKHSDUW
of an actuator that interacts directly with the world.

¿QVKHDUDFWXDWRU)6$$ELRORJLFDOO\LQVSLUHGGHVLJQWKDWFRQYHUWVVKHDU

VWUDLQIURPODWHUDOVWUHVVLQWKHOD\HUVRIWKH¿QLQWREHQGLQJ

ÀDVK PHPRU\ (UDVDEOH UHSURJUDPPDEOH FRPSXWHU PHPRU\ GHYHORSHG LQ

the early 1980s that is relatively fast, durable, and cheap.

À\ZKHHO$URWDWLQJPHFKDQLFDOGHYLFHXVHGWRVWRUHHQHUJ\WKURXJKURWDWLRQDO
inertia or, when its center of mass is located off-axis, create vibrations.

gantry robot/DUJHW\SHRI&DUWHVLDQFRRUGLQDWHPDQLSXODWRUURERWZKRVH
degrees of freedom are in translation and at right angles to each other.

global positioning system (GPS)1DYLJDWLRQDOV\VWHPXVLQJVDWHOOLWHVWKDW

locate a receiver in coordinates of the world geodetic system.

gyroscope 0HFKDQLFDO RU HOHFWURQLF URWDWLRQ VHQVRU WKDW LV XVHG IRU
maintaining or measuring orientation; often linked with an accelerometer for
more precise location in three-dimensional space.

Hall effect sensor$VHQVRUWKDWFRQYHUWVFKDQJHVLQDQH[WHUQDOPDJQHWLF

¿HOGLQWRFKDQJHVLQDYROWDJHRURWKHUHOHFWULFRXWSXW

homing 1DYLJDWLQJ WR D SUHYLRXV ORFDWLRQ XVXDOO\ WKH ORFDWLRQ ZKHUH D

Glossary

robot started.

230
humanoid A robot designed to look and function like a human. See
Actroid, android, and cyborg.

hydraulics Using pressurized liquid to create mechanical motion.

integrated development environment (IDE) 6RIWZDUH LQWHUIDFH WR

facilitate programming in a particular language.

inertial measurement unit (IMU) $Q LQWHJUDWHG VHQVRU WKDW PHDVXUHV

orientation, velocity, acceleration, and, sometimes, compass heading.

inertial navigation'HDGUHFNRQLQJXVLQJDFFHOHURPHWHUVDQGJ\URVFRSHV

kinematics 7KH VWXG\ RI PRWLRQ ZLWKRXW UHJDUG WR WKH IRUFHV JHQHUDWLQJ
them.

Kismet $XWRQRPRXV URERWLF KHDG FDSDEOH RI VRFLDO DQG HPRWLRQDO

interactions with humans that was built by Cynthia Breazeal in the late 1990s.

light-emitting diode (LED)6HPLFRQGXFWRUEDVHGOLJKWVRXUFH

Lunokhod 1 7KH ¿UVW URERWLF SODQHWDU\ URYHU GHSOR\HG RQ WKH 0RRQ LQ
1970 by the Soviet Union.

machine learning&RPSXWHUSURJUDPVZULWWHQWRPDNHDGMXVWPHQWVWRWKHLU
code, with or without direct feedback from a human, in order to improve
performance of the code itself or the robot that the code controls.

manipulator $ URERWLF DUP WKDW JUDVSV DQG PRYHV REMHFWV DOVR DQ\
stationary robot that has one or more such arms.

mechanoid$W\SHRIURERWEXLOWZLWKRXWLQVSLUDWLRQIURPELRORJ\

model-based robotics'HVLJQRIURERWVWKDWPD[LPL]HVWKHXVHRILQWHUQDO
world models, ongoing and sophisticated planning algorithms, and complex
goals and tasks.

231
 Moravec’s paradox ,QVLJKW IURP DUWL¿FLDO LQWHOOLJHQFH RI WKH V WKDW
WKH GLI¿FXOW SUREOHPV LQ URERWLFV DUH WKH HDV\ SUREOHPV IRU DQLPDOV DQG
humans—namely, sophisticated and skillful movements.

motherboard7KHPDLQDQGODUJHVWSULQWHGFLUFXLWERDUGLQDFRPSXWHU

motor6HH'&EUXVKHGPRWRUVHUYRPRWRUDQGVWHSSHUPRWRU

multimeter $Q LQVWUXPHQW XVHG WR PHDVXUH HOHFWULF YROWDJH UHVLVWDQFH

continuity, and current.

muscle wire$VKDSHPHPRU\DOOR\XVHGDVDOLQHDURUVSLUDOPRWRUIRUWKLQ
or distributed actuators.

QDYLJDWLRQDO¿[/RFDWLRQIRXQGXVLQJH[WHUQDOUHIHUHQFHSRLQWV

negative feedback control Regulatory process to maintain stable output

of the robot using the difference between the desired output and the
actual output.

neuroprosthetic7\SH RI URERWLF PDQLSXODWRU XVHG DV D UHSODFHPHQW OLPE

for human amputees and controlled by muscular or neural signals.

nickel-metal hydride battery$FRPPRQW\SHRIUHFKDUJHDEOHEDWWHU\

OpenSCAD)UHHVRIWZDUHIRUGHVLJQLQJWKUHHGLPHQVLRQDOREMHFWV

oscilloscope $Q LQVWUXPHQW XVHG WR PHDVXUH WKH G\QDPLF SURSHUWLHV RI
electric signals.

photoresistor7\SH RI VHQVRU WKDW FRQYHUWV FKDQJHV LQ OLJKW LQWHQVLW\ LQWR
changes in electric resistance.

pick and place )XQGDPHQWDO WDVN IRU URERWLF PDQLSXODWRUV LQYROYLQJ

Glossary

JUDVSLQJPRYLQJDQGWKHQUHOHDVLQJDQREMHFW

232
piezoelectric crystal7\SHRIPDWHULDOXVHGDVDVHQVRURUDQDFWXDWRUWKDW
transduces a change of shape into a voltage or a change in voltage into a
change in shape.

ping sensor An ultrasonic UDQJH¿QGHU

pneumatics Using pressurized gas to create mechanical motion.

potentiometer7\SHRIVHQVRUWKDWFRQYHUWVDPHFKDQLFDOURWDWLRQLQWRDQ
electric change in resistance; also known as a variable resistor.

probabilistic robotics$IUDPHZRUNIRUWKHGHVLJQRIURERWVWKDWH[SOLFLWO\
models the uncertainty inherent in the signals provided by sensors, the
models created by controllers, and the movement instantiated by actuators.

programmed article transfer &RPSXWHUFRQWUROOHG SLFNDQGSODFH WDVN

SHUIRUPHGE\WKH¿UVWFRPPHUFLDOO\VXFFHVVIXOURERWLFPDQLSXODWRU8QLPDWH

proprioception ,QWHUQDO VHQVLQJ RI WKH PRWLRQ DQG IRUFH RI WKH URERW¶V
MRLQWVDQGPRWRUV

prosody 3URSHUWLHV RI VSHHFK RWKHU WKDQ WKH ZRUGV RIWHQ FRQYH\LQJ
emotional content.

pulse-width modulation (PWM) 0HWKRG RI HQFRGLQJ LQIRUPDWLRQ LQ DQ

electric signal by varying the duration of square-wave pulses.

quadcopter (quadricopter)$URWDU\ZLQJHGDLUFUDIWZLWKIRXUSURSHOOHUV

rack and pinion$W\SHRIDFWXDWRUWKDWFRQYHUWVDPRWRU¶VURWDWLRQDOPRWLRQ

to a translational output.

UDQJH ¿QGHU $Q DFWLYH VHQVRU WKDW EURDGFDVWV XOWUDVRQLF RU RWKHU VRXQG
waves and then uses the time it takes for the echo to return to measure the
GLVWDQFHWRWKHUHÀHFWLQJVXUIDFH6HHsonar.

233
 robot A type of machine that can be remote controlled, partially
DXWRQRPRXV RU IXOO\ DXWRQRPRXV DV LW PRYHV LWVHOI RU REMHFWV LQ RUGHU WR
carry out tasks. While robots always have controllers and actuators, remote-
controlled robots may lack onboard sensors.

robotics 7KH ¿HOG RI VWXG\ DQG LQTXLU\ WKDW GHYHORSV SULQFLSOHV DQG
approaches for the design, fabrication, operation, and control of robots.

robot operating system (ROS)$QRSHQVRXUFHVRIWZDUHHQYLURQPHQWIRU

professional roboticists that integrates different languages, libraries, and
solutions.

Roomba7KH ÀRRUFOHDQLQJ URERW LQWURGXFHG LQ  E\ L5RERW WKH ¿UVW
fully autonomous home robot to achieve commercial success.

rotor7KHSDUWRID'&PRWRUWKDWVSLQV

sensor $Q\ GHYLFH WKDW GHWHFWV FKDQJHV LQ SK\VLFDO SURSHUWLHV RU HQHUJ\
patterns in the world or the robot and converts those into electric, chemical,
RU PHFKDQLFDO VLJQDOV XVDEOH E\ D FRQWUROOHU WR DGMXVW DFWXDWRU EHKDYLRU
and overall motion, update maps and other internal models, or provide
information for human operators.

sensor fusion7KHSURFHVVRIFRPELQLQJLQIRUPDWLRQIURPPXOWLSOHVHQVRUV
to create information that is not available from individual sensors alone.

sensorimotor circuits3DUWVRIDURERWRUDQLPDO
that link the information provided by sensors to
the motion produced by actuators.

service robot Any robot built to assist humans,

‹0DWFL6WRFN7KLQNVWRFN

excluding robots involved in manufacturing.

servomotor $Q HOHFWULF '& PRWRU WKDW XVHV

Glossary

sensory feedback from an internal potentiometer

to precisely control position and movement.

234
shaft encoder$VHQVRUWKDWFRQYHUWVWKHURWDWLRQVRIDZKHHOLQWRDQHOHFWULF
signal that represents the speed of the wheel’s rotation.

Shakey7KH¿UVWPRELOHDXWRQRPRXVURERWFRQWUROOHGE\DUHSURJUDPPDEOH
digital electronic computer.

simultaneous localization and mapping (SLAM) 6XE¿HOG RI QDYLJDWLRQ

in which the robot uses sensors and communication systems to know its
position as it creates a map of its surroundings.

Slocum glider An underwater robot that cycles its buoyancy from negative
to positive and converts the resulting vertical motion into horizontal
propulsion.

solder Metal alloys with low melting points used to fuse together separate
elements and, in electronics, create secure electric connections.

sonar6WDQGVIRUVRXQGQDYLJDWLRQDQGUDQJLQJW\SLFDOO\DQDFWLYHVHQVRU
system that broadcasts sound waves and then uses the pattern of the returning
HFKRWRPHDVXUHWKHGLVWDQFHDQGVL]HRIREMHFWV

spectrometer$Q\RIDODUJHFODVVRILQVWUXPHQWVGHVLJQHGWRPHDVXUHKRZ
the intensity of a physical property varies across a range of frequencies,
energies, or masses.

square-cube law 0DWKHPDWLFDO SULQFLSOH WKDW GHVFULEHV WKH UDWLR RI WKH
VXUIDFHDUHDWRWKHYROXPHRIDFODVVRIVLPLODUO\VKDSHGREMHFWVWKDWGLIIHU
in size alone.

Stanley $ IXOO\ DXWRQRPRXV FDU WKDW ZRQ WKH  '$53$ *UDQG
Challenge, created by Stanford Racing Team.

stator7KHVWDWLRQDU\SDUWRIDQHOHFWULFPRWRU

stepper motor$EUXVKOHVV'&PRWRUWKDWFDQURWDWHSUHFLVHO\ZLWKRXWWKH
need for sensory feedback.

235
 stereovision 9LVXDO VHQVLQJ WKDW XVHV WZR FDPHUDV IRFXVHG RQ WKH VDPH
REMHFW RU VFHQH WR SURYLGH LQIRUPDWLRQ DERXW GHSWK UDQJH DQG WKUHH
dimensional shape.

stigmergy ,QGLUHFW FRPPXQLFDWLRQ E\ DOWHULQJ WKH HQYLURQPHQW WR VLJQDO

other agents.

subsumption architecture$KLHUDUFK\RIEHKDYLRUVLQWKHSURJUDPPLQJRI
a robot that depends on input from sensors.

swarm robotics The coordination and control of multiple robots of similar

type, operating at the same time and place.

technological singularity 7KH VSHFXODWLYH LGHD WKDW DUWL¿FLDOO\ LQWHOOLJHQW

machines will exceed the capacity and control of humans and, by virtue of
those properties, radically alter civilization in unpredictable ways.

telepresence 8VLQJ VHQVRUV DQG LQVWUXPHQWV RQ URERWV WR REVHUYH DQG
measure without the need for a human to be physically present.

thermocouple A sensor that converts external temperature into an electric

signal.

three laws of robotics ,QWURGXFHG E\ ,VDDF $VLPRY LQ WKH VKRUW VWRU\
³5XQDURXQG´)LUVWODZ$URERWPD\QRWLQMXUHDKXPDQEHLQJRU
WKURXJK LQDFWLRQ DOORZ D KXPDQ EHLQJ WR FRPH WR KDUP 6HFRQG ODZ$
robot must obey the orders given it by human beings, except where such
RUGHUVZRXOGFRQÀLFWZLWKWKH¿UVWODZ7KLUGODZ$URERWPXVWSURWHFWLWV
RZQH[LVWHQFHDVORQJDVVXFKSURWHFWLRQGRHVQRWFRQÀLFWZLWKWKH¿UVWRU
second laws.

thrust$IRUFHWKDWSURSHOVDQDQLPDOYHKLFOHRUURERW

WLPHRIÀLJKW'XUDWLRQIRUDVLJQDOWRWUDYHOIURPLWVVRXUFHWRDVHQVRU
Glossary

torque Force applied through a moment arm to rotate or twist.

236
trilateration 0HWKRG IRU GHWHUPLQLQJ ORFDWLRQ XVLQJ WKH PDWKHPDWLFV RI
circles, spheres, and triangles.

transducer $Q\ GHYLFH WKDW FRQYHUWV HQHUJ\ LQ RQH IRUP WR HQHUJ\ LQ
another. For example, a touch sensor transduces the kinetic energy of
movement into electric signals; a motor transduces electric signals into
kinetic energy.

Turing test $OVR NQRZQ DV WKH LPLWDWLRQ JDPH D W\SH RI EHKDYLRUDO
assay proposed by Alan Turing to determine if a machine can demonstrate
linguistic behavior indistinguishable from that of humans.

uncanny valley+\SRWKHVL]HGE\0DVDKLUR0RULDVXGGHQORVVRIDI¿QLW\
that humans can feel as masks or robots become almost, but not quite, human.

visual odometry 0HWKRG IRU XVLQJ VXFFHVVLYH LPDJHV RI D URERW¶V

surroundings to measure its displacement.

voltage divider A simple circuit that uses two resistors, with resistance Z1
and Z2, to reduce the input voltage, Vin, to a lower output voltage, Vout, by the
IROORZLQJHTXDWLRQ

voltmeter $ GHYLFH WKDW PHDVXUHV WKH YROWDJH DFURVV WZR SRLQWV LQ DQ
electric circuit.

wheg An actuator combining the rotational simplicity of a wheel and the

obstacle-clearing capacity of legs.

zero-moment point &UHDWHG E\ 0LRPLU 9XNREUDWRYLü D PHFKDQLFDO

concept that calculates a critical position of legs and body in dynamic motion
in order to preserve stability of a humanoid robot.

237
 Answers

Lecture 1

No questions

Lecture 2

1. One trade-off of walking on two legs instead of four is that we start to

become lousy at moving around in trees. We trade agility in trees for a
toolmaking life on the ground. There is no animal—or robot—that can
do everything well. A robot may be designed to do almost anything, but
no single robot can do everything.

2. The Tadro on the right (with sensors pointed away from the light) would
VZLPDZD\IURPWKHOLJKW&DOFXODWHGLUHFWLRQWRWKHOLJKWDVIROORZV7DNH
the difference of the right sensor’s intensity and the left sensor’s intensity.
For the Tadro on the left, with sensors facing forward, a positive difference
means that the light is to the right. To move toward the light, that positive
difference is then converted into a signal to the motor to turn Tadro to
the right. A negative difference means that the light is to the left, and the
instruction would be to turn Tadro to the left. A difference of zero means
that the light is straight ahead, so Tadro should stay the course.

The Tadro on the left will swim straight as long as both sensors are
pointed away from the light. Once it passes the light, the sensor on the
left will register higher intensity than the one on the right, and Tadro will
turn to the left. But Tadro is moving away from the light and not toward
it. So, as it continues to arc left, the right sensor moves into position
to receive light that is nearly the same intensity as the left sensor. The
difference becomes zero, and this Tadro swims away.

Performing thought experiments like these is the equivalent of using

Answers

RXUPLQGVDVURERWLFVLPXODWRUV7KRXJKWH[SHULPHQWVDUHRIWHQWKH¿UVW
simulation that we run when we start to design a robot. This thought

238
 H[SHULPHQW VKRZV WKDW D UHFRQ¿JXUDWLRQ RI WKH ERG\ FDQ ³UHSURJUDP´
the robot’s behavior without touching the computer software on the
robot’s controller.

Lecture 3

1. A servomotor is a DC motor with a sensor and a controller. What makes

servomotors so useful is that the sensor, a potentiometer, feeds back the
position of the DC motor to the controller.

2. $ '& PRWRU ZRXOG EH VXI¿FLHQW IRU WKH WDQN WUDFN GULYH ³D´ ZKLOH
D VHUYRPRWRU ZRXOG EH VXI¿FLHQW IRU ³F´$ VWHSSHU PRWRU LV EHVW IRU
³E´WRSUHFLVHO\PRYHDQGWKHQKROGSRVLWLRQRIDURERWLFDUP6WHSSHU
motors move at slow speeds, provide higher torque than a servomotor,
DQGFDQEHSUHFLVHO\SRVLWLRQHGVRWKDW¶VEHVWIRU³E´

3. 7KH DQVZHU LV ³F´ OLJKWHPLWWLQJ GLRGH /(' :KHQ DQ /(' HPLWV
light, it is changing the world by altering the local energy patterns of
OLJKW ,Q WKLV FRXUVH ZH XVH D QDUURZHU GH¿QLWLRQ$Q DFWXDWRU LV WKH
system that creates movement, such as a wheel and its drive motor
DQG WUDQVPLVVLRQ7KH EURDG GH¿QLWLRQ WKRXJK FRPSOHPHQWV WKH ZD\
that sensors work. Sensors detect changes in the energy patterns in the
world. Actuators create energy changes in the world.

239
 Lecture 4

1. ,QDQVZHU³D´WKHURERWZRXOGEHVSLQQLQJLWVZKHHOVWU\LQJWRPRYH
IRUZDUGEXWQRWPDNLQJSURJUHVVEHFDXVHLWLVXSDJDLQVWDQREMHFW,Q
DQVZHU³F´WKHURERWZRXOGWXUQLQWRDZDOOUDWKHUWKDQDZD\IURPWKH
ZDOODVLQWKHFRUUHFWDQVZHU³E´

2. The difference in light intensity between the two sensors indicates the
bearing of the light. Tadro calculates the difference between the right
DQGOHIWVHQVRUVULJKWíOHIW x. The controller on the Tadro is set up to
use a positive difference as a signal to turn to the right. With a negative
GLIIHUHQFHULJKWíOHIW7DGURZLOOWXUQWRWKHOHIW1RWHWKDWLIWKHUH
LV QR GLIIHUHQFH ULJKW  OHIW DQG ULJKW í OHIW   WKHQ 7DGUR NHHSV
swimming straight.

That said, we create a one-eyed Tadro in this course that can work with a
single light sensor. The trick with a single sensor is to make the amount
of turning proportional to the amount of light reaching the sensor—so
keep moving and rotate slowly in a spiral motion.

Lecture 5

1. This is a trick question. All three of these answers are viable options,
depending on the task that your robot is trying to accomplish. The
DQVZHU³D´PLJKWEHWKHSHUIHFWVHWRILQVWUXFWLRQVIRUDURERWLFSODQH
.HHS À\LQJ VWUDLJKW DKHDG DW WKLV DOWLWXGH XQWLO IXUWKHU QRWLFH 7KH
DQVZHU ³E´ PLJKW EH SHUIHFW IRU D VHDUFKDQGUHVFXH URERW WKDW LV
ORRNLQJIRUDZDQGHULQJFKLOG6LWDQGZDLWWRVHHLIWKHFKLOGZDQGHUV
ZLWKLQVHQVRUUDQJH7KHDQVZHU³F´FRXOGDOVREHXVHIXOIRUDVHDUFK
and-rescue robot, but the problem with random searches over large
spaces is that they don’t systematically cover the area. This would work
EHVWLQDQHQFORVHGVSDFHVXFKDVDKRPHGXULQJD¿UHZKHQWKHURERW
is searching for the family cat, which might be frightened and hiding
under the sofa.
Answers

240
2. %XLOG ¿YH PRYHPHQW PRGXOHV DQG WKHQ FRPELQH WKRVH PRGXOHV ZLWK
VHQVRU WULJJHUV WR FUHDWH EHKDYLRUV 8VH ,)7+(1(/6( WR KDYH WKH
URERWGHFLGHZKHQWRH[HFXWHWKHGLIIHUHQWEHKDYLRUVLQD¿[HGSULRULW\
hierarchy.

 &UHDWH¿YHPRYHPHQWPRGXOHV

Cruise. Move straight ahead.

 (VFDSHBULJKW 0RYH VWUDLJKW EDFNZDUG IRU RQH VHFRQG DQG WKHQ
turn to the right 45 degrees and stop.
 (VFDSHBOHIW0RYHVWUDLJKWEDFNZDUGIRURQHVHFRQGDQGWKHQWXUQ
to the left 45 degrees and stop.
 $YRLGBULJKW7XUQGHJUHHVWRWKHOHIW
 $YRLGBOHIW7XUQGHJUHHVWRWKHULJKW

 &UHDWHEHKDYLRUV

 (VFDSH,IOHIWRUULJKWEXPSVHQVRULVWULJJHUHGWKHQHVFDSHBOHIWRU
HVFDSHBULJKW
 $YRLG,IOHIWRUULJKW,5LVWULJJHUHGWKHQDYRLGBOHIWRUDYRLGBULJKW
 &UXLVH,IQRVHQVRUVDUHWULJJHUHGWKHQFUXLVH

 &UHDWHWKH¿[HGSULRULW\GHFLVLRQPDNLQJSDUWRIWKHFRGH

If left or right bump sensor is triggered, then escape left or right

 (OVH
If left or right IR is triggered, then avoid left or right
 (OVH
Cruise.

7KHWULFNLVWRQHVWRQH,)7+(1(/6(VWDWHPHQWLQVLGHDQRWKHU7KLV
FUHDWHVD¿[HGSULRULW\KLHUDUFK\ZKHUHRQHEHKDYLRU(VFDSHRYHUULGHV
DQ\RWKHU$VORQJDV(VFDSHLVQ¶WWULJJHUHGWKHQ$YRLGRYHUULGHV&UXLVH

241
 Lecture 6

1. $QVZHU ³E´ D FRPSDVV KHDGLQJ WR WKH QH[W ZD\SRLQW JLYHV WKH
minimum amount of information that you need. In calm waters, if you
know the heading you need to sail, then as long as you stay on that
course, you’ll get to your waypoint. Map coordinates alone won’t get
you where you need to go unless you know where you are, too.

2. 7KHEHVWDQVZHULVXVXDOO\³F´ZDLWXQWLO\RXUXQLQWRVRPHWKLQJWKDW¶V
not on your map and then add it to your map. Ideally, we’d like to keep
WUDFN RI HYHU\WKLQJ LQ RXU ZRUOG DQVZHU ³D´ EXW WKDW LVQ¶W HI¿FLHQW
Depending on how quickly your world changes, you might spend all
of your time updating the model and very little time getting your work
done. There is a trade-off between the accuracy of your model and the
time that you have to perform your task. While monitoring a subset of
REMHFWV DQG SHRSOH PLJKW EH D JRRG FRPSURPLVH WKDW DSSURDFK ZLOO
RQO\ ZRUN LI \RX¶YH FRUUHFWO\ LGHQWL¿HG WKH OLNHO\ REMHFWV DQG SHRSOH
that do not change over time.

3. 0RGHOEDVHGV\VWHPVKDYHWKHEHQH¿WRIDOORZLQJWKHURERWWRXQGHUWDNH
WDVNV WKDW KDYH VSHFL¿F FRPSOLFDWHG JRDOV WKDW DUH VSHFL¿HG LQ WKH
software program. Progress toward that goal can be monitored by an
external observer. In contrast, the cost of a pure behavior-based system
LV WKDW WKH URERW¶V JRDOV DUH VSHFL¿HG RQO\ LQ WHUPV RI KRZ WR FKDQJH
EHKDYLRULQUHVSRQVHWRFKDQJHVLQVHQVRU\LQSXW7KLVEHQH¿WFRVWSDLU
can be turned on its head. Pure model-based systems have the cost that
they cannot respond in a timely fashion to unforeseen circumstances, so
WKH\WHQGWRKDYHKLJKUXQWLPHIDLOXUHUDWHV,QFRQWUDVWWKHEHQH¿WRI
a pure behavior-based system is that it is very robust, able to keep on
moving in the face of rapidly changing circumstances.

Lecture 7

No questions
Answers

242
 Lecture 8

1. In order to close the functional loop to make a behaviorally autonomous

robot, you need to add sensors and a controller, and you need all the
components working together.

Lecture 9

1. )RU³D´\RXFRXOGPRGLI\D5RRPEDE\SXWWLQJD'URS&DPRQLW)RU
³E´\RXFRXOGPRGLI\DTXDGFRSWHUWKDWKDVDFDPHUDWRDYRLGZDOOV
)RU³F´\RXQHHGDGHVLJQIURPVFUDWFK<RX¿UVWKDYHWRJHWWKURXJK
an ice layer that might be miles thick. Once through to the ocean, the
robot needs to be able to maneuver in water. Do you use the same robot
to do it all, or do you use a team of robots? Do you melt your way to
the ocean using a small nuclear reactor? There really is no off-the-shelf
solution. You’ll need a team of crack engineers, tons of money, and a
ride from NASA.

2. :KLOH 5D\%RW LV HQHUJ\ HI¿FLHQW DEOH WR JOLGH DQG DEOH WR VWD\ RQ
station for long periods of time, it doesn’t have the motors or the energy
supply to accelerate rapidly or cruise at high speeds.

3. There are other important features, but the following are three.

(1) Do you have a robot that people want to buy? You need a thorough
PDUNHWLQJVWXG\<RX¶GZDQWWRVHHWKHVSHFL¿FTXHVWLRQVWKDWZHUH
used on a survey, if they took that approach. A good survey asks the
same question in a variety of ways.

 , I\RXUURERWMRLQVDQH[LVWLQJFDWHJRU\KRZFDQ\RXGLIIHUHQWLDWH
your product? Companies making the next kind of vacuum cleaner,
for example, have an identity and shelf-space problem. Why buy
their vacuum cleaner when the Roomba is a proven commodity?
Why should a vendor stock your new machine when he or she has
no idea if someone will buy it? The new company needs a very
clear vision about how they will grab attention, induce a sale, and
then build loyal customers.

243
 (3) Having a management team that includes an experienced and
creative chief of technology and an experienced and creative chief
H[HFXWLYHRI¿FHU0RVWRIWKHWLPHLW¶VGLI¿FXOWWR¿QGRQHSHUVRQ
ZKR FDQ GR ERWK ZHOO &ROLQ $QJOH &(2 RI L5RERW KDV WDONHG
DERXW KLV GLI¿FXOWLHV LQ WUDQVIHUULQJ IURP EHLQJ D WHFK JX\ LQWR
being what he calls a vacuum salesman.

Lecture 10

1. Robots are best for tasks that require extended attention, endurance, and
reliability. Robots don’t sleep, get hungry, or get distracted.

2. (YHQWKRXJKLWLVYHU\GLI¿FXOWWRGRDQVZHU³F´UXQQLQJDQLQWHUQDO
PRGHORIRQHVHOIKDVDQXPEHURIEHQH¿WVRYHUWKHRWKHUWZRDQVZHUV
First, if you are making plans based on a model of yourself working
in the world, then that allows for different conditions to apply each
time you attempt the task. For example, when the library is crowded
with people before exams, the probability of a robot shelving books
LQDWLPHO\PDQQHUPLJKWEHORZEHFDXVHRIDOOWKHWUDI¿F6RUDWKHU
WKDQ VD\LQJ WKDW \RX FDQ QHYHU FRPSOHWH WKDW WDVN DQVZHU ³D´ \RX
rule it beyond your limits at that time, based on those conditions.
Second, your self model can explore different plans and compare their
probabilities of success. For example, two plans might vary only in the
sequence in which you visit different book drop-off points. Given the
state of the world (and perhaps your batteries, or what you are already
carrying) at the time, one sequence that didn’t work before might offer
a solution now.

Lecture 11

1. The robot must have enough kinematic degrees of freedom to reach

each block and place it on the display shelf. It must also have a gripper
to grasp and then release each block.
Answers

8QOLNH³EOLQG´SLFNDQGSODFHURERWVZKHUHWKHURERWGRHVQ¶WVHQVHWKH
SRVLWLRQRIWKHREMHFWWKLVEORFNVSHOOLQJURERWPXVWNQRZWKHSRVLWLRQ
of each block and the letter represented, which it does by (1) knowing

244
 WKHDQJOHRIHDFKRILWVMRLQWVDQGLWXVHVIHHGEDFNFRQWUROZLWKDQJXODU
position sensors; (2) having force-feedback sensors on the grippers to
VHQVHLIWKHREMHFWKDVEHHQJULSSHGKDYLQJDZRUOGPRGHOWKDWPDSV
the position of each letter block at all times.

Lecture 12

1. $ KDQGVRQ DSSURDFK ZLOO JLYH \RX D YHU\ FOHDU DQVZHU KHUH 3LFN XS
Roomba while it is on, and manually touch the bump sensor lightly
DQGWKHQPRUH¿UPO\REVHUYLQJWKHGLUHFWLRQLQZKLFKWKHZKHHOVDUH
spinning. Note that the moment you pick Roomba up, it will think that
it has detected an edge, so you’ll have to put a little masking tape on the
infrared sensors that do the edge detection. But now that you are holding
LW\RXFDQDSSURDFKWKHEXPSHUIURPEHORZZLWK\RXU¿QJHUDQGNQRZ
that forward-looking infrared detectors aren’t at work here.

Lecture 13

1. All of the following would help us build a fully autonomous robotic

VXUJHRQDSODQQLQJFRPSRQHQWRIWKHFRQWUROOHUWKDWLVIRUZDUGLQJWKH
next steps to an independent surgeon or surgical AI before action is taken
as extra screening precaution; (2) a model-based controller that uses and
updates a three-dimensional map of each patient; and (3) algorithms and
new types of sensors to identify tissue, organ, and cell types.

2. <RXPLJKWRSWIRUVWURQJHUMRLQWPRWRUVEHFDXVHPRVWVPDOODUPVOLNH
WKH'(.$$UPDUHQRWYHU\SRZHUIXO<RXPLJKWZDQWWRLPSURYHRU
increase the number or types of sensors on the arm so that the human
knows better—beyond visual feedback—what the arm is doing. Or, you
could decide to increase the autonomy of the arm so that it interacts with
the human in more complex ways. For example, what if the arm, using
signals from an eye-tracker system in the frame of your eyeglasses,
could anticipate what you are about to do based on the direction of your
gaze? Then, it could put itself in a state of preparation. Anticipation and
preparation require that the arm have a model of itself, a model of you,
and a model of how you interact.

245
 Lecture 14

1. A hybrid controller that combines features and fail-safes from both

systems would be best. A behavior-based controller creates important
UHÀH[HV VXFK DV DYRLGLQJ REVWDFOHV VWD\LQJ RQ WKH URDG DQG UHDGLQJ
signs. A model-based controller uses comprehensive road maps coupled
with GPS and roadside scanning to navigate carefully.

2. Radar. Here is what Dickmann has to say in an article about the

0HUFHGHVURERWLFFDUOLVWHGLQWKH6XJJHVWHG5HDGLQJIRUWKLVOHFWXUH
³8QOLNHRSWLFDOV\VWHPVUDGDURSHUDWHVZHOOQRPDWWHUZKDWWKHZHDWKHU
ZRUNLQJDVLWGRHVZLWKPLFURZDYHV´

Lecture 15

1. Notice that the question didn’t say if the autopilot was engaged. If
it is engaged, then you have a closed-loop system. Most commercial
DLUFUDIW DUH D FODVVLF FDVH RI ³PL[HG DXWRQRP\´ ERWK RSHQORRS and
closed-loop systems. Keep in mind that mixed autonomy systems
can operate in parallel, such as when you steer your car but the cruise
control has autonomy over speed, or in series, such as when the
plane’s pilot steers during takeoff but then hands the steering over to
the autopilot.

Lecture 16

1. Most of the time, we try to build underwater robots that are passively
³UROOVWDEOH´7KLVPHDQVWKDWDQVZHU³E´LVEHWWHUHYHQWKRXJK³D´LV
also feasible. Think about Robot Madeleine (see image).

She is highly maneuverable in roll, pitch, and yaw. She can roll
TXLFNO\E\KDYLQJWKHWZRÀLSSHUVRQRQHVLGHJHQHUDWHDGRZQZDUG
WKUXVW ZKLOH WKH WZR ÀLSSHUV RQ WKH RWKHU VLGH JHQHUDWH DQ XSZDUG
WKUXVW 6KH FDQ SLWFK TXLFNO\ E\ KDYLQJ WKH IURQW DQG EDFN ÀLSSHUV
Answers

generate vertical thrust in opposition. She can yaw quickly by

246
 © John Long.
KDYLQJWKHOHIWVLGHÀLSSHUVJHQHUDWHDIRUZDUGWKUXVWZKLOHWKHULJKW
VLGH ÀLSSHUV JHQHUDWH D UHDUZDUG WKUXVW $W WKH VDPH WLPH 5RERW
0DGHOHLQH¶V ÀDWWHQHG ERG\ FUHDWHV UHVLVWDQFH WR URWDWLRQ LQ UROO
and pitch that makes her stable in these directions when she isn’t
actively turning.

2. :KLOHDOORIWKHVHDUHWUXH³E´LVSUREDEO\WKHPRVWYH[LQJSUREOHP
7KHFKDOOHQJHLVPRVWGLI¿FXOWLQVDOWZDWHUZKHUHWKHVDOWVLQVROXWLRQ
increase electric conductance and, hence, the ability to short-circuit
HOHFWURQLFV %XLOGLQJ D ZDWHUWLJKW URERW LV GLI¿FXOW EHFDXVH \RX
QHHG DFFHVV WR WKH LQVLGHV RI WKH URERW IRU UHSDLUV 7KH MRLQWV DQG
seams that make that possible then become the places most likely
to leak.

247
 Lecture 17

1. All three answers offer potential advantages. Complex missions, such

DV KRZ WR H[SORUH DQRWKHU SODQHW EHQH¿W IURP KDYLQJ KXPDQV LQ WKH
loop. Telepresence enhances everything from the entertainment value of
recreational robots to the research value of rovers on Mars. Checking on
the robot when we do remote exploration at the bottom of the ocean or
on another planet is more complete when we keep humans in the loop.

Lecture 18

1. While all of these are true, the greatest vulnerability in time of war comes
IURPWKHKXPDQVLQWKHFRQWUROORRS:LWKZHDSRQVWKH¿UVWUXOHLVWR
not kill your own troops. In the laws of war, eliminating the deaths of
noncombatants is paramount. The navy has an advantage over the army
in that at least in open-ocean conditions, the number of noncombatants
is likely to be small. The next step might be to have the commanding
RI¿FHUGHFLGHWKHOHYHODQGW\SHRIDXWRQRP\WRJLYHWRWKHV\VWHPDQG
when to withdraw each increment of autonomy. For example, you could
LPDJLQH WKDW LI \RX VXGGHQO\ ZHUH WDNLQJ KHDY\ ¿UH \RX FRXOG PRUH
quickly mobilize your defenses using the fully autonomous system.

Lecture 19

1. First, calculate the kinetic energy in each case. The microbot has a mass
of 0.01 kilograms and a velocity of 1 meter per second. If we square
the velocity, we still get 1, and the product of 1 and 0.01 is the kinetic
HQHUJ\RIMRXOHVZKHUHMRXOHVDUHWKH6,XQLWIRUHQHUJ\

For the cheetah, its mass is 100 kilograms. Square the velocity of 10
meters per second and we get 100. The product of that square and the
PDVV LV  MRXOHV 6R HYHQ WKRXJK WKH FKHHWDK LV RQO\ JRLQJ 
WLPHVDVIDVWDVWKHPLFURERWÀ\HUZKHQWKH\LPSDFWWKHNLQHWLFHQHUJ\
of the cheetah is 1 million times greater.
Answers

248
2. A robot has to use its onboard energy source (battery or fuel) and convert
that potential energy into kinetic energy. This is where actuators get
involved and the square of velocity comes back to bite us. To go a little
bit faster takes a lot more kinetic energy.
More kinetic energy requires more fuel. 5000
4500

kinteic energy of robot

Look at the graph. The mass of our robot 4000
is 100 kilograms. The speed increases by 2 3500
orders of magnitude, from 0.1 to 10 meters 3000

(Joules)
per second. Over that same range of speed, 2500
2000
the kinetic energy increases from 0.5 to
1500
MRXOHVRURUGHUVRIPDJQLWXGH 1000
500
The trade-off is that when you go faster, 0
you run out of fuel faster. This might 0 5 10
speed of robot
seem like common sense, but there’s more (meters per second)
SK\VLFV IXQ KHUH %HFDXVH \RX XVH PRUH
fuel to move faster, you don’t move as far. That’s the real trade-off. For a
given amount of fuel, you can either choose to move fast or far.

Lecture 20

1. 7KH EHVW RSSRUWXQLW\ IRU VWLJPHUJ\ LV ZLWK VFHQDULR ³F´ 2QH RI WKH
huge challenges in any search-and-rescue situation is to quickly and
systematically cover an area. If these robots are working in snow, and
perhaps listening directly below for any signs of life, then the simplest
thing to do is to have them keep a set distance away from any other
robot’s trail. That trail could be the one that they made previously, or it
could be one from another robot.

:HFDQUXOHRXWDQVZHU³D´EHFDXVH&XULRVLW\LVQRWOLNHO\WRGRXEOH
back on its explorations. If you leave yourself a note, you have to go
EDFNDQGUHDGLWWRPDNHDQ\VHQVHRILW7KHRQHZD\WULSIRUWKHÀHHWRI
VHOIGULYLQJFDUVDQGWUXFNVORRNVDW¿UVWJODQFHOLNHWKHVDPHVLWXDWLRQ
but with many robots on the same road, they could help each other out.
For example, suppose that every time a robot encountered a dangerous
pothole, it placed a little radio beacon to mark the danger.

249
 Lecture 21

1. This is a tough, open-ended question. If you are physically minded, you

could think about life-forms being local dis-entropy machines, able to
harvest energy so that they can construct order in the form of complex
bodies. If you are biologically minded, you might focus on the idea of
reproduction and the transmission of information for self-production
from generation to generation. Or maybe you are putting this all together
and thinking about the different kinds of autonomy that we see in life-
IRUPVEHKDYLRUUHSURGXFWLRQGHYHORSPHQWDQGHYROXWLRQ:RXOG\RX
say that a group of robots fully possessing all those types of autonomy
would be functionally indistinguishable from life itself?

2. 7KHFRUUHFWDQVZHULV³E´'HVLJQVWKDWSHUIRUPWKHEHVWDUHFRPELQHG
and altered to create a new group of nonidentical robots. This isn’t
SKUDVHGLQWKHXVXDOZD\WKDW\RXPLJKW¿QGLQDELRORJ\WH[WERRN$OVR
WKHUHVSRQVHGLGQ¶WVSHFLI\WKHQDWXUHRIWKHMXGJLQJEXWLQQDWXUHZKDW
KDSSHQV RIWHQ LV WKDW SUHGDWRUV VHOHFW WKH ZLQQHUV VHH ¿JXUH 2WKHU
NLQGVRIVHOHFWLRQIRUFHVPD\RFFXU&RPSHWLWLRQIRUUHVRXUFHVHQHUJ\
can make some individuals winners and others losers. The winners are
the survivors.

predators survivors
select reproduce

survivors
become parents

population population
generation 1 generation 2
Answers

250
 Lecture 22

1. The simplest feature to track is the direction of a human’s gaze,

DQVZHU³D´:HNQRZIURPPDQ\SV\FKRSK\VLFDOH[SHULPHQWVWKDWWKH
eyes of humans automatically track to features and movements in the
environment. Social robots bank on this and capture a human’s attention
using their own movements, sounds of certain types, and even displays
of light or images.

2. Carry the pills. Remind the patient. Follow the patient, or wake up the
patient. Carry around water. Monitor the dispensing of the pills. Monitor
the taking of pills. Communicate with the physician any irregularities in
dosages or timing of the pills. Track the patient verbally and with visual
record keeping for side effects and progression or retreat of the illness.
Wireless monitoring of wearable devices to track body temperature,
sleep, hydration, blood pressure, and heart rate.

Lecture 23

1. Automatons from hundreds of years ago had very clever actuators,

bodies, and energy sources. But what tended to be missing were sensors
DSDUWIURPWKHRQRIIVZLWFKZKLFKZRXOGQ¶WFRQWULEXWHWREHKDYLRUDO
autonomy). The functional loop would be wide open. As far as we know,
Leonardo never solved the sensor problem.

But perhaps his biggest challenge was not being aware of electricity—
not only a source of energy but also the currency for information
throughout the modern robot. All sensor signals are converted to
patterns of electricity. All signals to the actuators are given as patterns
of electricity. All computations on the controller are carried out by
miniaturized electric switches.

251
 Lecture 24

1. $JULFXOWXUH ³G´ UHWDLO VHUYLFHV ³H´ SHUVRQDO DQG KRPH KHDOWK FDUH
³I´ PLOLWDU\ DQG GHIHQVH V\VWHPV ³J´ DQG ¿UVWUHVSRQGHU VHUYLFHV
³K´ DUH DOO DUHDV WKDW DUH VWLOO GRPLQDWHG E\ KXPDQV SHUIRUPLQJ
manual labor. Because the history of technology is in many ways the
story of how we use tools and machines to reduce the need for manual
ODERUFKDQFHVDUHWKDWWUHQGZLOOFRQWLQXH(PEHGGHGURERWLFV\VWHPV
will be the tools of change, because they are psychologically easier
for us to accept than a humanoid because we simply don’t see them.
Out of sight is out of mind. For psychological reasons as well, these
embedded robotic systems won’t be called robots. Instead, look for
WKHVHHXSKHPLVPVDGDSWLYHDVVLVWLYHDXWRPDWLFFRJQLWLYHLQWHOOLJHQW
and smart.

7KH RWKHU RSWLRQV²WUDQVSRUWDWLRQ ³D´ DVVHPEO\ DQG PDQXIDFWXULQJ

³E´DQGPHGLFDOGHYLFHV³F´²DUHDOUHDG\GRPLQDWHGE\HPEHGGHG
robotic systems.

2. $Q\ RI WKH DERYH &RPPHUFLDO URERWLFV LV IXQGDPHQWDOO\ OLNH DQ\
RWKHUEXVLQHVV<RXKDYHWR¿JXUHRXWZKDWSHRSOHZDQWZKDWWKH\DUH
willing to purchase, and how much they are willing to pay. That’s called
market analysis, which you’ll have to perform when you create your
¿UVWEXVLQHVVSODQ
Answers

252
 Bibliography

$QJHOHV -RUJH DQG )UDQN & 3DUN ³3HUIRUPDQFH (YDOXDWLRQ DQG 'HVLJQ
&ULWHULD´ Springer Handbook of Robotics ± %HUOLQ +HLGHOEHUJ
Springer, 2008. This is a formal treatment of the mechanical design of
robots. The emphasis on also designing the workspace is very helpful.

$QRQ\PRXV ³$Q$XWRPDWLF %ORFN6HWWLQJ &UDQH´ Meccano Magazine 23,

QR0DUFK$YDLODEOHDWKWWSZZZQ]PHFFDQRFRP00YLHZHU
SKS7KLVLVWKH¿UVWUHFRUGRI*ULI¿WK37D\ORU¶V³5RERW*DUJDQWXD´7D\ORU¶V
URERWLVRIWHQFRQVLGHUHGWREHWKH¿UVWSURJUDPPDEOHURERW

Banzi, Massimo. Getting Started with Arduino. 6HEDVWRSRO &$ 0DNH

Books, 2011. Within a few minutes of referencing chapters 1 through 4,
you can load code onto your Arduino and hook up sensors and motors. The
Arduino philosophy gets you to work making and tinkering.

Bekey, G. A. Autonomous Robots: From Biological Inspiration to

Implementation and Control. &DPEULGJH0$0,73UHVV&KDSWHU
is an introduction to robotic autonomy that includes explanations of control
theory and the parts needed to build an autonomous robot.

Bohm, Harry, and Vickie Jensen. Build Your Own Underwater Robot, and
Other Wet Projects. 9DQFRXYHU:HVWFRDVW:RUGV7KLVGRLW\RXUVHOI
book will help you avoid the catastrophic failures that only water can bring.

Braitenberg, Valentino. Vehicles: Experiments in Synthetic Psychology.

&DPEULGJH 0$ 0,7 3UHVV  /HDUQ WR XVH WKH VLPSOHVW DSSURDFK
possible to understand how robots behave with the famous thought
H[SHULPHQWVIRXQGLQFKDSWHUVWKURXJK%UDLWHQEHUJ¶VLGHDRID³YHKLFOH´
is that all you need to create an autonomous agent is a body with a sensor
attached to a motor.

253
 %UHD]HDO &\QWKLD ³&RJQLWLYH 0RGHOLQJ IRU %LRPLPHWLF 5RERWV´ &KDSWHU
9 in Biologically Inspired Intelligent Robots(GLWHGE\<RVHSK%DU&RKHQ
and Cynthia Breazeal.%HOOLQJKDP:$63,(7KH,QWHUQDWLRQDO6RFLHW\IRU
2SWLFDO(QJLQHHULQJ%UHD]HDOLVRQHRIWKHSLRQHHUVRIVRFLDOURERWLFV
and in this chapter, she details both the design of Kismet and the basic
principles behind building emotions into robots.

Brooks, Rodney. Flesh and Machines: How Robots Will Change Us. New
<RUN 3DQWKHRQ %RRNV  %URRNV LV ZULWLQJ IRU D JHQHUDO DXGLHQFH VR
he gives an approachable history of behavior-based robotics. In chapter 3,
Brooks explains the design of Grendel, a legged robot built for planetary
H[SORUDWLRQ*UHQGHOEHJDW6RMRXUQHUWKH¿UVWDXWRQRPRXVURYHURQ0DUV

%XUGHW(')UDQNOLQDQG7(0LOQHUHuman Robotics: Neuromechanics

and Motor Control. &DPEULGJH 0$ 0,7 3UHVV  ,Q FKDSWHU  WKH
authors explain how humans control the motion of their limbs and body as
well as how biological sensors and actuators are the basis for engineering
neuromechanical arms. In chapter 11, they explore neurorehabilitation
and robotics.

Bureau of Ordnance, Department of Navy. The Whitehead Torpedo, U.S.N.,

45 c/m. x 3.55 m. Mark I, Mark II, Mark III, and 45 c/m. x 5 m. Mark I.
General Description. Naval Torpedo Station, document no. 1598, 1898.
Available online at KWWSPDULWLPHRUJGRFZKLWHKHDG 7KH ¿UVW  SDJHV
take you through Mark I. Make sure to look at the plates, which offer some
lovely diagrams of the internal mechanisms and external parts.

&LXWL *$ 0HQFLDVVL DQG 3 'DULR ³&DSVXOH (QGRVFRS\ )URP &XUUHQW
$FKLHYHPHQWV WR 2SHQ &KDOOHQJHV´ IEEE Reviews in Biomedical
Engineering   ± <RX¶OO KDYH WR HLWKHU SD\ WR SXUFKDVH WKLV
article or get it through your library. The hassle is worth it if you want to read
a good overview of this emerging robotics discipline.
Bibliography

Clancey, William. :RUNLQJRQ0DUV9R\DJHVRI6FLHQWL¿F'LVFRYHU\ZLWKWKH

Mars Exploration Rovers.&DPEULGJH0$0,73UHVV7KHVFLHQWL¿F
mission of Spirit and Opportunity is explained in chapter 4. In chapter 6, the
rovers are on Mars and exploring.

254
Cook, Gerald. Mobile Robots: Navigation, Control and Remote Sensing.
+RERNHQ 1- -RKQ :LOH\  &KDSWHU  LV D SUDFWLFDO H[SRVLWLRQ RI
getting your robot to navigate; it includes coordinate systems, GPS, and
dead reckoning.

Craig, John. Introduction to Robotics: Mechanics and Control. 3rd ed.

8SSHU 6DGGOH 5LYHU 1- 3HDUVRQ 3UHQWLFH +DOO  ,Q FKDSWHU  WKH
basic problems and principles are laid out for kinematics and dynamics of
robotic manipulators.

Dawson, Michael, Brian Dupuis, and Michael Wilson. From Bricks to

Brains: The Embodied Cognitive Science of LEGO Robots(GPRQWRQ$%
Athabasca University Press, 2010. Chapter 6 is a great analysis of Grey
Walter’s tortoises. Dawson includes instructions for building a tortoise using
WKH/(*2™ Mindstorms™ robot kit. Chapter 7 includes the computer code
for the tortoise.

Devol, George C. 1961. Programmed article transfer. U.S. Patent 2,988,237.

This is a landmark patent for industrial robotics. The system Devol describes
GRHVQRWORRNOLNHDURERWVR\RX¶OOKDYHWRXVH\RXUGHWHFWLYHVNLOOVWR¿JXUH
out how sensors and actuators are coordinated.

'LFNPDQQ -XUJHQ 1LOV $SSHQGURW DQG &DUVWHQ %UHQN ³+RZ :H *DYH
6LJKW WR WKH 0HUFHGHV 5RERWLF &DU´ IEEE Spectrum KWWSVSHFWUXPLHHH
RUJWUDQVSRUWDWLRQVHOIGULYLQJKRZZHJDYHVLJKWWRWKHPHUFHGHVURERWLF
car. Posted July 24, 2014. Dickmann, one of the pioneers of autonomous
navigation in cars, explains the development of radar systems and the
importance of combining different types of sensor systems.

'\P &OLYH 3DWULFN /LWWOH DQG (OL]DEHWK 2UZLQ Engineering Design: A

Project-Based Introduction. 4thHG+RERNHQ1--RKQ:LOH\&KDSWHUV
1 and 2 emphasize the process of designing, including vocabulary and the
LPSRUWDQFH RI DVNLQJ TXHVWLRQV :KLOH QRW IRU URERWLFV VSHFL¿FDOO\ WKH
approach to engineering design is completely relevant. Chapters 4 through
6 help us clarify the design by focusing on function. Limits and constraints
help frame what is possible.

255
 Floreano, Dario, and Claudio Mattiussi. %LRLQVSLUHG$UWL¿FLDO,QWHOOLJHQFH
Theories, Methods, and Technologies &DPEULGJH 0$ 0,7 3UHVV 
7KLVLVDJRRGFRPSOHPHQWWR.HUQEDFK¶VRYHUYLHZRIWKH¿HOG³,QWURGXFWLRQ
WR &ROOHFWLYH 5RERWLFV 5HOLDELOLW\ )OH[LELOLW\ DQG 6FDODELOLW\´ )ORUHDQR
and Mattiussi introduce some mathematical details and provide examples of
robotic systems in which collective algorithms have been implemented.

Grundfest, W. S., J. W. Burdick, and A. B. Slatkin. 1994. Robotic endoscopy.

U.S. Patent 5,337,723. This patent set the stage for many later robots that go
into the body as either miniaturized vehicles or robotic arms.

Hanson, David, Daniela Rus, Steven Canvin, and Gernot Schmierer.

³%LRORJLFDOO\ ,QVSLUHG 5RERWLFV$SSOLFDWLRQV´ &KDSWHU  LQ Biologically
Inspired Intelligent Robots (GLWHG E\ <RVHSK %DU&RKHQ DQG &\QWKLD
Breazeal. %HOOLQJKDP :$ 63,( 7KH ,QWHUQDWLRQDO 6RFLHW\ IRU 2SWLFDO
(QJLQHHULQJ%\IRFXVLQJRQWKHPRGXOHVDQGWDVNVSHFL¿FDSSOLFDWLRQV
of robots, this chapter provides several different perspectives on design.

Hickman, Ryan, James Kuffner, James Bruce, Chaitanva Gbarpure, Damon

Kohler, Arshan Poursohi, Anthony Francis, and Thor Lewis. 2014. Shared
robot knowledge base for use with cloud computing system. U.S. Patent
8,639,644 B1. This patent laid the groundwork for cloud robotics and for a
FROOHFWLYHURERWLF³PLQG´GDWDEDVH

Hornyak, Timothy. Loving the Machine: The Art and Science of Japanese
Robots. 1HZ<RUN.RGDQVKD,QWHUQDWLRQDO&KDSWHUJHWVXVXQGHU
the skin of the Actroid androids to understand how they are actuated with
pneumatic systems, plus the metaphysical questions that drive creator
Hiroshi Ishiguro’s development of lifelike robots.

+XDQJ +XL0LQ (OHQD 0HVVLQD DQG -DPHV $OEXV Autonomy Levels for
Unmanned Systems (ALFUS) Framework. Volume II: Framework Models.
Version 1.0. NIST Special Publication 1011-II-1.0, 2007. Prepared for the
Bibliography

U.S. military, this document grapples with the complexities of autonomy.

7KHLU IUDPHZRUN XVHV WKUHH SDUDPHWHUV IRU GHVLJQ GHFLVLRQV KXPDQ
independence, mission complexity, and environmental complexity.

256
L5RERW ³L5RERW© 2ZQHU¶V 0DQXDOV DQG 4XLFN6WDUW *XLGHV´ 2ZQHU¶V
0DQXDO IRU 5RRPED KWWSKRPHVXSSRUWLURERWFRPDSSDQVZHUVGHWDLODB
LGaLURERW&$(RZQHUVPDQXDOVDQGTXLFNVWDUWJXLGHV :KLOH
this might not seem like exciting reading, it actually is pretty cool to look
at the capacities of these robots, how they work, and how you can do some
simple troubleshooting and repair.

-DUQRW&KDUOHV³+LVWRU\´&KDSWHULQIntroduction to Unmanned Aircraft

Systems(GLWHGE\5LFKDUG%DUQKDUW6WHSKHQ+RWWPDQ'RXJODV0DUVKDOO
DQG (ULF 6KDSSHH %RFD 5DWRQ )/ &5& 3UHVV  -DUQRW SURYLGHV D
concise, detailed history of UAVs, including some of the context behind
Sperry’s aerial torpedo patent of 1917.

Jones, J. L., A. M. Flynn, and B. A. Seiger. Mobile Robots: Inspiration

to Implementation. 2nd HG :HOOHVOH\ 0$ $. 3HWHUV  &KDSWHU
2 contains the instructions for building your own Tutebot and a great
explanation of the workings of the electronics. Chapter 5 reviews a range
of sensors and explains how they work. Chapter 6 offers insights into
actuators built to move robots around, including wheels and legs. Chapter
7 explains how DC and servomotors work and how to size them properly
for your robot.

Karvinen, Kimmo, and Tero Karvinen. Make: Getting Started with Sensors.
6HEDVWRSRO&$0DNHU0HGLD&KDSWHUVDQGZLOOJHW\RXEXLOGLQJ
simple sensor circuits immediately. And you will understand them by
building them.

.HUQEDFK 6HUJH ³,QWURGXFWLRQ WR &ROOHFWLYH 5RERWLFV 5HOLDELOLW\

)OH[LELOLW\ DQG 6FDODELOLW\´ &KDSWHU  LQ Handbook of Collective
Robotics: Fundamentals and Challenges. (GLWHG E\ 6HUJH .HUQEDFN
6LQJDSRUH 3DQ 6WDQIRUG 3XEOLVKLQJ  &ROOHFWLYH URERWLFV WKH
RYHUDUFKLQJ¿HOGWKDWLQFOXGHVVZDUPURERWLFVH[DPLQHVV\VWHPVWKDWDUH
cooperative, networked, swarming, and organized around the principle of
small-world networks.

257
 /RQJ -RKQ + -U ³%LRPLPHWLFV 5RERWLFV %DVHG RQ )LVK 6ZLPPLQJ´ ,Q
Encyclopedia of Fish Physiology: From Genome to Environment, edited by
$3)DUUHOO±6DQ'LHJR$FDGHPLF3UHVV/RQJWDNHV\RX
WKURXJKWKHGHWDLOHGSURFHVVRIGHVLJQLQJDQGEXLOGLQJD¿VKOLNHURERWDQG
reviews some of the recent history in biomimetic robotics.

———. Darwin’s Devices: What Evolving Robots Can Teach Us About

the History of Life and the Future of Technology.1HZ<RUN%DVLF%RRNV
2012. Long explains why and how to evolve robots, focusing on how this
helps biologists answer questions about how extinct animals evolved. The
book explores notions of intelligence in animals and robots and how you can
design robots that mimic life-forms.

Monk, Simon. Hacking Electronics: An Illustrated DIY Guide for Makers

and Hobbyists1HZ<RUN0F*UDZ+LOO(GXFDWLRQ&KDSWHUVDQG
2 instruct how to assemble a kit with tools and electronic parts. Stripping
wire and soldering, invaluable skills for building or hacking, are also
introduced. Chapter 2 introduces electricity and circuit diagrams. In
Chapter 2, Monk explains how to read a circuit diagram. Chapter 3 gets
XVKDFNLQJDQGPDNLQJFLUFXLWV0RQN¶VSXVKOLJKWKDFNFDQEHPRGL¿HG
to make a light-sensitive circuit that turns a tank track actuator into a
light-sensitive robot.

———. Programming Arduino: Getting Started with Sketches. 1HZ<RUN

McGraw-Hill. With more detail than Banzi’s Getting Started with Arduino,
this is a great companion. Chapters 1 through 3 do the important work of
teaching you some of the formal structures of programming in general and
the C computer languages in particular.

0RUL 0DVKLPRUL ³7KH 8QFDQQ\ 9DOOH\´ IEEE Robotics & Automation

Magazine -XQH  ± 7UDQVODWHG E\ .DUO ) 0DF'RUPDQ DQG
1RUUL .DJHNL $YDLODEOH DW KWWSVSHFWUXPLHHHRUJDXWRPDWRQURERWLFV
KXPDQRLGVWKHXQFDQQ\YDOOH\7KLVLVDQLQÀXHQWLDOSDSHULQURERWLFVZLWK
Bibliography

its ideas spreading throughout the worlds of entertainment and psychology.

Social robotics, and recognition of the importance of our emotional
relationship with robots, begins here.

258
Murphy, Robin. Introduction to AI Robotics. &DPEULGJH 0$ 0,7 3UHVV
&KDSWHUSUHVHQWVWHFKQLFDOGHWDLOVDERXWWKH³ZKHUHDP,"´SUREOHP
of topological navigation.

Nourbakhsh, Illah. Robot Futures &DPEULGJH 0$ 0,7 3UHVV 

Chapter 6 is the bright spot in the book; it focuses on the possibilities that
robotics provides to empower individuals.

Pfeifer, Rolf, and Josh Bongard. How the Body Shapes the Way We Think: A
New View of Intelligence. &DPEULGJH0$0,73UHVV&KDSWHUVDQG
RIIHUDJUHDWLQWURGXFWLRQWRWKHFRQFHSWRIHPERGLPHQW%HKDYLRUUHVXOWV
from the physical interactions of robots and their environments.

3LQKHLUR $ 9 ' +DQ : 0 6KLK DQG + <DQ ³&KDOOHQJHV DQG
2SSRUWXQLWLHVIRU6WUXFWXUDO'1$1DQRWHFKQRORJ\´Nature Nanotechology
±7KLVUHYLHZSURYLGHVWKHEDFNJURXQGWRXQGHUVWDQGKRZ
ZHFDQEXLOG'1$VSLGHUVDQGPROHFXODU³URERWV´

5H\QROGV &UDLJ ³)ORFNV +HUGV DQG 6FKRROV $ 'LVWULEXWHG %HKDYLRUDO

0RGHO´ SIGGRAPH ’87: Procs. 14 Ann. Conf. Computer Graphics
Interactive Tech±$YDLODEOHDWKWWSZZZFVWRURQWRHGXaGW
VLJJUDSKFRXUVHFZU:KLOHZULWWHQIRUZRUNHUVLQFRPSXWHUJUDSKLFV
this seminal paper in swarm robotics enumerates the three principles for
ZKLFK5H\QROGVDQGKLV%RLGVKDYHEHFRPHIDPRXVFROOLVLRQDYRLGDQFH
(2) velocity matching, and (3) centering.

Rosheim, Mark. Leonardo’s Lost Robots %HUOLQ 6SULQJHU 6FLHQFH 

Business Media, 2006. In chapter 3, Rosheim reconstructs Leonardo’s robot
knight. This reads like a detective story, and it’s a great way to appreciate
Leonardo’s brilliance and Rosheim’s creativity.

———. Robot Evolution: The Development of Anthrobotics 1HZ <RUN

-RKQ :LOH\  6RQV  (YHQ WKRXJK PXFK KDV FKDQJHG VLQFH WKLV
book was published in 1994, the problems of building humanoids have
not. As Rosheim steps through the history of humanoids, which he calls
³DQWKURERWV´KHDOVRVSHOOVRXWGHVLJQJRDOVDQGSULQFLSOHV

259
 Scherz, Paul, and Simon Monk. Practical Electronics for Inventors. 3rd ed.
1HZ<RUN0F*UDZ+LOO(GXFDWLRQ7KHIURQWÀ\OHDYHVDQGFKDSWHUV
1 and 2 provide more detail than Monk’s Hacking Electronics. Chapter 2 is
D VHULRXV LQWURGXFWLRQ WR HOHFWULFLW\ DQG HOHFWURQLFV &KDSWHU  ³+DQGV2Q
(OHFWURQLFV´ LV D PXVWUHDG LI \RX DUH JRLQJ WR EXLOG URERWV IURP VFUDWFK
Foremost are important precautions about safety. Learn how multimeters and
oscilloscopes work. This is full of details about tools and designing your
workspace that can keep you occupied for hours.

Singer, P. W. :LUHGIRU:DU7KH5RERWLFV5HYROXWLRQDQG&RQÀLFWLQWKHst
Century.1HZ<RUN3HQJXLQ3UHVV&KDSWHUVWKURXJKSURYLGHDQ
excellent introduction to robotics in war, including a brief history.

6PLWKHUV7³$XWRQRP\LQ5RERWVDQG2WKHU$JHQWV´Brain and Cognition

QR±7KLVSDSHUUHYLHZVXVHVRIWKHWHUP³DXWRQRPRXV´
While we use this term to mean that a robot has self-control, Smithers argues
that true autonomy requires a robot that can learn. He calls this type of
DXWRQRP\³VHOIUXOLQJ´

6SHUU\(OPHU:LUHOHVVFRQWUROOHGDHULDOWRUSHGR863DWHQW
937. Filed in 1917, this patent covers what today we would call radio-
controlled planes. Introduced with military applications in mind, Sperry’s
other innovation was the gyroscopic sensors that could be used for feedback
WRKDYHWKHSODQHÀ\DXWRQRPRXVO\

6WDPS -LPP\ ³$ %ULHI +LVWRU\ RI 5RERW %LUGV 7KH (DUO\ *UHHNV DQG
5HQDLVVDQFH $UWLVWV +DG %LUGV RQ 7KHLU %UDLQV´ Smithsonian Magazine,
0D\KWWSZZZVPLWKVRQLDQPDJFRPDUWVFXOWXUHDEULHIKLVWRU\
RIURERWELUGVTON;3M9,'8S;6WDPSWDNHVXVEDFNWR
%&(ZKHQWKH*UHHNPDWKHPDWLFLDQ$UFK\WDVRI7DUHQWXPLVSXUSRUWHG
WR KDYH EXLOW D À\LQJ GRYH :H OHDUQ RI +HUR RI$OH[DQGULD¶V GHVLJQV IRU
pneumatically powered mechanical animals.
Bibliography

Thrun, S., W. Burgard, and D. Fox. Probabilistic Robotics&DPEULGJH0$

0,7 3UHVV  $YDLODEOH DW KWWSPLWSUHVVPLWHGXERRNVSUREDELOLVWLF
robotics. The central mission of this new approach in robotics is to quantify
uncertainty. Chapter 1, available for free, sets the stage.

260
7KUXQ6HEDVWLDQHWDO³6WDQOH\7KH5RERW7KDW:RQWKH'$53$*UDQG
&KDOOHQJH´Journal of Field RoboticsQR±$YDLODEOH
DW ZZZURERWLFVXVFHGXaPDMDWHDFKLQJFVSDSHUVWKUXQVWDQOH\
pdf. Written shortly after Thrun and Stanford Racing won the 2005 DARPA
challenge, this paper is now a classic in robotics.

U.S. Department of Defense. Unmanned Systems Integrated Roadmap

FY2013–2038. 2SHQ3XEOLFDWLRQ5HIHUHQFH1XPEHU67KLVRIIHUV
a fascinating look at what the U.S. Department of Defense is making public
about their plans for robots in defense systems.

YDQ GH 0ROHQJUDIW 5HQH ³)LQDO 3URMHFW 5HSRUW 5RER(DUWK´$YDLODEOH DW

KWWSURERHDUWKRUJZSFRQWHQWXSORDGVGRFXPHQWSGI 5RER(DUWK
can be credited with creating, in a formal and systematic way, the idea of
cloud robotics.

Vernon, David. $UWL¿FLDO&RJQLWLYH6\VWHPV$3ULPHU&DPEULGJH0$0,7

Press, 2014. Chapter 9 offers a concise and formal overview of cooperating,
with a focus on social interaction, helping and being helped, collaboration,
MRLQWDFWLRQDQGDWWHQWLRQ

Vogel, Steven. Comparative Biomechanics: Life’s Physical World. 2nd

ed. 3ULQFHWRQ 1- 3ULQFHWRQ 8QLYHUVLW\ 3UHVV  :KHQ ZH WDON DERXW
extremes of size, we invoke the mathematics of scaling. In appendix 3, Vogel
gives a concise introduction to the essential mathematics and some of the
strange things that happen when nature starts to compensate for physics.

:DOWHU:*UH\³$Q,PLWDWLRQRI/LIH´6FLHQWL¿F$PHULFDQQR
±:DOWHUWDONVDERXW³HOHFWURPHFKDQLFDOHYROXWLRQ´:HPHHW(OVLHDQG
(OPHUKLV³V\QWKHWLFDQLPDOV´DQG:DOWHUGHVFULEHVWKHLUGHVLJQDQGEHKDYLRU

:HUIHO-XVWLQ.LUVWLQ3HWHUVRQDQG1DJSDO5DGKLND³'HVLJQLQJ&ROOHFWLYH
%HKDYLRU LQ D 7HUPLWH,QVSLUHG 5RERW &RQVWUXFWLRQ 7HDP´ Science
  ± 7KLV LV DQ H[FLWLQJ SURMHFW WKDW SXWV WKH FRQFHSW RI
stigmergy to work, showing that robots can leave behind a trail for one
DQRWKHUXVLQJWKHVWUXFWXUHWKDWWKH\DUHEXLOGLQJ²SOXVVRPHVLPSOHWUDI¿F
rules they enact that allow the group to coordinate its activities.

261
 Wood, Gaby. Edison’s Eve: A Magical History of the Quest for Mechanical
Life. 1HZ<RUN$OIUHG$.QRSI7KLVKLVWRU\UHPLQGVXVRIWKHGHHS
desire of humans to invent living machines. Wood brings to life the ancient
inventors, such as Descartes and Vaucanson, who, working in clockwork,
were laying the groundwork for modern robotics and trying to build
living machines.

:RRG5REHUW5DGKLND1DJSDODQG*X<HRQ:HL³7KH5REREHH3URMHFW
,V %XLOGLQJ )O\LQJ 5RERWV WKH 6L]H RI ,QVHFWV´ 6FLHQWL¿F $PHULFDQ 303,
no. 3 (2013). Wood and his colleagues explain the design and function
of Robobees.

;X<DQJVKHQJ+XLKDQ4LDQDQG;LQ\X:XHousehold Service Robotics.

6DQ'LHJR&$(OVHYLHU&KDSWHULQWURGXFHVWKHSURFHVVRIEXLOGLQJ
DPDSDQGWKHQXVLQJWKHPDSIRUSODQQLQJDSDWKDQGDYRLGLQJREMHFWV

Zaloga, Steven. Unmanned Aerial Vehicles: Robotic Air Warfare 1917–

20072[IRUG8.2VSUH\3XEOLVKLQJ7KLVLVDEULHIKLVWRU\RIGURQHV
ZLWKJUHDWSLFWXUHVDQGLOOXVWUDWLRQV,W¶VDOPRVWOLNHD¿HOGJXLGHWRKLVWRULF
UAVs—concise and with explanatory visuals.
Bibliography

262