Sådhanå (2018) 43:102 Ó Indian Academy of Sciences

https://doi.org/10.1007/s12046-018-0918-7 Sadhana(0123456789().,-volV)FT3
](0123456789().,-volV)

Development of a compliant legged quadruped robot

M M GOR1,*, P M PATHAK1, A K SAMANTARAY2, K ALAM1, P KUMAR1, D ANAND1,
P VIJAY1, R SARKAR1, J-M YANG3 and S W KWAK4
1
Indian Institute of Technology Roorkee, Roorkee, India
2
Indian Institute of Technology Kharagpur, Kharagpur, India
3
Kyungpook National University, Daegu, South Korea
4
Keimyung University, Daegu, South Korea
e-mail: [email protected]; [email protected]; [email protected];
[email protected]; [email protected]; [email protected]; [email protected];
[email protected]; [email protected]; [email protected]

MS received 19 May 2016; revised 11 November 2017; accepted 1 June 2018; published online 18 June 2018

Abstract. This paper presents the detailed steps for design and development of a compliant legged fault
tolerant quadruped robot where each leg has two links and two motorized revolute joints for locomotion. The
body and upper links of legs are rigid whereas the lower link of each leg is compliant. Amble gait is demon-
strated on the developed robot. Safety and reliability are the most critical issues for the quadruped robot. During
the failure of any joint, performance of quadruped robot is affected. In this paper, locked joint failure is also
discussed. Strategies for fault tolerant control of the quadruped are developed and experimentally validated. The
developed robot can be used for various hardware-in-the-loop controller prototyping such as reconfiguration,
fault tolerant control, and posture control, etc. pertaining to quadruped robots.

Keywords. Quadruped robot; compliant leg; locked joint failure.

1. Introduction joints in the flexion/extension plane of the leg. The hip

abduction/adduction joint was actuated by brushless elec-
Quadruped robots have many practical advantages over tric motor which provides constant output torque. LittleDog
other mobile robots. However, development and operation [4, 5] has 12 DOFs and each joint is operated by high gain
of working model of a quadruped robot is not easy. More servo motors. Sensors mounted on the robot measure body
and more quadrupeds are developed now-a-days. Some of orientation, joint angles and ground-foot contact. Sensing,
them are Baby Elephant [1], Cheetah-cub [2], HyQ [3], communication and actuators are controlled by an onboard
LittelDog [4, 5] and Tekken [6, 7]. Main intensions behind computer. Tekken [6, 7] is a light weight (4.3 kg) manually
their development were to explore their use in hazardous operated compliant legged quadruped robot. It has 16 DOFs
and unstructured environments. with 3 joints around pitch axis (ankle, knee and hip) and 1
Hydraulically actuated Baby Elephant [1] was designed hip joint around yaw axis at each leg.
to work as mechanical carrier. It has compliant legs and 12 Locked joint failure is a kinematic failure under which a
degrees-of-freedom (DOFs). In [1], multi-body dynamic joint cannot move and is locked in place. However, body
simulation software was used to compare experimental supporting ability is preserved. Due to this characteristic
results with simulations. Electrically actuated Cheetah-cub feature, the failed leg still contributes partially towards the
[2] was designed for fast locomotion. Cheetah-cub’s leg has robot locomotion. The locked joint failure is very common
spring loaded pantograph mechanism with multiple seg- and occurs due to friction in the bearings, slipping of
ments. Robot’s self-stabilizing properties was demonstrated clutches, broken gear teeth jamming the motor, and
in hardware model and in simulation carried out in Webots misalignment/bending on the rotor shaft. In this situation,
software. HyQ [3] has 12 DOFs and it was designed to locomotion can be carried out by either implementing fault
perform highly dynamic tasks like jumping and running. It tolerant gait pattern or applying control laws to the
has both the hydraulic and electric actuation system. During redundant hardware. Different fault tolerant gaits have been
running and jumping, generated impact forces were suggested by Lee [8] and Yang [9–11] for locked joint
absorbed by hydraulic actuation mounted on hip and knee failures. Fault accommodation through reconfiguration is
accomplished by on-the-fly deployment/activation of
*For correspondence alternate standby devices, called hardware redundancies, in

1
102 Page 2 of 18 Sådhanå (2018) 43:102

results. Finally, section 4 provides conclusions and draws

perspectives.

2. Development of quadruped robot

This section discusses the quadruped specification and

design studies. As per our design objective, the quadruped
Figure 1. Compliant link: (a) without spring, (b) with spring. should be able to walk with different static and dynamic
gait patterns. At the initial stage it is difficult to decide the
physical specification. Design process was initiated from
the commercially available components size and specifi-
cations. As the robot is compliant legged, we start from a
commercially available compliant link.

2.1 Compliant link

As natural evolution, humans and animals have legs with
spring like behavior that enables them to absorb shock on
impact, maintain stability, and also to initiate sudden
motion. Generally, mechanical springs are used to intro-
duce compliance in the structure of robot [7, 16]. Com-
pliance in the link improves locomotion of quadruped
robot. But over compliance reduces locomotion speed and
Figure 2. (a) Common kinematic structure of quadruped, (b) increases posture disturbance. So, optimizing the compli-
kinematic structure of developed quadruped. ance is required for a specific robot configuration. Such a
design is facilitated by simulating the model of quadruped
[13, 14] with some initial assumptions of link and body
the place of faulty actuators and adapting control laws to parameters and then evolving the design through parameter
the modified system’s architecture [12]. variations. Number of simulations can be carried out by
This paper introduces the quadruped robot, developed as varying the stiffness and resistance, and its values can be
a part of Indo-Korea joint research project at Robotics and finalized for maximum locomotion speed and minimum
Control Lab of Indian Institute of Technology Roorkee, posture disturbance of robot. This optimum stiffness and
India. It is an electrically actuated compliant legged load coming on each leg becomes the key parameters for
quadruped robot with two DOFs per leg. Main objective designing the spring used in the compliant link. Number of
behind this development is to explore untouched research turn of spring ‘n’ can be decided as,
areas in quadruped robots such as reconfiguration, fault
tolerant control, posture control, etc. The design is based on dGd
n¼ ð1Þ
extensive modeling and simulation studies published earlier 8WC 3
[13–15]. In this paper, fault tolerant control through
where, d is axial deflection derived from load(W)/stiff-
reconfiguration is proposed. A novel robot-on-robot type
ness(K), G is the modulus of rigidity of spring material, d is
moving appendage, which is realized in the form of a two
wire diameter, C is spring index derived from coil diameter
DOF robot with a payload at its end, is proposed. This
(D)/wire diameter(d). The robot developed in this research
appendage device may serve as a redundant hardware,
uses a commercially available prismatic link as shown in
which is to be activated only when the base quadruped
figure 1(a). Internal hydraulic pressure provides damping
robot experiences a locked joint failure.
and a spring is used to produce compliance in the link as
This paper is organized as follows. In section 2, the entire
shown in figure 1(b) with the above discussed design
process of development of the quadruped robot is pre-
considerations.
sented. It covers details about the compliant link, kinematic
structure, joint torque estimation, actuation system, con-
trollers and the fabrication process. Section 2 also presents
the experimental results when the robot moves in amble
2.2 Kinematic structure
gait. In section 3, fault tolerant control through reconfigu- The kinematic structure of a quadruped robot should be
ration by using hardware redundancy is discussed. This is simple and yet allow the robot to perform a wide range of
further validated through simulation and experimental tasks. As shown in figure 2(a), the hip and knee joints in the
Sådhanå (2018) 43:102 Page 3 of 18 102

Figure 3. Leg configuration: (a) forwardforward, (b) backward-backward, (c) forward-backward, (d) backward-forward.

Figure 4. The four running states of bounding gait [15].

backward-forward configuration is most beneficial one as it

reduces slippage between feet and ground and improves the
stability of motion. Therefore, backward-forward leg con-
figuration is adopted in this research.

2.3 Joint torque estimation

Before finalizing size and type of actuation system it is
necessary to estimate the joint torque that is required for
desired tasks and the joint motors are to be selected based
on the estimated torque demand on those. For torque esti-
mation, a four-legged robot model with bounding gait
presented in [15] is utilized. Its four phases, i.e., back
Figure 5. Sagittal plane model of quadruped [15]. stance, flight after back stance, front stance and flight after
front stance which are shown in figure 4 [15].
Figure 5 shows sagittal plane model of quadruped robot
[15] which is considered for the dynamic model develop-
sagittal plane and one hip joint on the vertical perpendicular ment. All links and body are assumed to be slender and
plane is a common kinematic design structure. Thus, there uniform. Nomenclatures used in this paper are listed at the
are 3 DOFs per leg. However, as an initial attempt, the end of paper. Torque at each joint is calculated by using the
robot is constructed with 2 DOFs per leg whose corre- equations shown in ‘‘Appendix A’’. Movement starts from
sponding kinematic structure is shown in figure 2(b). Pro- the initial position where it is assumed that each joint has
vision is kept in the design to include the third degree of position 0 rad, velocity 0 rad/s and acceleration 3 rad/s2. It
freedom in the future. is assumed that the robot has to move on a maximum of
Legs may be attached with the body in different ways as 0.25 rad sloped surface. Torque values obtained for the said
shown in figure 3. If knee joint points to the front of the movement are 5.83 Nm, 2.81 Nm, 6.38 Nm and 8.37 Nm
robot then it is termed as forward configuration and if it for joints 1, 2, 3 and 4, respectively. Estimating the higher
points to the back of the robot then it is termed as backward side of torque is a safe part of the design process. Con-
configuration. Thus, four possible configurations are for- sidering a factor of safety, the estimated torque is finalized
ward-forward, backward-backward, forward-backward and as 15 Nm. Thus, actuation system must be able to produce
backward-forward. It is reported in literature [17] that 15 Nm torque.
102 Page 4 of 18 Sådhanå (2018) 43:102

Table 1. Major specifications of Maxon motor unit [23]. [20] in which each leg made up of four bar link which is
actuated by three single acting pneumatic cylinders and a
Parameter Value wheel attached at the leg tip is actuated by electric actua-
Brushless EC-4 pole motor (Maxon) tion. Pneumatic actuation system does not have the problem
Nominal voltage 48 V of oil leakage and oil flammability, and there is no need of
No load speed 16,500 rpm return lines for the fluid. However pneumatic actuation is
No load current 422 mA not the best choice for lower power-to-weight ratio,
Nominal speed 15,800 rpm dynamic locomotion and quick response (due to com-
Nominal torque (max. continuous torque) 118 m Nm pressibility of air).
Stall torque 3430 m Nm Electric actuation system is most commonlyused in
Max. efficiency 89% robots. Now-a-days, electric motors are available in big
Encoder with Line Driver (Maxon)
range of sizes and performance; they are more compact and
Counts per turn 1000
economical. More and more quadrupeds are using electric
Number of channel 3
Planetary gearhead GP42C (Maxon) actuation system. Some of them are Scout-II [21], Tekken
Reduction 230:1 II [7], KOLT [16] and Cheetah-cub [2]. In case of electric
Number of stages 4 actuation system, compliance in the legs is realized with
Maximum continuous torque 15 Nm springs for energy efficient locomotion. Compliance in the
legs also helps to protect the gear unit of electric actuator
from excessive impact force.
In the present work, electric actuation system is used.
Several kinds of electric motors are available and they are
distinguished by current profile (AC/DC), number of coils,
their configuration and type of synchronization. There are
AC motors and brushed and brushless DC motors. It is
found that brushless DC motors are most suitable for
robotics application as they offer high speed, torque and
efficiency, and low maintenance in comparison to other
motors. Brushless DC motors have electronically con-
trolled commutation systems. Due to absence of brushes,
they are faster and more efficient because brushes create
friction. Also, maintenance is less as there is no need to
periodically replace brushes as in a brushed motor.
However, brushless DC motors are costlier and their
controllers are more complicated [22]. Looking at the
Figure 6. Electric actuation system for quadruped robot [23].
positive side of brushless DC motors, brushless DC
motors made by Maxon are used in this research. Motor
unit comprising motor, gear head and encoder are chosen
in such a way so that the entire unit can supply torque as
2.4 Actuation system estimated in previous section. Table 1 shows the major
specifications of the Maxon motor and controller unit.
Hydraulic, pneumatic and electric actuations are usually
Figure 6 shows Maxon motor, gear head and encoder
used in the robots. Each actuation system has its own merits
parts separately and also as a combined unit.
and demerits.
Quadruped robot developed by Raibert [18], Titan XI
[19] and Baby Elephant [1] are actuated by hydraulic
actuation system. HyQ [3] is also partially actuated by
2.5 Controllers
hydraulic system. Hydraulic actuation system generally Controllers are the brain of the entire quadruped system. To
offers high force. Due to low compressibility of fluid, maintain better compatibility with the actuation system,
hydraulic actuators also give a quick response. However, Maxon controller has been used. For the position control, a
there are drawbacks of hydraulic actuation system such as discrete PID controller with anti-windup, acceleration feed
oil leakages, change of oil viscosity with temperature, and forward and velocity feed forward is implemented using a
non-linear characteristics of the hydraulic system leading to digital signal processor (DSP) where the sample time (TS)
difficulty in control system development. is taken as 1 ms, which is much smaller than the mechan-
Pneumatic actuation is also a choice for quadruped ical time constant of a typical drive system. As per Maxon
robots. Hirose laboratory of the Tokyo Institute of Tech- documentation, compatible controller for selected drive is
nology has developed a hybrid quadruped robot Airhopper EPOS2 P 24/5, which is a freely programmable positioning
Sådhanå (2018) 43:102 Page 5 of 18 102

Table 2. Major specification of Maxon controller unit [23]. 2.6 Sensors

Parameter Value Information about body’s angular and translational
motion is obtained by MPU 6050 sensor. The InvenSense
EPOS2 P 24/5 Controller (Maxon) MPU-6050 [24] sensor contains a MEMS accelerometer
Operating voltage 11–24 VDC
and a MEMS gyro in a single chip. It is very accurate, as
Digital inputs 6
it contains 16-bits analog to digital conversion hardware
Analog inputs 2
Digital outputs 4 for each channel. Therefore it captures the x, y and z
Interface RS232, CAN and USB 2.0 channels at the same time. This sensor is mounted on the
top of the robot body. Over and above, a 3D motion
capture system may also be used to determine position
and velocity of leg tip. Commercially available PTI
controller [23]. Major specifications of this controller are Visualeyez motion trackers [25] is used in this research
shown in Table 2. To control eight motor units one need because they are highly accurate and calibration free
eight controllers. There is one master and seven slave tracking system. These trackers are able to continuously
controllers. Each controller controls one motor unit. Master collect information about the markers which are attached
controller does positioning control of its own motor as well on the leg tip.
as sends signal to other slave controllers to perform as per
the program. Other characteristic of master controller
EPOS2 P are IEC 61131-3 programmable, CANopen
master function, multiple axis system via CAN Bus 2.7 Fabrication and final assembly
CANopen, point to point control unit and interpolated
After finalizing link, actuation system and controllers, a
position mode. Master controller communication with
CAD model of the quadruped is prepared. Specifications of
computer, its own motor and other controller is shown in
the developed quadruped robot are listed in table 3. Ini-
figure 7.
tially, as shown in figure 9, possible arrangements of the
EPOS studio is used for programming according to
controller on the robot are evaluated. Finally, a cabinet is
IEC61131-3. Many decentralized controls can be easily
prepared to arrange all eight controllers on the quadruped
managed with complex programs in EPOS studio. Desired
as shown in figure 10. In the developed quadruped robot
input data are given in terms of encoder increment which is
each leg has two electric motors per leg, but provision is
called quad count (qc). Entire process of locomotion
also kept (as shown in the CAD model) to introduce third
planning and execution are presented in the flow chart as
degree of freedom by fitting third motor at each leg. After
shown in figure 8.
satisfactory visualization of CAD model, fabrication of

Figure 7. Controller connections with motors [23].

102 Page 6 of 18 Sådhanå (2018) 43:102

Figure 8. The flow chart of quadruped locomotion planning and execution.

Table 3. Specifications of quadruped robot developed at IIT quadruped robot was started. Patterns were prepared for
Roorkee, India. hub and upper link and they were casted from aluminum by
sand casting process and then they were machined for final
Parameter Value sizing. Robot chassis was prepared from aluminum chan-
Length of upper link 0.225 m nels. Lower link is ready made purchased and then it is
Length of lower (compliant) link 0.190 m modified by introducing the spring for leg compliance. The
Body length 0.500 m cabinet is fabricated from Galvanized Iron Sheet. Two
Body width 0.420 m stands are prepared from mild steel to hold robot while it is
Body thickness 0.065 m not moving. The final shape of the developed robot is
Quadruped robot weight (approx.) 15 kg shown in figure 11.
Main power source 24VDC, 63

Figure 9. CAD models for possible arrangements of controllers on quadruped robot.

Sådhanå (2018) 43:102 Page 7 of 18 102

center of body, top body and upper links are rigid, joint
rotation allows rotation of link about one axis only, robot is
walking on hard surface and on even terrain and external
force and moments effects are negligible. However, real
situation differs.

3. Locked joint fault accommodation

3.1 Strategies for locked joint fault tolerant

A moving appendage device containing two-DOF planar
robot with the payload at the end is introduced here to
accommodate locked joint failure. Entire planar robot is
arranged on the top of the quadruped robot body. The
payload attached at the tip serves as an inertial element and
its movement causes the change in the position of center of
Figure 10. Solid model of quadruped robot. gravity (CG). Figure 14 shows the schematic diagram of a
quadruped robot with moving appendage device in which
{A} is the inertial frame of reference and {B} is the body
frame. In addition, {P0} is the fixed frame which coincides
with the body frame, {P1} is the frame at the joint between
fixed frame and first link of two DOF robot, {P2} is the
frame at the joint between first and second link, and {P3} is
the frame attached at the tip of second link of two DOF
planar robot.
Governing equations for angular velocity propagation
(AVP) of planar robot can be derived as [26],
iþ1 A
ð xiþ1 Þ ¼ iþ1
i Ri ðA xi Þ þ iþ1 ði xiþ1 Þ ð2Þ

where, iþ1 ðA xiþ1 Þ is the angular velocity of (i?1)th link

with respect to the inertial frame and expressed in (i?1)th
frame; iþ1
i R represents the transformation from the body
frame {i} to the frame {i?1}; i ðA xi Þ is the angular velocity
of the ith link with respect to the inertial frame {A} and
expressed in ith frame and iþ1 ði xiþ1 Þ is the angular
Figure 11. Quadruped robot on stand. velocity of (i?1)th link as observed from ith link and
expressed in (i?1)th frame.
Equation (2) can be written for links 1 and 2 of two-DOF
2.8 Results planar robot as,

The operation of the developed quadruped was tested for ð xP1 Þ ¼ PB1 RB ðA xB Þ þ P1 ðB xP1 Þ
P1 A
ð3Þ
various gait patterns. Amble gait, which is statically
stable gait in which legs operate one by one in 1–4–2–3 ð xP2 Þ ¼ PP21 RP1 ðA xP1 Þ þ P2 ðP1 xP2 Þ
P2 A
ð4Þ
sequence, is considered in this paper as a test case. Here,
experimental results are shown for amble gait. Figure 12 Frame {P3} has same angular velocity as frame {P2},
shows a few snaps of quadruped robot walking with amble because both these frames are attached on the same link.
gait. It takes 3.3 s to complete one cycle. Experiment data Thus,
for joint rotations for five cycles are shown in figure 13. P3 A
ð xP3 Þ ¼ P2 ðA xP2 Þ ð5Þ
Figure 12 shows robot forward motion. In five cycles, it
is found to move by around 0.39 m. Surface condition plays Here, ð xP1 Þ ¼ ½ 0 0 h_ P1 T
P1 B
and
major role in the locomotion of robot. Also, there are T _
P2 P1 _
ð xP2 Þ ¼ ½ 0 0 hP2  , where hP1 represents the
always assumptions considered for preparation of mathe-
matical model like mass center of link is located at the mid angular velocity of the frame {P1} and similarly h_ P2 rep-
of its length, center of gravity of top body is located at the resents angular velocity of the frame {P2}.
102 Page 8 of 18 Sådhanå (2018) 43:102

Figure 12. Quadruped robot walking.

For translational velocity propagation (TVP), the gov- joint in that leg. Assuming that robot is walking with trot
erning equations for the link tip velocity and the link CG gait. In the first phase of trot gait, diagonally opposite
velocity of planar robot can be given as [26], legs move forward while in the second phase, the same
leg tips touch the ground and body propagation takes
A A
ð Viþ1 Þ ¼ A ðA Vi Þ þ Ai R½i ðA xi Þ  i ði Piþ1 Þ ð6Þ place. In the case of joint failure, there is no guarantee of
proper leg tip contact on the ground for body propagation
where, A ðA Viþ1 Þ represent the translational velocity of
and thus the locomotion is severely affected. In the
frame {i?1} with respect to inertial frame and expressed in
proposed control strategy, the two-DOF robot payload
the inertial frame; A ðA Vi Þ represent the translational
position is controlled in such a way that it moves towards
velocity of frame {i} with respect to inertial frame and
the hip joint of failed leg during body propagation to
expressed in the inertial frame;i ði Piþ1 Þ refers to the position ensure proper contact of failed leg tip on the ground and
vector of link frames {i?1} with respect to its previous towards center of the body during the forward movement
frames {i} and expressed in frame {i}. Equation (6) can be phase of the failed leg.
written in for two-DOF planar robot as,
For example, if joint 1 of leg 1 has failed then the desired
A A X and Y coordinates of payload mass during the first phase,
ð VP2 Þ ¼ A ðA VP1 Þ þ AP1 R½P1 ðA xP1 Þ  P1 ðP1 PP2 Þ ð7Þ
i.e., during leg 1 forward movement, are Xdes = - 0.098 m
A A and Ydes = - 0.013 m in X and Y directions, respectively.
ð VP3 Þ ¼ A ðA VP2 Þ þ AP2 R½P2 ðA xP2 Þ  P2 P2
ð PP3 Þ ð8Þ
These limits are decided from constrain of physical model.
If link lengths lP1 and lP2 are taken along the principal X- The main purpose is to keep the mass near the center during
axis of the links then they can be represented in vector form first phase while during the second phase, i.e., contact of leg
as, P1 PP2 ¼ ½ lP1 0 0 T , P2 PP3 ¼ ½ lP2 0 0 T . The tip with the ground,
moving appendage device is activated only when fault   
R1x
occurs. If one of the legs joint is locked then the leg can still Xdes ¼ ðlP1 þ lP2  0:005Þ sin tan1 ð9Þ
R1y
move in a restricted way with the help of the other working
Sådhanå (2018) 43:102 Page 9 of 18 102

Figure 13. Joints rotation: (a) Leg 1 joint 1; (b) Leg 1 joint 2; (c) Leg 2 joint 1; (d) Leg 2 joint 2; (e) Leg 3 joint 1; (f) Leg 3 joint 2; (g)
Leg 4 joint 1; (h) Leg 4 joint 2.

  
1 R1x
Ydes ¼ ðlP1 þ lP2  0:005Þ cos tan ð10Þ
R1y

where, R1x and R1y are the distance of hip joint from body
center. To avoid singularity problems, an offset of 0.005 m
is introduced in Eqs. (9) and (10). Similar control laws are
expressed for joint failure in other legs. The above dis-
cussed control law is activated only after a joint failure is
detected in a particular leg. There is a bank of control laws
implemented in the form of if-then blocks and the appro-
priate control law is executed by the decision support sys-
tem based on the results of the fault diagnosis tools in the
system supervision module.
For the above desired coordinates of a planar robot tip,
necessary joint rotations can be derived using the Jacobian.
The Jacobian is the mapping between velocities in the joint Figure 14. Schematic diagram of a quadruped robot with moving
space to the Cartesian space. Equation (8) is utilized to appendage device.
102 Page 10 of 18 Sådhanå (2018) 43:102

Figure 15. Multi-bond graph of quadruped robot with two-DOF planar robot.

derive the Jacobian matrix. For position control of two- the Jacobian which then evaluates the required efforts at the
DOF planar robot tip with reference to time, desired posi- joints for appendage movement. The required efforts for the
tion of the tip discussed above are used as the reference. first and second joints can be evaluated as
Here, the reference velocity is compared with the actual
planar robot tip velocity and error values are sent to a PID
controller. The PID controller sends corrective signals to V1 ¼ CX b12 þ CY b11 þ CZ b10 ð11Þ
Sådhanå (2018) 43:102 Page 11 of 18 102

Table 4. Input parameters.

Parameters Value
Leg parameters
First link length of leg (l1) 0.225 m
Mass of first link (Ml1) 1.11 kg
Mass of cylinder part of the prismatic link (Mc) 0.3 kg
Mass of piston part of the prismatic link (Mp) 0.2 kg
Inertia of Link 1
Ixx1 0.013346 kg m2
Iyy1 0.007396 kg m2
Izz1 0.011563 kg m2
Inertia of cylinder part of the prismatic link
Ixxc=Izzc 0.005144 kg m2
Iyyc 0.000487 kg2
Inertia of piston and piston rod of the prismatic link
Ixxp=Izzp 0.00168 kg m2
Iyyp 0.000025 kg m2
Stiffness of the prismatic link (kf) 5000 N/m
Damping of the prismatic link (Rf) 274 Ns/m
Contact point stiffness at the piston cylinder of the prismatic link (kb) 108 N/m
Contact point resistance at the piston and cylinder of the prismatic link (Rb) 103 Ns/m
Length of piston and piston rod of the prismatic link (lp) 0.1 m
Distance of cylinder CG from the end frame of prismatic link (lcg) 0.05 m
Distance of piston CG from the end frame of the prismatic link (lpg) 0.07 m
Mass of piston & piston rod of the prismatic link (mp) 0.2 kg
Mass of cylinder part of the prismatic link (mc) 0.3 kg
Position of the cylinder end point with respect to the body fixed frame at the mass center in meter, (x2, y2, z2) (0.0, - 0.05, 0.0)
Position of the piston end point with respect to the body fixed frame at the mass center in meter, (x3, y3, z3) (0.0, 0.07, 0.0)
Common parameters
Mass of body (Mb) 6.94 kg
Inertia of body
Ixb 0.1470 kg m2
Iyb 0.1045 kg m2
Izb 0.2466 kg m2
Ground damping in x, y, z direction (Rgx, Rgy, Rgz) 1000 Ns/m
Ground stiffness in z direction (Kgz) 106 N/m
Controller parameters
Proportional gain of controller(Kp) 150
Derivative gain of controller (Kd) 90
Integral gain of controller (Ki) 40
Joint actuator parameters of quadruped
Motor constant (Kt) 0.0276 Nm/A
Motor armature resistance(Rm) 0.386 X
Motor inductance(Im) 0.001 H
Gear ratio (n) 230
Two-DOF planar robot parameters
Link 1 and Link 2 length of planar robot (lP1, lP2) 0.055 m, 0.140 m
Mass of link of planar robot (MPl1, MPl2) 0.05 kg, 0.06 kg
Payload mass (Mp) 0.5 kg
Inertia of both link of planar robot (Ix, Iy, Iz) 0.0001 kg m2
Motor constant (Kt) 0.02 Nm/A
Motor armature resistance (Ra) 0.1 Ohms
Gear ratio 100
Bearing resistance (Rb) 0.01 Ns/m
Proportional and integral gain values(Kpp), (Kip) 250, 6700
102 Page 12 of 18 Sådhanå (2018) 43:102

Figure 16. Body CG movement in (a) Y and (b) Z directions.

Figure 17. Joints rotation: (a) Joint 1; (b) Joint 2.

Figure 18. Two-DOF Planar robot joints rotation: (a) Joint 1; (b) Joint 2.
Sådhanå (2018) 43:102 Page 13 of 18 102

Figure 19. Two-DOF Planar robot tip movement in (a) X and (b) Y direction.

and a simple presentation of a bond graph model can be

carried out in multi bond graph form. Modeling of a
quadruped robot consists of a modeling of translational and
angular dynamics of robot body and legs. The detailed
discussion of bond graph model of quadruped robot is
covered in our previous publications [13, 14]. Model of the
two-DOF planar robot, discussed in Sect. 3.1, is appended
on the bond graph model of the quadruped robot given in
[13, 14]. Using common flow junction (1 junction), com-
mon effort junction (0 junction), one port elements viz.,
Resistor (R), Inertia (I), Capacitor (C), Source of Effort
(SE) and Source of Flow (SF); two port elements viz.,
transformer (TF) and gyrator (GR), the bond graph model
of the two DOF planar robot is developed. As shown in
figure 15, AVP of link P1 indicates modeling of angular
velocity propagation of P1 link as per Eq. (3), and AVP of
link P2 indicates modeling of angular velocity propagation
of P2 link Eq. (4). Similarly, TVP of links P1 and P2
indicate modeling of translational velocity propagations as
per Eqs. (7) and (8), respectively, for links P1 and P2.
Figure 20. Experiment set-up.
Modeling of geared motors of two-DOF robot is same as
modeling of geared motor of quadruped robot discussed in
our publication [14]. The translational velocity and the
V2 ¼ CX b15 þ CY b14 þ CZ b13 : ð12Þ angular velocity vectors have been resolved into three
mutually perpendicular components, and their dynamics
Where CX, CY and CZ are corrective signal received from
can be modelled with Euler junction structure (EJS). EJS of
PID in X, Y and Z directions, respectively. Detailed
two DOF planar link P1 and P2 is modelled as same as EJS
description of coefficients b10, b11, b12, b13, b14 and b15
of quadruped robot link discussed in [14]. The pad is used
used in Eqs. (11) and (12) are shown in ‘‘Appendix B’’.
to avoid differential causality. Pads are artificial compli-
ances/lumped flexibilities that can be used in bond graph.
The Jacobian sub-model shows the mapping between
3.2 Dynamic model of the robot velocities in the joint space to the Cartesian space. This sub
To validate the above discussed reconfiguration strategy, a model can be developed using Eq. (8). PID controller sub
bond graph model is developed. Bond graph is a tool used model as shown in figure 15 is also developed which is
to model physical systems through power interaction. responsible for the required effort at joint P1 and P2 of two
Multi-bond graph presentation of quadruped robot with DOF planar robot. PID controller sub model is developed
two-DOF planar robot is shown in figure 15. A compact using Eqs. (11) and (12).
102 Page 14 of 18 Sådhanå (2018) 43:102

Figure 21. Joints rotation of physical model of quadruped robot during experiment: (a) joint 1 of leg 1 and 2; (b) joint 1 of leg 3 and 4;
(c) joint 2 of leg 1 and 2; (d) joint 2 of leg 3 and 4.

Figure 22. Joint rotation of physical model of two-DOF planar robot during experiment: (a) joint 1; (b) joint 2.

3.3 Simulation results of locked joint fault Simulation is carried out for six cycles. Each cycle is of
accommodation through reconfiguration 1.7 s duration. Initially, it is assumed that there is no fault.
Figure 16 shows the quadruped body CG motion in the
The control law discussed in section 3.1 for fault accom- Y and Z directions. Figures 17(a) and (b) show the joint 1
modation through reconfiguration is implemented in the and 2 rotations, respectively, of all leg joints of the quad-
model developed in section 3.2. Parameter used for the ruped. Initially, all the joints are working properly and the
simulations are listed in table 4. moving appendage device is switched off. Immediately
A fault is intentionally introduced during the locomotion after the joint failure (joint locked), the moving appendage
of the quadruped robot and the locomotion performance is device starts working. For demonstration of the control
observed with and without the moving appendage.
Sådhanå (2018) 43:102 Page 15 of 18 102

Figure 23. Tip movement of physical model of two-DOF planar robot during experiment in (a) X and (b) Y.

Figure 24. Snaps taken during experiment of reconfiguration using two-DOF planar robot.

strategy, the locked joint failure is introduced at joint 2 of contribution of the moving appendage device then it can
leg 1. Figure 18 shows two-DOF planar robot joint rotation traverse 0.362 m distance, whereas it can travel up to
for joint 1 and 2, while figure 19 shows the planar robot tip 0.441 m in 10.2 s with the controlled motion of the moving
position in the X and Y local frames with respect to time, appendage. Thus, it shows overall improvement in loco-
respectively, in which the moving appendage starts working motion because of the control strategies implemented with
immediately after the failure is detected at 2.4 s (automatic the moving appendage device.
failure detection is not implemented here, so the switching
condition is hard-coded). As per the control strategy, the
appendage device remains near the center during leg 1 3.4 Experimental validation of locked joint fault
forward motion, while it tries to move towards the hip joint accommodation through reconfiguration
of leg 1 during body propagation. Because of the weight of
the payload mass, the controller ensures proper contact of Above discussed simulation results are also verified
leg tip with the ground and effective body propagation. If through experimental results. A Quadruped robot as dis-
the quadruped robot continues locomotion without cussed in section 2 is used. A two-DOF planar robot with a
102 Page 16 of 18 Sådhanå (2018) 43:102

payload at its end is constructed and mounted on the Appendix A

quadruped robot as shown in figure 20. Links of this planar
robot are fabricated from mild-steel and 0.5 kg mass is kept Equation of motions for back stance phase and front stance
in stainless box attached at the end of robot. Both the links phase
are operated through 12 V DC motor (Brushed DC motor
with 100:1 metal gearbox) attached at their joints. These s1 ¼ ½l1 ðmb þ m1 þ m2 Þ sinðh2 Þ€l2 þ ½l21 ðmb þ m1 þ m2 Þ
motors get power through 20 A Dual DC motor driver  l22 ðmb þ 2m1 þ m2 Þ  Ll1 ðmb þ 2m1 þ 2m2 Þ sinðh1 Þ
which is controlled by Ardiuno Mega ADK controller.  2l1 l2 ðmb þ m1 þ m2 Þ cosðh2 Þ
Leg joints rotations are shown in figure 21 for both
þ ðm1 þ m2 Þl21 cosðh1 þ h3 Þ
healthy and faulty state while planar robot joint rotations
þ ðm1 þ m2 Þl1 l2 cosðh1  h2 þ h3 Þ
are shown in figure 22 for faulty state and robot tip position
in X and Y directions are shown in figures 23(a) and (b), þ m2 l1 l0 cosðh1 þ h3  h4 Þ
respectively. It has been observed that without such moving þ m2 l2 l0 cosðh2  h1 þ h4  h3 Þ
appendage device with failed leg quadruped travels around  Ll2 ðmb þ 2m1 þ 2m2 Þ
0.3 m distance in five cycles while with appendage device sinðh1  h2 Þh€ þ ½l2 ðmb þ m1 þ m2 Þ þ m1 r2
1 1
it travels around 0.35 m distance. Figure 24 shows few
þ l22 ðmb þ 2m1 þ m2 Þ þ 2l1 l2 ðmb þ m1 þ m2 Þ
snaps taken during experiment on reconfiguration for
locked joint failure using two DOF planar robot. Fig- cosðh2 Þh€1 þ ½l2 ðmb þ 2m1 þ m2 Þ
2

ure 24(a) shows planar robot payload position near the  l1 l2 ðmb þ m1 þ m2 Þ cosðh2 Þh€2
 
center, Figures 24(b) and (c) payload moves away from the  2ðmb þ 2m1 þ m2 Þl2 l_2 h_ þ h_ 2  h_ 1
center and figure 24(d) shows payload moves near the
center. þ ðmb þ 2m1 þ 2m2 Þ
h i
Ll1 h_ 2 cosðh1 Þ  Ll2 h_ 2 cosðh1  h2 Þ
h  
4. Conclusions þ ðmb þ m1 þ m2 Þl1 l_2 2h_ 2  2h_ 1 þ 2h_ cosðh2 Þ
  i
þl2 h_ 2 h_ 2  2h_ 1 þ 2h_ sinðh2 Þ
This paper presents the entire process of development of a h
quadruped robot. This work may serve as a guideline for þ ðm1 þ m2 Þ l21 h_ 2 sinðh1 þ h3 Þ
those who are interested to develop quadruped robot for i
their research purpose. Here, experimental results of robot l1 l2 h_ 2 sinðh1  h2 þ h3 Þ
h
walking with amble gait are demonstrated. There are cer- þ m2 l1 l0 h_ 2 sinðh1 þ h3  h4 Þ
tain issues like body disturbance during dynamic walk and i
different types of joint failures. A strategy for locked joint þl0 l2 h_ 2 sinðh2  h1 þ h4  h3 Þ
failure through reconfiguration by using hardware redun-    
þ g ðmb þ m1 þ m2 Þl1 sin h1  h þ hp
dancy is demonstrated. Simulation and experimental results
þðmb þ 2m1 þ m2 Þl2 sinðh2  h1 þ h þ hp Þ
show the success of developed fault tolerant control strat-
egy. This quadruped robot with the added two-DOF robot s2 ¼ ½ðmb þ 2m1 þ m2 Þl22 þ ðmb þ 2m1 þ 2m2 ÞLl2
on its platform or elsewhere may be used to explore various
sinðh1  h2 Þ þ ðmb þ m1 þ m2 Þl1 l2 cosðh2 Þ
other control strategies such as posture control, fault tol-
 ðm1 þ m2 Þl1 l2 cosðh1  h2 þ h3 Þ
erant and reconfiguration strategies for other types of joint
failures and/or multiple joint failures.  m2 l0 l2 cosðh2  h1 þ h4  h3 Þh€
þ ½l22 ðmb þ 2m1 þ m2 Þ  l1 l2 ðmb þ m1 þ m2 Þ
cosðh2 Þh€1 þ ½l2 ðmb þ 2m1 þ m2 Þ þ m2 r 2 h€2
2 2
Acknowledgements
þ ðmb þ 2m1 þ 2m2 ÞLl2 h_ 2 cosðh1  h2 Þ
 2
The research work presented here was funded by the þ ðmb þ m1 þ m2 Þl1 l2 h_  h_ 1 sinðh2 Þ
Department of Science and Technology (DST), India under h i
Indo–Korea Joint Research in Science and Technology vide þ ðm1 þ m2 Þ l1 l2 h_ 2 sinðh1  h2 þ h4 Þ
Grant No. INT/Korea/P–13. Mr M M Gor is thankful to G. h i
þ m2 l0 l2 h_ 2 sinðh2  h1 þ h4  h3 Þ
H. Patel College of Engineering and Technology, Gujarat, 
India for allowing him to carry out research work at IIT  g ðmb þ 2m1 þ m2 Þl2 sinðh2  h1 þ h þ hp Þ
 
Roorkee. The work of Mr J M Yang and Mr S W Kwak was þ 2ðmb þ 2m1 þ m2 Þl2 l_2 h_ þ h_ 2  h_ 1
supported by the Korea government (MEST) vide the
National Research Foundation of Korea Grant No. NRF–
2011–0027705.
Sådhanå (2018) 43:102 Page 17 of 18 102

s3 ¼ ½l1 ðmb þ m1 þ m2 Þ sinðh2 Þ€l2 þ ½l21 ðmb þ m1 þ m2 Þ s4 ¼ ½ðmb þ 2m1 þ m2 Þl22 þ ðmb þ 2m1 þ m2 Þ
 l22 ðmb þ 2m1 þ m2 Þ  Ll1 ðmb þ 2m1 þ 2m2 Þ sinðh3 Þ Ll2 sinðh3  h4 Þ
þ 2l1 l2 ðmb þ m1 þ m2 Þ cosðh4 Þ þ ðmb þ m1 þ m2 Þl1 l2 cosðh4 Þ  ðm1 þ m2 Þl1 l2
þ ðm1 þ m2 Þl21 cosðh3 þ h1 Þ
cosðh4  h3  h1 Þ þ m2 l0 l2 cosðh4  h3 þ h2  h1 Þh€
þ ðm1 þ m2 Þl1 l2 cosðh4  h3  h1 Þ  m2 l1 l0 
þ l22 ðmb þ 2m1 þ m2 Þ  l1 l2 ðmb þ m1 þ m2 Þ
cosðh3 þ h3  h4 Þ  m2 l2 l0 cosðh4  h3 þ h2  h1 Þ 
 Ll2 ðmb þ 2m1 þ 2m2 Þ sinðh3  h4 Þh€ cosðh4 Þh€1 þ l22 ðmb þ 2m1 þ m2 Þ h€4
  
þ l21 ðmb þ m1 þ m2 Þ þ m1 r12 þ l22 ðmb þ 2m1 þ m2 Þ þ 2ðmb þ 2m1 þ m2 Þl2 l_2 h_ þ h_ 4  h_ 3

þl1 l2 ðmb þ m1 þ m2 Þh€3 þ l2 ðmb þ 2m1 þ m2 Þ
2 þ ðmb þ 2m1 þ m2 ÞLl_2 h_ sinðh3  h4 Þ
l1 l2 ðmb þ m1 þ m2 Þh€4 þ ½2ðmb þ 2m1 þ m2 Þl2 l_2
  h þ ðmb þ 2m1 þ 2m2 ÞLl2 h_ cosðh3  h4 Þ
h_ þ h_ 4  h_ 3 þ ðmb þ 2m1 þ 2m2 Þ Ll1 h_ 2 cosðh3 Þ  2
i þ ðmb þ m1 þ m2 Þl1 l2 h_ 3  h_ sinðh4 Þ
Ll2 h_ 2 cosðh3  h4 Þ þ ðmb þ m1 þ m2 Þl1   
h   þ ðmb þ 2m1 þ 2m2 ÞL l2 h_ h_ þ h_ 4  h_ 3
l_2 2h_ 4  h_ 3 þ h_ cosðh4 Þ 
  i
cosðh3  h4 Þ  l_2 h_ sinðh3  h4 Þ
þl2 h_ 4 h_ 4  h_ 3 þ h_ sinðh4 Þ
h h i
þ ðm1 þ m2 Þ l21 h_ 2 sinðh3 þ h1 Þ þ ðm1 þ m2 Þ l1 l2 h_ 2 sinðh4  h3  h1 Þ
i h
þl1 l2 h_ 2 sinðh4  h3  h1 Þ þ m2 l0 l_2 h_ cosðh4  h3 þ h2  h1 Þ
h   
þ m2 l1 l0 h_ 2 sinðh3 þ h1  h2 Þ l_2 h_ cos h4  h3  h2 þ h1 þ 2 h  hp
   i
þl0 l_2 h_ cos h4  h3  h2 þ h1 þ 2 h  hp þl2 h_ 2 sinðh4  h3 þ h2  h1 Þ
i 
l0 l2 h_ 2 sinðh4  h3 þ h2  h1 Þ  g ðmb þ 2m1 þ m2 Þl2 sinðh4  h3 þ h  hp Þ
  
þ g ðmb þ m1 þ m2 Þl1 sin h3  h þ hp

þðmb þ 2m1 þ m2 Þl2 sinðh4  h3 þ h  hp Þ

Appendix B

Coefficients used in equations to determine the required

efforts at the joints of the two-DOF planar robot

Coefficient Equation
b10 ðsinðhÞ sinðhP1 Þ þ sinðwÞ cosðhÞ cosðhP1 ÞÞlP1
þ ½ð sinðhÞ cosðhP1 Þ þ sinðwÞ cosðhÞ sinðhP1 ÞÞ sinðhP2 Þ
þðsinðhÞ sinðhP1 Þ þ sinðwÞ cosðhÞ cosðhP1 ÞÞ cosðhP2 ÞlP2
b11 ½ cosðhÞ sinð/Þ sinðhP1 Þ þ ðsinðwÞ sinðhÞ sinð/Þ þ cosðwÞ cosð/ÞÞ cosðhP1 ÞlP1
þ ½ðcosðhÞ sinð/Þ cosðhP1 Þ þ ðsinðwÞ sinðhÞ sinð/Þ þ cosðwÞ cosð/ÞÞ sinðhP1 ÞÞ sinðhP2 Þ
þð cosðhÞ sinð/Þ sinðhP1 Þ þ ðsinðwÞ sinðhÞ sinð/Þ þ cosðwÞ cosð/ÞÞ cosðhP1 ÞÞ cosðhP2 ÞlP2
b12 ½ð cosðhÞ cosð/Þ sinðhP1 Þ þ sinðwÞ sinðhÞ cosð/Þ cosðhP1 Þ  cosðwÞ sinð/Þ cosðhP1 ÞÞlP1
þ ððcosðhÞ cosð/Þ cosðhP1 Þ þ ðsinðwÞ sinðhÞ cosð/Þ  cosðwÞ sinð/ÞÞ sinðhP1 ÞÞ sinðhP2 Þ
þð cosðhÞ cosð/Þ sinðhP1 Þ þ ðsinðwÞ sinðhÞ cosð/Þ  cosðwÞ sinð/ÞÞ cosðhP1 ÞÞ cosðhP1 ÞlP2
b13 ½ð sinðhÞ cosðhP1 Þ þ sinðwÞ cosðhÞ sinðhP1 ÞÞ sinðhP2 Þþ
ðsinðhÞ sinðhP1 Þ þ sinðwÞ cosðhÞ cosðhP1 ÞÞ cosðhP2 ÞlP2
b14 ½ðcosðhÞ sinð/Þ cosðhP1 Þ þ ðsinðwÞ sinðhÞ sinð/Þ þ cosðwÞ cosð/ÞÞ sinðhP1 ÞÞ sinðhP2 Þ
þð cosðhÞ sinð/Þ sinðhP1 Þ þ ðsinðwÞ sinðhÞ sinð/Þ þ cosðwÞ cosð/ÞÞ cosðhP1 ÞÞ cosðhP2 ÞlP2
b15 ½ðcosðhÞ cosð/Þ cosðhP1 Þ þ ðsinðwÞ sinðhÞ cosð/Þ  cosðwÞ sinð/ÞÞ sinðhP1 ÞÞ sinðhP2 Þ
þð cosðhÞ cosð/Þ sinðhP1 Þ þ ðsinðwÞ sinðhÞ cosð/Þ  cosðwÞ sinð/ÞÞ cosðhP1 ÞÞ cosðhP2 ÞlP2
102 Page 18 of 18 Sådhanå (2018) 43:102

Nomenclature IEEE/RSJ International Conference on Intelligent Robots

and Systems, (IROS 2004). pp. 986–991
L Half the distance between the hip joints [8] Lee Y-J and Hirose S 2000 Three-Legged Walking for Fault
Tolerant Locomotion of a Quadruped Robot with Demining
mb Body mass
Mission. In: Proceedings of the 2000 IEEE/RSJ International
rb Body radius of gyration
Conference on intelligent Robots and Systems, pp. 973–978
k Spring stiffness [9] Yang J-M 2002 Fault-tolerant gaits of quadruped robots for
b Damping coefficient locked joint failures. IEEE Transactions on Systems, Man
lo Free leg length (zero spring force) and Cybernetics-Part C, 32(4): 507–516
l1 Link 1 length [10] Yang J-M 2003 Crab walking of quadruped robots with a
l2 Link 2 length locked joint failure. Advanced Robotics, 17(9): 863–878
m1 Mass of link 1 [11] Yang J-M 2008 Two-phase discontinuous gaits for quad-
m2 Mass of link 2 ruped walking machines with a failed leg. Robotics and
hp Inclination of plane w.r.to horizontal Autonomous Systems, 56(9): 728–737
h Body angle w.r.to inclined plane [12] Krishnan V L, Pathak P M, Jain S C and Samantaray A K
2011 Reconfiguration of four-legged walking robot for
h1 Angle of rotation of back leg (link 1)
actuator faults. Proceedings of the Institution of Mechanical
h2 Angle of rotation of back leg (link 2)
Engineers, Part I: Journal of Systems and Control Engi-
h3 Angle of rotation of front leg (link 1) neering, 225: 1–16
h4 Angle of rotation of front leg (link 2) [13] Gor M, Pathak P, Samantaray A, Yang J and Kwak S 2013
s1 Torque applied at back leg (link 1) Dynamic Modeling and Simulation of Compliant Legged
s2 Torque applied at back leg (link 2) Quadruped Robot. In: Proceeding of the 1st International and
s3 Torque applied at front leg (link 1) 16th National Conference on Machines and Mechanisms (iNa-
s4 Torque applied at front leg (link 2) CoMM2013), IIT Roorkee, India, Dec 18–20 2013. pp. 7–16
hP1 Angle of rotation of frame {P1} of two DOF [14] Gor M, Pathak P, Samantaray A, Yang J and Kwak S 2015
planar robot Control oriented model-based simulation and experimental
hP2 Angle of rotation of frame {P2} of two DOF studies on a compliant legged quadruped robot. Robotics and
Autonomous System 72: 217–234
planar robot
[15] Ganesh Kumar K and Pathak P 2013 Dynamic modelling and
/; h; w Z-Y-X Euler angles
simulation of a four legged jumping robot with compliant
legs. Robotics and Autonomous Systems 61(2013): 221–228
[16] Estremera J and Waldron K 2008 Thrust control, stabiliza-
tion and energetics of a quadruped running robot. Interna-
References tional Journal of Robotics Research 27: 1135–1151
[17] Xiuli Z, Haojun Z, Xu G, Zhifeng C and Liyao Z 2005 A bio-
[1] Chen X, Gao F, Qi C and Zhao X 2013 Spring parameters logical inspired quadruped robot: structure and control. IEEE
design to increase the loading capability of a hydraulic quad- International Conference on Robotics and Biomimetics 387–392
ruped robot. In: Proceeding of the International conference on [18] Raibert M, Chepponis M and Brown B 1986 Running on four
Advanced Mechatronics Systems, Luoyang, China, pp. 535–540 legs as though they were one. IEEE Journal of Robotics and
[2] Spröwitz A, Tuleu A, Vespignani M, Ajallooeian M, Badri E Automation, RA-2. 70–82
and Ijspeert A 2013 Towards dynamic trot gait locomotion: [19] Hodoshima R, Doi T, Okamoto T and Mori J 2004 Devel-
Design, control, and experiments with Cheetah-cub, a com- opment of TITAN XI: a Quadruped Walking Robot to Work
pliant quadruped robot. The International Journal of on Slopes. In: Proceeding of IEEE/RSJ International Con-
Robotics Research, 32(8): 932–950 ference on Intelligent Robots and Systems, September
[3] Semini C, Tsagarakis N, Guglielmino E, Focchi M, Cannella 28-October 2, 2004, Sendal, Japan. pp. 792–797
F and Caldwell D 2011 Design of HyQ - a hydraulically and [20] Tanaka T and Hirose S 2008 Development of leg-wheel
electrically actuated quadruped robot. Proceedings of the hybrid quadruped airhopper - design of powerful light-
Institution of Mechanical Engineers, Part I: Journal of weight leg with wheel. In: Proceedings of the IEEE/RSJ
Systems and Control Engineering. 225(6): 831–849 International Conference on Intelligent Robots and Systems
[4] Byl K 2008 Metastable Legged-Robot Locomotion. Ph.D. (IROS), 2008, pp. 3890–3895
Thesis, Massachusetts Institute of Technology [21] Poulakakis I, Smith J and Buehler M 2005 Modeling and
[5] Shkolnik A, Levashov M, Manchester I and Tedrake R 2011 experiments of untethered quadrupedal running with a
Bounding on rough terrain with the LittleDog robot. The bounding gait: The scout II robot. International Journal of
International Journal of Robotics Research. 30: 192–215 Robotics Research 24: 239–256
[6] Fukuoka Y and Kimura H 2009 Dynamic locomotion of a [22] Kurfess T 2004 Robotics and Automation Handbook. Boca
biomorphic quadruped ‘Tekken’ robot using various gaits: Raton: CRC Press
walk, trot, free-gait and bound. Applied Bionics and [23] http://www.maxonmotor.in/
Biomechanics 6: 63–71 [24] http://www.invensense.com/mems/gyro/mpu6050.html
[7] Kimura H and Fukuoka Y 2004 Biologically inspired adap- [25] http://www.ptiphoenix.com/products/
tive dynamic walking in outdoor environment using a self- [26] Craig, J J 2005 Introduction to Robotics Mechanics and
contained quadruped robot:’Tekken2’. In: Proceeding of Control, 3rd ed. USA: Pearson Education