See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330132363

Design, Development and Experimental Realization of a Quadrupedal Research

Platform: Stoch

Preprint · January 2019

CITATIONS READS

0 1,626

8 authors, including:

Dhaivat Dholakiya Shounak Bhattacharya

Arizona State University National University of Singapore
8 PUBLICATIONS   10 CITATIONS    10 PUBLICATIONS   12 CITATIONS   

SEE PROFILE SEE PROFILE

Ajay Gunalan Abhik Singla

Istituto Italiano di Tecnologia Indian Institute of Science
2 PUBLICATIONS   3 CITATIONS    11 PUBLICATIONS   18 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Walking robot View project

Cable-driven continuum robot View project

All content following this page was uploaded by Shounak Bhattacharya on 08 January 2019.

The user has requested enhancement of the downloaded file.

 Design, Development and Experimental Realization of a Quadrupedal
Research Platform: Stoch
Dhaivat Dholakiya, Shounak Bhattacharya, Ajay Gunalan, Abhik Singla,
Shalabh Bhatnagar, Bharadwaj Amrutur, Ashitava Ghosal, Shishir Kolathaya

Abstract— In this paper, we present a complete description

of the hardware design and control architecture of our custom
built quadruped robot, called the Stoch. Our goal is to realize
a robust, modular, and a reliable quadrupedal platform, using
which various locomotion behaviors are explored. This platform
arXiv:1901.00697v1 [cs.RO] 3 Jan 2019

enables us to explore different research problems in legged

locomotion, which use both traditional and learning based tech-
niques. We discuss the merits and limitations of the platform
in terms of exploitation of available behaviours, fast rapid
prototyping, reproduction and repair. Towards the end, we
will demonstrate trotting, bounding behaviors, and preliminary
results in turning. In addition, we will also show various gait
Fig. 1: Figure showing our custom built quadrupedal robot
transitions i.e., trot-to-turn and trot-to-bound behaviors.
Stoch, and its rendered CAD model in SOLIDWORKS
Keywords: Quadrupedal Robot, Legged Locomotion,
environment.
Robot Design

I. I NTRODUCTION With a view toward realizing state of the art walking

There have been multiple types of quadruped robots in controllers in a modular, and a low-cost walking platform,
varying sizes over the past ten years, namely, MIT Cheetah we developed a quadrupedal robot called the Stoch. This
[1], Starl ETH [2], Spot mini [3], ANYmal [4], LaikaGo [5], manuscript primarily serves as a detailed description of the
Minitaur [12] etc. Most of these works have mainly focused hardware and software design process and construction of
on answering crucial questions on control and hardware this platform (see Figure 1), which complements the results
design for walking, and have also demonstrated multiple that were shown in [14]. This hardware features modular
types of functionalities both in simulation and experiments. central body and a five bar co-axial leg design. Which are
Despite their impressive results, it is important to note that fabricated using easily available off the shelf materials like
these robots cost to the tune of minimum $30,000, are carbon-fiber tubes and custom 3D PLA parts. Carbon fiber
expensive to manufacture, and also require advanced sensing tubes, which are strong and light, serve as the structural
and actuation. members and leg links. Similarly, the 3D printed parts, man-
There are also quadrupeds that are less resource intensive ufactured in-house, act as the joints. The software consists
such as Tekken [6], Cheetah-cub [7], Onchila [8], Mini- of a real-time trajectory generator, which are then passed
cheetah [9], Aibo [10], Rhex [11], Open cat [13] some of as commands to the individual servo drives. The trajectory
which are open sourced, but not easy to reproduce. In the generator performs reasonably well at control frequencies on
current state of research in robot locomotion, a beginner many off-the-shelf servos (50 Hz). In addition to realizing
usually faces a stiff learning curve to design and fabricate a robust gaits in the real hardware, the central pattern generator
robust and a reliable quadrupedal walking platform. Crossing (CPG) based trajectory generator [15], [7], [8] also enables 1.
this barrier is not only limited by the individual’s capacity Realization of derived walking gaits like bound, walk, gallop,
but also by the requirement of resources. Thus, a robot which and turning in hardware with no additional training, and 2.
is simpler, modular, affordable, easily repairable, and yet Smooth transitions from one type of gait to another. This is
facilitating complex behaviors is largely preferred. mainly motivated by the various locomotion behaviors real-
ized by [16], [15] in their custom hardware. These various
This work is supported by the Robert Bosch Center for Cyber Physical behaviors are then composed together and tele-operated by
Systems, Bangalore, India
Dhaivat Dholakiya, Shounak Bhattacharya, Ajay Gunalan, Abhik a remote control device.
Singla are with the Robert Bosch Centre for Cyber-Physical Systems, The paper is organized as follows: Section II details the
IISc, Bangalore, India. E-mail: {dhaivatd, shounakb, ajayg, design and hardware of Stoch. Section III describes the
abhiksingla}@iisc.ac.in.
Ashitava Ghosal is with the Faculty of Mechanical Engineering, software architecture, on-board compute platform, actuators,
Bharadwaj Amrutur is with the Faculty of Electrical & Computer En- communication interfaces, and the sensory inputs. Section IV
gineering, Shalabh Bhatnagar is with Computer Science & Automa- contains a detailed description of the control algorithm used.
tion, and Shishir Kolathaya is an INSPIRE Faculty Fellow at the
Robert Bosch Center for Cyber Physical Systems, IISc, Bengaluru, India Finally Section V describes the experimental results of Stoch
{shalabh,amrutur,asitava,shishirk}@iisc.ac.in demonstrating multiple types of gaits and the associated
 Fig. 2: Left image shows the physical hardware of Stoch, and the right image shows its rendered exploded view.

transitions. flexion/extension in each joint. As shown in Figure 4, the

Stoch is 0.4 m long in total, with 0.25 m between front and
II. M ECHANICAL D ESIGN hind legs. It has a maximal hip height of 0.24 m with legs at
Stoch is a quadrupedal robot (see Figure 1) designed and maximum stretch, a standing height of 0.22 m, and an overall
developed in-house at Indian Institute of Science (IISc), height of 0.35 m from ground to the top of electronics and
Bangalore, India. The design philosophy behind Stoch is cables. The robot has 0.24 m of lateral spacing between left
based on modularity, lightweight construction, ease of man- and right leg planes axes, and a maximum width of 0.27 m.
ufacturing, rapid repair and reproduction. The robot weighs around 3.1 kg and 3.7 kg with the battery-
The robot design can be understood as an assembly of a pack. The estimated static centre of mass (COM) of the
central body module and four leg modules. Figure 2 shows given system (in standing configuration similar to Figure 4)
an exploded view showing all of the modules of Stoch. The is situated 35 mm below the hip axis.
leg modules can be used in several types of morphologies
as shown in Figure 3. The body consists of carbon fiber Parameter Value
total leg length 230 mm
(CF) hollow tubes along its length as strong and lightweight leg segment length 120 mm
structural members. The front and back CF plates along with min./max. hip joint -45◦ / 45◦
3D printed poly lactic acid (PLA) hubs form rigid central min./max. knee joint -70◦ / 70◦
max hip speed 461◦ /s
frame structure. The body also houses all electronic parts, max knee speed 461◦ /s
cable routing and battery power supply. A salient feature of max hip torque 29 kg-cm
this design is that the leg modules slide and ‘sit’ on CF tubes max knee torque 35 kg-cm
motor power (servo) 16 W
placed along the body length. Thus, the geometrical distance
between front and hind legs can be modified by changing the TABLE I: Hardware specifications of leg.
CF tube lengths or the assembly positions of the leg modules.
Similarly, the distance between the left and right legs, which B. Leg Design
is constrained by the CF plates, can easily be customized. Every leg has two degrees of freedom, one each for hip and
This feature enables user to rapidly play and experiment with knee. Detailed view of the leg assembly is shown in Figure 5.
the basic geometry of legs and body module. The key specifications of leg and actuators are summarized
In this section, first we discuss the overall geometry in Table I. The emphasis in the leg design was to keep the
of the mechanical structure, then focus on its individual inertia of the moving segments minimal. To achieve this,
components, i.e., body and leg, design. Finally, we describe all the actuators and transmissions are mounted to the main
the actuator and sensor arrangement on the legs. body. The leg mechanism is designed as a co-axial five bar
linkage[12]. This is beneficial to ensure fast swing motions
A. Geometry of Stoch with minimum leg inertia.
The Stoch is designed equivalent to the size of a miniature To fabricate the mechanism, the linkages are formed with
Pinscher dog. There are four identical legs and eight actua- carbon fiber hollow tubes connected by 3D printed poly
tors mounted in the robot, which are distributed equally. Each lactic acid (PLA) connectors. This arrangement enables us
leg has a hip and a knee joint, and the actuators provide to modify the link lengths with minimal effort. To further
Fig. 3: Modular Configurations of Stoch: (a) X configuration, (b) O configuration, (c) XO configuration, (d) Natural
Quadruped configuration.

Fig. 4: Physical robot dimensions of Stoch.

reduce the swing leg inertia, we’ve used CF tubes as journal achieved three main objectives using this architecture. First,
bearings and shaft at pivot joints instead of regular metal ball tele-operation of the robot, which is implemented via SSH
bearings. This drastically reduces the weight of the robot. over wireless communication (Figure 7). Second, the low-
level implementation of the walking controller, which passes
C. Actuation and Sensing Arrangement in Legs
on the desired joint configurations as commands to the joint
Figure 5 (i) and (ii) depicts the assembly arrangement actuators. And third, integrating sensory feedback with the
of servo actuators to the leg linkages and rotary encoders. walking controller. The last one is achieved by reading and
For the joint actuation we have used standard servo motors recording different sensors without experiencing significant
(Knee: Robokits 35 kg − cm High Torque Servo∗ , Hip: delay.
Kondo 2350HV Servo† ). The knee actuators are directly The encoders, servos and other parts selected for this work
coupled with the 3D printed links, which in turn move the are readily available, off-the-shelf components. This enabled
coupler links and eventually the knee joints of the legs. The a quick adoption, rapid prototyping while minimizing the
joint angle sensor (rotary encoder) for knee is connected total cost. Also, the library associated with the hardware is
with the linkage via a laser cut steel gear. Similarly, the readily available on GitHub, reducing the overhead associ-
hip actuator transfers mechanical power to the hip link via ated with firmware development.
a simple gear train (ratio 1 : 1), whereas the hip encoder
is directly coupled to monitor the hip servo angle via a 3D A. Tele-operation
printed part. The walking controller operates by changing internal pa-
III. E LECTRONIC S YSTEM A RCHITECTURE rameters of the Central Pattern Generator (CPG), such as the
frequency, phase difference and amplitude. These changes
In this section, we describe the electronic system architec-
result in generation of different behaviors and motions. In
ture in light of the objectives achieved in Stoch. The platform
this robot, these parameter changes, are made during the
∗ https://robokits.co.in/ process execution via SSH. We selected the keyboard as the
† https://kondo-robot.com/ input device for remote operation. This method of control
Fig. 5: (a) shows the physical leg assembly, (b) shows the rendered counterpart of the leg assembly. (b) has three parts: (i)
sensor mounting and actuation hub, (ii) coupling and transmission of linkages with actuators and rotary encoders, and (iii)
exploded view of knee joint and CF journals.

requires continuous process execution, without any block- position of each motor shaft is measured by externally
age. Thus, we have used non-blocking API’s from POSIX, coupled absolute magnetic encoders (Bourns EMS22a) at
namely, “select” to implement on Linux. Similar features can resolution of 1024 counts per revolution. Each encoder is
be obtained in Windows by using multi-threading. read sequentially by the RPi3 via a multi-duplex, SPI Bus.
B. Motor actuation The low-level drivers (in C) are used to pass the values to
the primary process (in Python) in real-time using shared
The actuators used in this robot are PWM enabled servo memory. In Comparison with other Inter-Process Communi-
motors. To actuate these servos, we have used Adafruit PWM cation (IPC) like socket, pipe, etc; the information doesn’t go
drivers (16-Channel 12-bit: PCA968), connected with the through the kernel, enabling direct access of virtual memory
central compute platform (Raspberry Pi3) over I2C Bus. In to both processes. This results in a low latency which is
this robot, there are two sets of servos (see section II-C). vital for controlling the robot. We used the limit switches
These servos operate at 11.1 V and 7.4 V respectively. for self-calibration and failure prevention at the joints. These
Equipment Details switches are connected to an analog to digital converter(4
Joint encoder Bourns EMS22a Channel 16 Bit) on the I2C Bus.
IMU MPU-9150
ADC ADS1115
PWM driver Adafruit PCA968
Servo Kondo 2350HV
Robokits RKI-1203 The center of mass of the robot, is estimated by the inertial
Computer Raspberry Pi 3b measurement unit (IMU). The states observed by the IMU
TABLE II: Electronics hardware specifications. are three axis acceleration, three axis angular velocity etc.
Using sensor fusion, the IMU can also estimate the roll,
C. Sensory feedback pitch, yaw of the robot. The IMU, MPU-9150, is placed
The sensory inputs contain signals from joint encoders, close to the geometric center of the chassis and is connected
inertial measurement unit and limit switches. The angular to the I2C bus.
 Fig. 6: Schematics of Electronics and communication network

IV. WALKING C ONTROLLER where, ω, ωd are the current and target oscillation frequency
of the legs, in radian per second. φi is the phase of the
In this section, we describe the design of the control
ith internal pattern generator. φ is the vector containing the
architecture of Stoch. The controller is based on the use
φi . Φi is the phase difference between the ith leg and the
of nonlinear coupled oscillators, called as Central Pattern
first leg. Ki,j are coupling constants. Equations 1 and (4)
Generator (CPG) [16], [17], [8]. This approach uses a group
represent low pass filter in time domain [16]. These filters
of polynomials associated with the coupled oscillators to
ensure there are no discontinuous changes in the variables.
generate the desired patterns at the end point. We will first
The low pass filter in equation 4 [17] is modified such that,
describe the methods used to develop the rhythmic patterns
periodic signal of any desired frequency can pass through but
using CPG, and then discuss inverse kinematics used to
undesired signal of the high frequency can not. In eqaution 4,
transform these patterns into desired joint angles.
X, Xd are the current and the desired state values and X0d =
∂Xd /∂φ. However, eqation 4 can only be used if sensory
A. Central pattern generator
feedback is enabled. When sensory feedback is disabled, we
Central pattern generators (CPGs) are neural networks directly feed the desired values to the inverse kinematics
capable of producing coordinated patterns of rhythmic ac- solver. The constants αω , αR are used to determine the
tivity without any rhythmic inputs from sensory feedback or corner frequency of these filters. ∆ is the discrete or event
from higher control centers [16]. In this work, the neural based input provided by the user or the sensors on-board.
network has been replaced with nonlinear coupled differen- In this work, the filters are used in places where a discrete
tial equations. When a signal is provided with oscillation signal is added by the user or by some feedback in response
frequency (ω), the CPG begins to produce rhythmic signals to an event.
of a predetermined pattern. The oscillation frequency updates
After the phase (φi ) is determined, any function described
the phases (φ) and the coupling ensures the phase offsets (Φ)
in form of φi can be used to create the endpoint behaviour
between the oscillators. This difference in phase offset results
of the ith leg. We have
in the generation of different types of gaits. The equations
for the CPG are given as

ω̇ = αω (ωd − ω) (1) Xd,i = Xd (φi ) = [x(φi ) y(φi )] (5)

N n
X
X
φ̇i = ω + Ki,j sin (φi − φj + Φi − Φj ) (2) x(φi ) = Wx,j,i fx,j (φi ) (6)
j=1 j=1
n
φi = mod(φi , 2π) (3) X
y(φi ) = Wy,j,i fy,j (φi ), (7)
Ẋ = φ̇X0d + αX (Xd − X) + ∆ (4) j=1
and their derivatives are given by each robot has a different reference frame associated with
legs, the joint angles are transformed to obtain motor angles
Ẋd,i = φ̇i X0d (φi ) = φ̇i [x0 (φi ) y 0 (φi )] (8) by using
n
X
0
ẋ(φi ) = φ̇i Wx,j,i fx,j (φi ) (9) Mhip = π − qhip , MKnee = −qknee ,
j=1
Xn where Mhip and MKnee are the motor signals generated
0 before transformation to a pulse width modulation (PWM)
ẏ(φi ) = φ̇i Wy,j,i fy,j (φi ) (10)
j=1 signal required for motor drivers.
In the above equations, x(φi ), y(φi ) represent the desired
end point position values of the leg from the hip joint
of the ith leg. This scheme can also be used to generate
r(φi ), θ(φi ) in polar co-ordinates which are used in the
reinforcement learning framework [14]. The constant vectors,
Wx,i,j , Wy,i,j , are the weights associated with the basis
functions, fx,j , fy,j . The basis functions can range from
kMPs [18], [14], trigonometric functions, Bezier polynomials
[19], time responses of any dynamical system, to even data
points. For example, a circular end point trajectory with
trigonometric functions fx,1 = fy,1 = sin φi and fx,2 =
fy,2 = cos φi would need a weight W = [Wx ; Wy ] =
[1 0; 0 1]. The patterns generated in Cartesian or in polar
co-ordinates are then transformed into joint co-ordinates via
inverse kinematics.
B. Inverse kinematics
Having obtained the coordinates (either in the form of
polar coordinates (r, θ) or Cartesian coordinates (x, y) of end Fig. 8: Inverse kinematics of robot leg
points), we generate the joint angles via inverse kinematics.
Figure 8 describes the leg configuration and the variables. V. E XPERIMENTAL R ESULTS
If the end points of the robot are described in Cartesian
In this section, we will describe various strategies under-
coordinates, then 11-14 are used to determine the joint angle.
taken to generate the different types of robot behaviors. First,
In case the description is in polar co-ordinate, equation11 can
we discuss the various types of flat ground walking gaits.
be avoided, since l3 and θ are directly provided. The inverse
Next, we describe the abduction free turning, and then finally
kinematics are given by
p y describe the transitions between these gaits.
l3 = x2 + y 2 , θ = arctan (11) 1) Flat ground walking: On the flat plane, we have
 2 x introduced 6 different types of gaits, trot, gallop, bound,
2 2

l − l1 − l2 walk, modified trot 1 and modified trot 2. We have used the
Φ = arccos 3 (12)
2l1 l2 basis function based approach [14] to store the information
 
l2 sin Φ about the trajectories in a few constant values. These constant
ζ = arctan (13)
l1 + l2 cos Φ values are stored in the weight matrix (Wx , Wy ). Using the
qhip = θ + ζ, qknee = Φ + qhip , (14) weights, we can obtain the desired end point trajectories and
the velocities. The target end point trajectory for the trot gait
where, l1 , l2 are the hip and knee lengths of the robot. Since is shown in Fig.9. We get
5
X
xi = fkM P,j (φi )Wx,i,j (15)
j=0
5
X
yi = fkM P,j (φi )Wy,i,j , (16)
j=0

where, xi , yi are the target x, y coordinates of the ith leg.

fkM P,j (φi ) is the j th kinematic motion primitive [14], which
essentially is a polynomial of φi .
Table III presents the comparison of speeds obtained
for various gaits. A more detailed description of the basis
functions are given in [14].
Fig. 7: Controller and teleoperation schematic
 −0.16 −0.16
shown. Further, the animation and physical assembly is also
shown in the video for easy understating of the Stoch design.
−0.18 −0.18
Y (m)

Y (m)
VI. CONCLUSION
−0.2 −0.2
This work presents a custom built quadruped robot along
−0.22 Front left −0.22 Hind left with its design, software, control framework and experimen-
−0.05 0 0.05 0.1 −0.05 0 0.05 0.1 tal validation on the hardware. Compared to other existing
X (m) X (m)
−0.16 −0.16 quadruped robots, our platform requires less resource and
costs under USD 1000. Several gaits were realized such as
−0.18 −0.18 walk, trot, gallop, and bound. Additionally, the robot can turn
Y (m)

Y (m)
with a small radius via CPG based gait transitions, without
−0.2 −0.2
the use of abduction. The robot reached a maximum forward
speed of 0.6 m/s. An open-loop controller was used to set
−0.22 Front right −0.22 Hind right
−0.05 0 0.05 0.1 −0.05 0 0.05 0.1
the speed and direction in all these experiments. Future work
X (m) X (m) involves increasing the payload, and also incorporate external
Fig. 9: Endpoint trajectories learned from the reinforcement sensors for closed-loop control of the quadruped.
learning based training[14] ACKNOWLEDGMENT
Gait type Trot Walk Gallop Bound Modified Trot 1 Modified Trot 2
Speed (m/s) 0.60 0.51 0.51 0.55 0.62 0.59 We acknowledge Ashish Joglekar, Rokesh Laishram and
TABLE III: Walking speed of Stoch obtained during different Balachandra Hegde for help in software development, Car-
gait experiments. The last two columns are the result of bon fibre component manufacturing and PCB design.
reshaping of the end point trajectories of trot. All the gaits R EFERENCES
are shown in the video.
[1] H.-W. Park, S. Park, and S. Kim, “Variable-speed quadrupedal bound-
ing using impulse planning: Untethered high-speed 3d running of
2) Turning: The current model of Stoch, does not contain mit cheetah 2,” in Robotics and automation (ICRA), 2015 IEEE
international conference on. IEEE, 2015, pp. 5163–5170.
any abduction. This limits us to realization of abduction [2] M. Hutter, C. Gehring, M. Bloesch, M. A. Hoepflinger, C. D. Remy,
free/locked turning. We set the X axis (forward-backward) and R. Siegwart, “StarlETH: A compliant quadrupedal robot for fast,
movement of the desired leg to zero, and use efficient, and versatile locomotion,” in Adaptive Mobile Robotics.
World Scientific, 2012, pp. 483–490.
xturn,i = αt,i xi (17) [3] “SpotMini robot,” https://www.bostondynamics.com/spot-mini/.
 [4] M. Hutter, C. Gehring, D. Jud, A. Lauber, C. D. Bellicoso, V. Tsounis,
J. Hwangbo, K. Bodie, P. Fankhauser, M. Bloesch et al., “Anymal-a
[1, 0, 0, 1]
 left turn
highly mobile and dynamic quadrupedal robot,” in 2016 IEEE/RSJ
αt = [0, 1, 1, 0] right turn (18) International Conference on Intelligent Robots and Systems (IROS).
IEEE, 2016, pp. 38–44.

[1, 1, 1, 1] no turn

[5] “Laikago robot,” http://www.unitree.cc/.
[6] H. Kimura and Y. Fukuoka, “Biologically inspired adaptive dynamic
where αt is the vector containing the turning coefficients and walking in outdoor environment using a self-contained quadruped
xturn,i is the target x axis position of the ith leg. To avoid robot:’Tekken2’,” in Intelligent Robots and Systems, 2004.(IROS
discontinuous changes in the motion, we pass the αt through 2004). Proceedings. 2004 IEEE/RSJ International Conference on,
vol. 1. IEEE, 2004, pp. 986–991.
a low pass filter, equation 1. This results in a smooth trot-turn [7] A. Spröwitz, A. Tuleu, M. Vespignani, M. Ajallooeian, E. Badri, and
transition. A. J. Ijspeert, “Towards dynamic trot gait locomotion: Design, control,
3) Gait transition: In this section, we will provide a gen- and experiments with cheetah-cub, a compliant quadruped robot,” The
International Journal of Robotics Research, vol. 32, no. 8, pp. 932–
eral methodology for transitioning between the flat-ground 950, 2013.
walking gaits. In section IV-A, we described the phase offset, [8] A. T. Sprowitz, A. Tuleu, M. Ajaoolleian, M. Vespignani, R. Moeckel,
Φ, as a constant set of values which can be changed to obtain P. Eckert, M. D’Haene, J. Degrave, A. Nordmann, B. Schrauwen et al.,
“Oncilla robot: a versatile open-source quadruped research robot with
various gaits. This change can be achieved, on the fly, by compliant pantograph legs,” Frontiers in Robotics and AI, vol. 5, p. 67,
passing the desired phase offset through a low pass filter: 2018.
[9] W. Bosworth, S. Kim, and N. Hogan, “The mit super mini cheetah:
Φ̇i = αΦ (Φd,i − Φ) (19) A small, low-cost quadrupedal robot for dynamic locomotion,” in
Safety, Security, and Rescue Robotics (SSRR), 2015 IEEE International
where Φd,i is the desired phase offset of the ith leg, and Φi Symposium on. IEEE, 2015, pp. 1–8.
[10] “AIBO robot,” https://us.aibo.com/.
is the current phase offset of the ith leg. This allows for a [11] Rhex, “RHex robot,” https://www.bostondynamics.com/rhex/.
smooth change in the phase values. [12] A. De and D. E. Koditschek, “Vertical hopper compositions for preflex-
ive and feedback-stabilized quadrupedal bounding, pacing, pronking,
A. Video Result and trotting,” The International Journal of Robotics Research, vol. 37,
no. 7, pp. 743–778, 2018.
Video results of walking experiments on the Stoch [13] “OpenCat robot,” https://www.hackster.io/petoi/opencat-845129.
are available on https://youtu.be/Wxx9pwwTIL4. [14] A. Singla, S. Bhattacharya, D. Dholakiya, S. Bhatnagar, A. Ghosal,
B. Amrutur, and S. Kolathaya, “Realizing learned quadruped locomo-
Specifically, we show trot (Fig. 10), bound, gallop, walk gaits tion behaviors through kinematic motion primitives,” arXiv preprint
on the robot. Additionally turning and gait transitions are also arXiv:1810.03842, 2018.
 Fig. 10: Trot sequence of Stoch on flat terrain

[15] A. T. Sprowitz, A. Tuleu, A. J. Ijspeert et al., “Kinematic primitives

for walking and trotting gaits of a quadruped robot with compliant
legs,” Frontiers in computational neuroscience, vol. 8, p. 27, 2014.
[16] A. J. Ijspeert, “Central pattern generators for locomotion control in
animals and robots: a review,” Neural networks, vol. 21, no. 4, pp.
642–653, 2008.
[17] M. Ajallooeian, J. van den Kieboom, A. Mukovskiy, M. A. Giese, and
A. J. Ijspeert, “A general family of morphed nonlinear phase oscillators
with arbitrary limit cycle shape,” Physica D: Nonlinear Phenomena,
vol. 263, pp. 41–56, 2013.
[18] F. L. Moro, A. Spröwitz, A. Tuleu, M. Vespignani, N. G. Tsagarakis,
A. J. Ijspeert, and D. G. Caldwell, “Horse-like walking, trotting, and
galloping derived from kinematic motion primitives (kmps) and their
application to walk/trot transitions in a compliant quadruped robot,”
Biological cybernetics, vol. 107, no. 3, pp. 309–320, 2013.
[19] G. A. Castillo, B. Weng, A. Hereid, and W. Zhang, “Reinforcement
learning meets hybrid zero dynamics: A case study for RABBIT,”
arXiv preprint arXiv:1810.01977, 2018.

View publication stats