Control Systems for Robots

Prof. Robert Marmelstein

CPSC 527 – Robotics
Spring 2010
 Introduction to Robot Control

 We want robots to do things that might otherwise be

done by intelligent, physical beings
 Biological nervous systems are often thought of as
control systems for living organisms
 To control a robot in this manner, we need to:
 Define the state of the robot
 Sense the state of the robot
 Compare the state of the robot to a desired state
 Decide what actions need to be taken to achieve the desired
state and issue the associated commands
 Translate those command into physical action
 Measure the effect of those physical actions
 Robot State Description

 Physical States:
 Position
 Orientation (pose)
 Velocity Assumed vs. Actual
 Acceleration
 Sensor states
 Actuator States
 Internal States:
 Plans
 Tasks
 Behaviors
 Control Engineering

 Control engineering is the application of

mathematical techniques to the design of algorithms
and devices to control processes or pieces of
machinery.
 It almost always requires a model of the entity that is
being controlled
 If a system can be modeled by a set of linear
differential equations there are well understood
techniques for getting exact analytical solutions, and
so designing controllers so that the output of the
system is the required one
 Example: Forced Spring
 A forced spring can be modeled by a linear differential equation

k
m

b
x

d 2x dx
F (t )  m 2  b  kx
dt dt
 Real-World Systems

 Unfortunately, most real-world

systems are non-linear in nature
 Example: Pendulum

 In these cases, the nonlinear system is

often approximated by a linear system
 For the Pendulum, assume sin() ≈ ,
which yields:
 Control System Components
 Target Value – The desired operating point of the overall
system, which is speed.
 Measured Value – The actual operating point of the system. It
is affected by external factors such as hills, and internal factors
such as the amount of fuel delivered to the engine.
 Difference Value – This is the difference between the target
value and the measured value. Translates into feedback.
 Control Input – This is the main adjusting point of the control
system. The amount of fuel delivered to the engine is the
primary control input to the cruise control
 Control Algorithm – Determines how to best regulate the control
input to make the difference value as close to zero as
possible. It does this by periodically looking at the difference
value and adjusting the control input
 Feedback Defined
 Feedback: continuous monitoring of the sensors and reacting
to their changes.
 Feedback control = self-regulation
 Two kinds of feedback: positive and negative
 Negative feedback acts to regulate the state/output of the
system
 e.g., if too high, turn down, if too low, turn up
 thermostats, toilets, bodies, robots...
 Positive feedback acts to amplify the state/output of the
system
 e.g., the more there is, the more is added
 lynch mobs, stock market, ant trails...
 Often results in system instability
 Open Loop Controller

 The Open Loop Controller (OLC) is the simplest kind

 The controller sends an input signal to the plant
 It does not compensate for disturbances that occur after the
control stage
 Actual effects are assumed – not measured
 No feedback to match actual to intended
 Open Loop Controller (cont.)

 The OLC is commonly used for behavior-based

systems
 If a trigger condition is met , then the behavior is activated
 Behavior is performed until the condition is no longer met
 If the condition is not met, the (assumedly) some other
behavior is activated
 You would likely use an OLC if you have no way of
measuring your operating point (e.g., the value you
are trying to control)
 Problem with Open Loop Controllers

 The effectiveness of OLCs are very context

dependent
 The amount of force that is applied has different
effects dependent on the surface type
 Closed Loop Controller

 In Closed-loop control, the output is sensed and compared with

the reference. The resulting error signal is fed back to the
controller [feedback].
 Components:
 Reference – Desired State
 Controller – Issues Commands
 Plant – Actuator
 Closed Loop Controller (cont.)
 [Negative] Feedback keeps the operation of the system smooth and
stable

 Closed Loop Controller issues:

 How quickly will the system respond to error?
 How long will it take the system to reach equilibrium?
 What, if any, residual error will remain?
 Under what conditions will the system become unstable?
 Example of a Closed Loop Controller

Input Error Output

Computation Actuator Control Velocity
+ (Brain) (Auto Pedals) (Engine Power)
- Actual
Desired
Position Position

Sensing
(Eyes)
Velocity

Time
Velocity decreases as the
car gets closer to the
desired position

Velocity
 Bang-Bang Controller (BBC)

 The simplest type of closed-loop controller is the Bang-

Bang controller. It typically consists of two states and a
target value
 The BBC typically monitors one item (quantity) of interest
—its job is to keep that quantity at a certain target value

temp > target

 If the quantity is too high or
low (vs. the target) the BBC
compensates to change it
 The system continually OFF ON
transitions between states,
often abruptly
temp < target
 Taking the Edge off Bang-Bang Control
with Hysteresis

 Hysteresis provides a sort of "guard

band" around the desired set point of
the system.
 In other words, when the temperature goes above (or
below) the desired control point, there is a margin
which needs to be exceeded before compensation is
applied
 The result a lag which causes the system to run much
smoother, avoiding the jerkiness of a purely Bang-
Bang controller
 Leveraging Hysteresis

Single Threshold System Use of temperature dead zone

(no hysteresis) to induce hysteresis
 Closed Loop Controller Issues

JLJ Text – Fig 2.5

Low Gain – Sluggish High Gain – Unstable

 Proportional Integral Derivative (PID)
Controller
 Closed loop controllers that only use
proportional control can easily become
unstable if the gain is too high or sluggish
if the gain is set too low
 PID controllers help solve this problem
 It use the measured error compute an
input to the Plant based on three distinct
controls: Proportional, Integral and
Derivative (see right)
 Proportional control – Computed based on the actual error (times
a gain factor). Thus, The larger the error, the bigger correction
the control will make
 Serves to control response time to error
 For high gains or large errors, tends to overshoot and oscillate the
desired output
 There is typically a steady state error that cannot be corrected
 Proportional Integral Derivative (PID)
Controller
 Integral control – Reduces steady-
state error by adding (integrating) the
actual errors over time.
 Once the error reaches a predetermined
threshold, the controller will compensates
 Too little can result in undershoot; too

much can result in overshoot

 Derivative (D) control – Computed based of difference between
current and previous error. Thus, the output of this control is
proportional to the change in error
 Prevents oscillations due to overshoot
 Reduced settling time by giving a better dynamical response
 Generally, this control has a positive effect
 PID Controller (cont.)
 The tunable factors are:
 Kp – Proportional Gain factor
 KI – Integral Gain Factor
 KD – Differential Gain Factor
 These factors are cross-coupled, so the performance of the
system cannot be optimized by tuning each factor independently
 Some systems can be engineered without all three PID
components
 The P components is always required, but P controllers alone can
result in instability
 PI is not accurate but converges quickly
 PD converges relatively quickly reducing oscillations as it
approaches the goal.
 PID accurately maintains a position, but isn’t very fast.
 Helpful PID Terms
 Gain(s) --The parameter(s) that determine the magnitude of the
system’s response.
 Gain values determine whether or not the system stabilizes or
fluctuates.
 Finding effective gains is a trial and error process, requiring testing
and recalibration.
 Proportional Gain – When the value of the gain is proportional
to the error.
 Damping – The process of systematically decreasing a system’s
fluctuations
 A system is damped if it does not oscillate out of control.
 Generally, the gains have to be adjusted for a system to be
damped
 Steady State Error – The amount of error that remains after the
system has reached equilibrium
PID Controller (cont.)

• KP – Proportional Gain
• KI – Integral Gain
• KD – Derivative Gain
 PID Controller Response Curve

Controlled
variable
Overshoot
Steady state error
%
Reference

Transient State Steady State

Time
Settling time
 PID Response Curve (cont.)

 Rise Time (Tr) – The time for the plant output y to rise beyond 90%
of the desired level for the first time
 Overshoot – How much the peak level is higher than the steady
state, normalized against the steady state
 The time required for the output to reach its maximum level is called the
Peak Time (Tp)
 Settling Time (Ts) – The time it takes for the system to converge to
its steady state
 Transient State – The period from the detection of error until its
approximate correction, resulting in the steady state
 Steady-state Error – The difference between the steady-state output
and the desired output.
Effect of Increasing PID Factors

KP KI KD

NT: No trend
Control Performance – Proportional
Control Source: CUNY – Dr. Jizhong Xiao

KP = 20 KP = 50

KP = 200
KP = 500
 Control Performance – Integral Control
(KP = 100)

KI = 50 KI = 200

Source: CUNY – Dr. Jizhong Xiao

Control Performance – Derivative Control
(KP = 100, KI =200) Source: CUNY – Dr. Jizhong Xiao

KD = 2 KD = 5

KD = 10 KD = 20
 Optimizing Performance

 PID Tuning – By Hand

 Boost Kp until it oscillates
 Boost KD to stop oscillation, back off Kp by 10%
 Dial in KI to Hold position or velocity smooth
 Trial and error
 PID tuning – By Design
 Zeigler-Nichols Method (next slide)
 Other:
 Work to minimize environmental interference and sensor
error (two are typically coupled)
 Smart design helps too
 Zeigler-Nichols Tuning Rule for PID
Controllers

Yields ~25%
overshoot and good
settling time
 Why Care about the PID Controller?

 Because PID Controllers are everywhere!

 Due to its simplicity and excellent if not optimal performance in
many applications, PID controllers are used in more than 95%
of closed-loop industrial processes.
 It can be tuned by operators without extensive background in
Controls, unlike many other modern controllers that are much
more complex but often provide only marginal improvement.
 In fact, most PID controllers are tuned on-site.
 The lengthy calculations for an initial guess of PID parameters
can often be circumvented if we know a few useful tuning rules.
This is especially useful when the system is unknown
 Non-Linear Control

 Linear controllers are generally valid only over small

operational ranges.
 Hard non-linearity cannot even be approximated by
linear systems.
 Model uncertainties can sometime be tolerated by
the addition of non-linear terms.
 Non-linear systems often have multiple equilibrium
points, plus periodic, or chaotic attractors.
 In these systems, small disturbances (even noise)
can induce radically different behaviors.
 Control System Non-Linearity Issues

 Saturation – Occurs when the

input signal to a certain device
exceeds the ability of the device
to process it
 Input – sensors
 Output – motors
 For output, saturation means that the required
compensation can no longer be applied to the control
system
 In general, it is good practice to limit the signal to the
saturation value in software
 When an input reaches the saturation point, it no longer
provides a reliable estimate of the real world
 Non-Linearity Issues (cont.)

 Dead Zone – Because the sensitivity

of actuators is limited, not every
non-zero input will result in action.
The Dead Zone is the +/- region
above a zero (0) input that will
result in no actuator movement.

 Backlash – Term describing actuator

hesitation and overshoot caused by
small gaps between motor gears
 Can result in small, but unnecessary,
oscillations of the actuator position
 Self-Balancing Robot Lab

 Problem – Create a PID controller that will keep an

NXT robot balanced on two wheels
 Use a light sensor to determine balance:
 Light diminishes as it tilts backward
 Light gets stronger as it tilts forward
 Collect reference value when the robot is perfectly balanced
 PID controller will use the light sensor data to
compute the compensation for the motors (plant)
 Move backward if tilting back
 Move forward if tilting forward
Self-Balancing Robot (Pics)

Light Sensor
 [Simplified] PID Controller Algorithm

// Get Balance Reading for Light Sensor

midVal = Sensor_Read(LS_Port);

// Now Compute Power (PID) Value for Balancing Robot

while (true) {
lightVal = Sensor_Read(LS_Port);
error = lightVal – midVal;
diffErr = error – prevError;
intErr = intErr + error;
pVal = kp * error;
iVal = (ki*intErr)*dampFactor;
dVal = kd*diffErr;
prevErr = error;
power = (pVal+iVal+dVal)*scale;
MotorFWD(motorPorts, power);
}
 Other Material

 PID Controller for Dead Reckoning

 CMU PID Tutorial
 DePaul Self-Balancing Robots – Trials and
Tribulations
Questions?