Robotics and Autonomous Systems 61 (2013) 221–228

Contents lists available at SciVerse ScienceDirect

Robotics and Autonomous Systems

journal homepage: www.elsevier.com/locate/robot

Dynamic modelling & simulation of a four legged jumping robot with

compliant legs✩
Ganesh Kumar K., Pushparaj Mani Pathak ∗
Robotics and Control Laboratory, Mechanical and Industrial Engineering Department, Indian Institute of Technology, Roorkee – 247667, India

article info abstract

Article history: Legged locomotion is used by most animals and human beings on Earth. Legged locomotion is preferred
Received 27 March 2012 over the wheeled locomotion as it can be used for both flat and rough terrains. In this paper, an attempt
Received in revised form has been made for the dynamic modelling and simulation of four legged jumping robots with compliant
23 July 2012
legs. Sagittal plane and bounding gait has been considered for the simulation. For energy saving, passive
Accepted 10 September 2012
Available online 8 November 2012
dynamics has been used with the help of compliant legs (linear spring). Different state variables have
been obtained for analysis. Control strategies have been implemented on dynamic modelling for forward
Keywords:
velocity control.
Passive dynamics © 2012 Elsevier B.V. All rights reserved.
Legged locomotion
Sagittal plane
Control

1. Introduction leg. Cherouvim and Papadopoulos [5] explored the mechanism of

energy transfer between the single actuated degrees of freedom
Research on a compliant legged robot has been started in recent (DOF) of a one-legged hopping robot and the remaining unactuated
decades. Four legged walking robots provide better stability and DOFs. Berkemeier [6] performed an analytical study on a simplified
payload capacity over other legged robots. Animals have passive quadrupedal running robot. Although expressions for exact maps
elements such as elasticity in muscles, tendons, etc. In the case could not be obtained, approximate maps for bounding and
of quadrupeds, it can be done by replacing the rigid legs with pronking were derived and used to predict the behaviour of
compliant ones. Compliant elements such as springs, dampers, the different running parameters. Raibert et al. [7] made a four
etc. can be used for making the compliant legs. Several attempts legged walking robot named Big Dog. Their control system leans
are being made for power autonomous and passive motion. These the quadruped forward while climbing slopes, leans the body
kinds of robots can be used to travel in rocky, muddy and sloped backwards while descending slopes, and leans it sideways while
terrains. walking along the contour line. Zhang et al. [8] did the study on a
Miura and Shimoyama [1] made the first dynamically stable compact, four legged machine called RUSH with only one actuator
Biped BIPER, consisting of three motors. Raibert [2] worked per compliant leg. They constructed RUSH to study autonomous
on a dynamically stable quadruped, consisting of hydraulically and efficient running on flat and rough terrain. Chatzakos and
actuated and one passive prismatic DOF per leg. The robot was Papadopoulos [9] attempted to set the basis for a systematic
made to pace, bound and trot through a three part controller. approach in designing quadruped robots employing a dynamically
Ahmadi and Buehler [3] presented a control strategy for a one stable quadruped running in the sagittal plane with a bounding
legged running robot with compliant elements. They designed gait which is a simple model commonly used to analyze the basic
a controller which was based on the online calculations of the qualitative properties of quadruped gaits that use the legs in pair.
desired passive dynamic motion. Papadopoulos and Buehler [4] Pongas et al. [10] made a four legged walking robot named Little
considered a four legged walking robot named Scout II with a Dog that uses the robust quadruped walking gait for traversing
hip actuator and an additional compliant prismatic joint for each rough terrain. Yi [11] proposed a reliable gait planning and control
for a miniaturised quadruped robot pet. The reliability of his
proposed algorithm was ensured at two levels: trajectory planning
✩ Part of this paper appeared in the 15th National Conference on Machines &
and joint control. Radkhah et al. [12] designed a biologically
inspired compliant four legged robot. They summarised some
Mechanisms 2011, IIT Madras, Chennai, India.
∗ Corresponding author. basic principles of legged locomotion in animals and discussed the
E-mail addresses: [email protected] (K. Ganesh Kumar), application of the principles to the design and fabrication of a four
[email protected], [email protected] (P.M. Pathak). legged robot.
0921-8890/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.robot.2012.09.025
222 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228

In the previous work, the various factors such as masses of legs, Notation for parameters
friction, impact force and inclined ground surface have not been Fi,f —Impact force at the front leg
considered. In this paper, dynamic modelling has been carried out τ1 —Torque applied at back leg (link 1)
for a quadruped with compliant legs by incorporating the above τ2 —Torque applied at back leg (link 2)
factors. Modelling has been made by considering the quadruped in τ3 —Torque applied at front leg (link 1)
sagittal plane with each leg having two links in which the legs with τ4 —Torque applied at front leg (link 2)
ground contact are the compliant ones. Simulation and animation t—Flight time
models have been made in the Simulink package and successfully
simulated. A controller has been made for controlling forward Fig. 2 shows the robot which is considered for dynamic
velocity by a PD controller. modelling and trajectory control. In dynamic modelling, masses of
the legs, locomotion in inclined plane, friction and impact forces
have been considered. The masses of the legs are assumed to be
2. Dynamic modeling with compliant legs concentrated on the toe to make a stable motion. The robot is
assumed to make locomotion without slip. The impact force acts
Dynamic modelling of a four legged jumping robot with on the robot along the bottom links of the leg. It is used to store
compliant legs is done for a bounding gait. For modelling, the the strain energy. Frictional force acts opposite to the direction
sagittal plane has been considered. Fig. 1 shows the different of motion. Each leg of the robot can be considered as a parallel
phases involved in a bounding gait. manipulator comprising of two links connected through a revolute
Notations used in the derivation of dynamic modelling are given joint. The XY coordinated system used for the dynamic modelling
below. They remain the same throughout the dynamic modelling is shown at the contact toe in Fig. 2. X represents the position of the
and control strategy. robot in the horizontal direction and Y represents the position of
the robot in the vertical direction. The state variables represented
Notation for state variables
in figures are represented by, θ as body pitch, θ1 as angle of rotation
θ –Body angle with respect to inclined plane
of back leg (link 1), θ2 as angle of rotation of back leg (link 2), θ3 as
θ1 —Angle of rotation of back leg (link 1)
angle of rotation of front leg (link 1), θ4 as angle of rotation of front
θ2 —Angle of rotation of back leg (link 2)
leg (link 2) and l2 as length of the leg which is making contact with
θ3 —Angle of rotation of front leg (link 1)
the ground.
θ4 —Angle of rotation of front leg (link 2)
l2,b —Back leg length of link 2 Eq. (1) shows the equation of motion in the simplified matrix
l2,f —Front leg length of link 2 form for stance phases of a four legged jumping robot with
compliant legs. It represents the dynamic equation of motion for
Notation for parameters
both back stance and front stance phases. The detailed derivation
L—Half the distance between the hip joints
for the back stance phase is given in Appendix A.1.
mb —Body mass
I—Body moment of inertia about the center of mass 4 
4 4 
4
rb —Body radius of gyration
 
Aij (q)q̈j + Bij (q̇j )q̇i + Gi = τi . (1)
k —Spring stiffness i =1 j =1 i=1 j=1
b—Damping coefficient
lo —Free leg length (zero spring force) The above equation is a second order, nonlinear differen-
l1 —Leg length of link 1 tial equation. In this Aij (q) represents the coefficient of iner-
m1 —Mass of leg link 1 tia, Bij represents the centrifugal and Coriolis force coefficient
m2 —Mass of leg link 2 and Gi represents the gravitational force. For back stance phases
I1 —Moment of inertia of link 1 (I1 = m1 l1 2 ) generalised coordinates of the quadruped are represented by
T
I2 —Moment of inertia of link 2 (I2 = m2 l2 2 ) l2,b , θ , θ1 , θ2

q = and the applied force/torque are repre-
θp —Inclination of plane w.r.to horizontal T
sented by τi = Fi,b , 0, τ1 , τ2 . For front stance phases gen-

θtd,n —Touchdown angle of leg at nth flight
eralised coordinates of the quadruped are represented as q =
θtd,n+1 —Touchdown angle of leg at (n + 1)th flight T
l2,f , θ , θ3 , θ4

θlo,1b —Lift-off angle rotation of back leg (link 1) and the applied force/torque is represented as
T
θlo,2b —Lift-off angle of rotation of back leg (link 2) τi = Fi,f , 0, τ3 , τ4 .

θlo,3b —Lift-off angle of rotation of front leg (link 1) Equation of motion for flight after back stance phase can be
θlo,4b —Lift-off angle of rotation of front leg (link 2) written as
xi —Body Cartesian coordinates from the supporting toe (i)  
Xb —X value of the back leg Xlo + Ẋlo t
Xf —X value of the front leg Cartesian coordinates P = 1 (2)
− gt 2 + Ẏlo t + Ylo
Xlo —X value at lift-off 2
Xtd —X value at touchdown  
Ẋlo
Xfl —X value during the flight phase Velocities Ṗ = (3)
Xs —X value during the stance phase −gt + Ẏlo
Xtd,n+1 —Horizontal position of tip at (n + 1)th flight Body pitch θ = θlo,b + θ̇lo,b t (4)
Xtd —Horizontal position of tip at nth flight
Xa —Actual position of body in X direction Body pitch speed θ̇ = θ̇lo,b (5)
Xd —Desired position of body in X direction Leg angle θ1 = θlo,1b + θ̇1b t θ2 = θlo,2b + θ̇2b t
Ẋa —Actual Velocity of body in X direction
Ẋd —Desired Velocity of body in X direction θ3 = θlo,3b + θ̇3b t θ4 = θlo,4b + θ̇4b t (6)
KP —Constant for Proportional controller Leg angle speed θ1 = θ̇1b θ2 = θ̇2b
KD —Constant for Differential controller
Fi,b —Impact force at the back leg θ3 = θ̇3b θ4 = θ̇4b . (7)
 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228 223

Fig. 1. The four running states of a bounding gait.

Fig. 2. Sagittal plane model of quadruped.

Eqs. (2)–(7) represents the equation of motion for flight phases.

It represents the coordinates and rotation of the quadruped body
and leg, also its derivatives.

3. Controllers

Forward velocity can be controlled by changing the touchdown

angle. Energy required for a bounding gait is mainly obtained by
passive dynamics. Strain energy stored in the spring can be used for
flight phases. Fig. 3 Shows the leg position at touchdown. Energy
stored in the spring can be varied by changing the foot position
at touchdown. During flight the leg actuator is actuated to the
required touchdown angle. Fig. 3. Leg position at touchdown.
In closed loop controllers, touchdown angle can be controlled
by the PD controller, when the quadruped is in flight phase.
where, Xtd,n+1 is the position of the tip at the (n + 1)th phase and
Touchdown angle can be calculated for the forward speed from
Xtd,n is the position of tip at the nth phase.
Ahmadi’s expression [3].
A proportional derivative (PD) controller can be implemented to So, touchdown angle becomes,
obtain the desired leg touchdown position. The difference between  
KP KD
Ahmadi’s controller and our controllers is the passive term. Also, θtd,n+1 = sin −1
sin(θtd,n ) + (Xd − Xa ) + (Ẋd − Ẋa ) . (9)
the controller is extended from single leg to quadruped. In this l0 l0
controller, leg position at the last flight position is considered.
Whenever the robot’s current forward speed falls below the
Xtd,n+1 = Xtd,n + KP (Xd − Xa ) + KD (Ẋd − Ẋa ) (8) desired one, the leg touchdown angle reduces and vice-versa.
224 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228

Table 1
Initial conditions of state variables.
a
State variables θ θ1 θ2 θ3 θ4 Back leg Front leg
length length

Initial conditions 0 rad 0 rad 0 rad 0 rad 0 rad 0.65 m 0.65 m

Table 2
Physical parameters of quadruped.
Parameter Symbol Value

Body mass mb 23 kg
Leg mass of link 1 ml 0.85 kg
Leg mass of link2 m2 0.82 kg
b
Body inertia I 1.091 kg m2
Leg inertia for link 1 Il 0.020 kg m2
Leg inertia for link 2 I2 0.019 kg m2
Hip length 2L 0.8 m
Leg length link 1 l1 0.325 m
Leg length link 2 l2 0.325 m
Friction factor µ 0.35
Leg damping constant b 30 N s/m
Leg spring constant K 3600 N/m

4. Simulation and Results c

The Simulink model has been made in the MATLAB/Simulink

package. It represents the dynamics equations of motion of a
bounding gait.
Once the Simulink model is made, it is simulated for the set of
initial conditions. While starting the simulation it is assumed that
the robot is in home position. So the set of initial conditions for the
quadruped is given in Table 1.
The parameters of quadruped used in the simulations are given
in Table 2. It is decided according to the required quadruped speed, d
flight distance required and physical property of the steel and
spring steel.
In the open loop controller, the input values are given during
stance and flight phases. The values of lift off angles for back leg and
front leg are 20 deg and 25 deg. The stance torque values applied
are τ1 = 15 N m, τ2 = 30 N m, τ3 = 25 N m and τ4 = 50 N m.
During stance phase the energy is stored in the spring due to the
impact of the robot with the ground. Also, torque is applied on
the legs when additional energy is required. Energy stored during
stance phases is used for jumping during flight phases. In flight e
phases, the motor is asked to rotate to the required angle. During
stance phases, the robot follows the dynamics of corresponding
stance phases,. It follows flight dynamics for the flight phases.
Fig. 4 shows the variation of body pitch and leg angles with
respect to time. Graphs shown on left represent state variables
for complete gait cycles used in the simulation which shows
the variation in complete simulation, whereas Poincare maps are
shown on the right for the 10th and 11th bounding gait cycle which
clearly show the difference between the different phases of the
bounding gait. Legs are rotating either clockwise or anti-clockwise Fig. 4. Time in seconds vs. body pitch and leg angles for bounding gait.
depending on the phases. It is clear that during the back stance
phase, link 2 of the back leg is subjected to compression and the
motor applies torque at the back leg. When phase change occurs, It is observed that the robot makes the locomotion for the
the graph also changes its nature accordingly. constant input given. In real cases the external factors influence the
Fig. 5 shows the variation of position and velocity of body and robot’s performances. So the feedback controller is required for the
legs with respect to time. Graphs shown on the left represent desired output.
position and velocities of the robot body and legs in the horizontal Fig. 6 shows the forward velocity of the quadruped which
direction, whereas the right side represents the same in the vertical is controlled by a closed loop PD controller. Desired velocity is
direction. It shows that back leg tip position has been taken as the obtained by measuring the distance covered by the robot. The
reference axis and it is 0.4 m in the X direction and 0.65 m in the desired velocity of the robot during the flight phase is 2.5 m/s up
Y direction. It almost covers 16 m in just 10 s. It consumed only to 6.2 s. After that it has to be 1.6 m/s. The error signal is generated
16 bounding gait cycles. Average velocity of the robot is around by subtracting the actual position from the desired position. So the
1.5 m/s. The speed of the robot cannot be varied as required, as it touchdown angle is changed which makes an impact on forward
works on the open loop controller. velocity and the expected speed is achieved in a few running cycles.
 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228 225

a b

c d

e f

g h

Fig. 5. Time in seconds vs. position and velocity of body and legs.

Fig. 9 shows the snapshots taken when the quadruped makes

the locomotion on an inclined surface with different phases and
times. Simulation was done on an inclined surface of 0.1 radian.

5. Conclusion

Dynamic modelling of the quadruped is done by considering

various factors such as masses of the legs, inclination of the
plane, friction, impact force, etc. It resulted that the robot ran at
approximately 1.5 m/s which was achieved just in a 16 bounding
gait cycle. It was animated successfully which gives a clear view of
Fig. 6. Time in seconds vs. body velocity in m/s for forward velocity controller.
the running cycle.
In this paper a PD control algorithm has been used to control the
Then the touchdown angle is increased, so the forward speed is forward speed. The algorithm was implemented in the simulated
reduced to 1 m/s. model.
Fig. 7 shows the model of the quadruped used for the animation. Quadrupeds with compliant legs were successfully simulated
A four legged walking robot with flexible legs is animated by and animated. Future work will be to verify this experimentally.
providing the position and angle of rotation made by the body and
legs. All these parameters vary as the time progresses. A simulink Appendix
package is used for animating a quadruped.
Fig. 8 shows the snapshots taken when the quadruped makes A.1. Equation of motion for Back stance phase
the locomotion (animation) in a flat surface for different phases.
These snapshots make it clear all the phases of the bounding gait The equations of motion are derived for the back stance phase
have been implemented successfully in a quadruped. model of quadrupeds. The generalised coordinate vector q for this
226 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228

Fig. 7. Model of the quadruped used in animation.

Fig. 8. Snapshots of bounding gait when locomotion is in a flat surface.

Fig. 9. Snapshots of bounding gait when locomotion is on an inclined surface.

system is chosen to be q = [l2 , θ , θ1 , θ 2 ]T . In terms of generalised × sin(θ1 − θ2 ) + Ll̇2 θ̇ cos(θ1 − θ2 ) − l1 l2 (θ̇1 − θ̇)
coordinates, the dynamics equations of motion can be derived as 
follows: × (θ̇ + θ̇2 − θ̇1 ) cos(θ2 ) − l1 l̇2 (θ̇1 − θ̇ ) sin(θ2 ) .
Cartesian coordinates for body Potential Energy of the body
The body Cartesian coordinates can be expressed at the contact

Vbb = mb g L sin(θ + θp ) + l1 cos(θ1 − (θ + θp ))
point of the back leg’s toe and the ground as follows,

L cos(θ + θp ) + l1 sin θ1 − (θ + θp )  + l2 cos(θ2 − θ1 + (θ + θp )) .
  
−l2 sin(θ2 − θ1 + (θ + θp )) 
 
[Pbb ] =  . By obtaining the kinetic and potential energy for all the legs, the

L sin(θ + θ ) + l cos(θ − (θ + θ ))
p 1 1 p
Lagrangian function becomes:
+l2 cos(θ2 − θ1 + (θ + θp ))
1  ˙2 2
These can be differentiated with respect to time, in order to give
L = θ L (mb + 4 m1 + 4m2 ) + (θ̇1 − θ̇ )2 l1 2 (mb + m1 + m2 )
2
the body velocities in the x- and y-directions, 
  + l2 2 (θ̇ + θ̇2 − θ̇1 )2 (mb + 2m1 + m2 )
Ṗbb
+ l̇2 2 (mb + 2m1 + m2 ) + mb rbf 2 θ˙2 + m1 rf 1 2 θ̇1 2
−Lθ̇ sin(θ + θp ) + l1 (θ̇1 − θ̇ ) cos(θ1 − (θ + θp ))
 

 −l2 (θ̇2 − θ̇1 + θ̇ )  + m2 rf 2 2 θ̇2 2 + l1 2 θ˙2 (m1 + m2 ) + m2 l0 2 θ˙2
cos(θ2 − θ1 + (θ + θp )) − l̇2 sin(θ2 − θ1 + (θ + θp ))
 
= .

 Lθ̇ cos(θ + θp ) − l1 (θ̇1 − θ̇ ) sin(θ1 − (θ + θp ))  + (mb + 2m1 + 2m2 ) Ll1 θ̇ (θ̇ − θ̇1 ) sin(θ1 )
−l2 (θ̇ + θ̇2 − θ̇1 )
  
sin(θ2 − θ1 + (θ + θp )) + l̇2 cos(θ2 − θ1 + (θ + θp )) + Ll2 θ̇ (θ̇ + θ̇2 − θ̇1 ) sin(θ1 − θ2 ) + Ll̇2 θ̇ cos(θ1 − θ2 )

Kinetic Energy of the body can be expressed as, + (mb + m1 + m2 ) l1 l2 (θ̇ − θ̇1 )(θ̇ + θ̇2 − θ̇1 ) cos(θ2 )
1  
Tbb = mb L2 θ̇ 2 + l1 2 (θ̇1 − θ̇ )2 + l2 2 (θ̇ + θ̇2 − θ̇1 )2 + l̇2 2 + l1 l̇2 (θ̇ − θ̇1 ) sin(θ2 )
2
 
+ rb 2 θ̇ 2 + 2 −Ll1 θ̇ (θ̇1 − θ̇ ) sin(θ1 ) + Ll2 θ̇ (θ̇ + θ̇2 − θ̇1 ) + (m1 + m2 ) −2Ll1 θ˙2 sin(θ3 ) − l1 2 (θ̇ − θ̇1 )
 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228 227

× θ̇ cos(θ1 + θ3 ) + l1 l̇2 θ̇ sin(θ1 − θ2 + θ3 ) + (m1 + m2 )l1 [−l̇2 (2θ̇ − θ̇1 + θ̇2 ) cos(θ1 − θ2 + θ3 )
+ l2 (2θ̇ − θ̇1 + θ̇2 )(θ̇1 − θ̇2 ) sin(θ1 − θ2 + θ3 ) + l̇2 (θ̇1 − θ̇2 )

− l1 l2 (θ̇ + θ̇2 − θ̇1 )θ̇ cos(θ1 − θ2 + θ3 )
 × cos(θ1 − θ2 + θ3 )] + m2 l0 l1 θ̇1 (2θ̇ − θ̇1 ) sin(θ1 + θ 3 −θ4 )
+ m2 −2Ll0 θ˙2 sin(θ4 − θ3 ) + l1 l0 (θ̇1 − θ̇ )θ̇ + m2 l0 [−l̇2 (2θ̇ − θ̇1 + θ̇2 ) cos(θ2 − θ1 + θ4 − θ3 )
× cos(θ1 + θ3 − θ4 ) − l0 l̇2 θ̇ sin(θ2 − θ1 + θ4 − θ3 ) + l2 (2θ̇ − θ̇1 + θ̇2 ) sin(θ2 − θ1 + θ4 − θ3 ) + l̇2 (θ̇1 − θ̇2 )
 × sin(θ2 − θ1 + θ4 − θ3 )] + g [(mb + 2m1 + 2m2 )L
− l2 l0 θ̇ (θ̇ + θ̇2 − θ̇1 ) cos(θ2 − θ1 + θ4 − θ3 )
× cos(θ + θp ) + (mb + m1 + m2 )l1 sin(θ1 − (θ + θp ))
+ l1 l0 θ̇ 2 cos(θ4 ) − g (mb + 2m1 + 2m2 )L sin(θ + θp )

− (mb + 2m1 + m2 )l2 sin(θ2 − θ1 + θ + θp ) + (m1 + m2 )l1
+ (mb + m1 + m2 )l1 cos(θ1 − (θ + θp )) × sin(θ + θp +θ 3 ) − m2 l0 sin(θ4 − (θ + θp +θ 3 ))] = 0.
− (m1 + m2 )l1 cos(θ3 + θ + θp ) − m2 l0
The equation of motion for generalised Coordinate θ1 becomes:
× cos(θ4 − (θ3 + θ + θp )) + (mb + 2m1 + m2 )l2 
 1 [−l1 (mb + m1 + m2 ) sin(θ2 )] l̈2 + −l1 2 (mb + m1 + m2 )
× cos(θ2 − θ1 + θ + θp ) − K (l2 − l0 )2 .
2
− l2 2 (mb + 2m1 + m2 ) − Ll1 (mb + 2m1 + 2m2 ) sin(θ1 )
The equation of motion for generalised Coordinate l2 becomes:
 − 2l1 l2 (mb + m1 + m2 ) cos(θ2 ) + (m1 + m2 )l1 2
(mb + 2m1 + m2 )l̈2 + (mb + 2m1 + 2m2 )L cos(θ1 − θ2 ) × cos(θ1 + θ3 ) + (m1 + m2 )l1 l2
+ (mb + m1 + m2 )l1 sin(θ2 ) + (m1 + m2 )l1 × cos(θ1 − θ2 + θ3 ) + m2 l1 l0 cos(θ1 + θ3 − θ4 ) + m2 l2 l0
 × cos(θ2 − θ1 + θ4 − θ3 ) − Ll2 (mb + 2m1 + 2m2 )
× sin(θ1 − θ2 + θ3 ) − m2 l0 sin(θ2 − θ1 + θ4 − θ3 ) θ̈ 
× sin(θ1 − θ2 ) θ̈
− [(mb + m1 + m2 )l1 sin(θ2 )] θ̈1 − [(mb + 2m1 + m2 )] 
× l2 (θ̇ + θ̇2 − θ̇1 )2 − (mb + 2m1 + 2m2 )Lθ̇ 2 sin(θ1 − θ2 ) + l1 2 (mb + m1 + m2 ) + m1 r1 2 + l2 2 (mb + 2m1 + m2 )
− (mb + m1 + m2 )l1 (θ̇ − θ̇1 )2 cos(θ2 ) + (m1 + m2 )l1 θ̇ 2 
+ 2l1 l2 (mb + m1 +m2 ) cos(θ2 ) θ̈1
× cos(θ1 − θ2 + θ3 ) + m2 l0 θ̇ 2 cos(θ2 − θ1 + θ4 − θ3 )
+ g (mb + 2m1 + m2 ) cos(θ2 − θ1 + θ + θp )

+ −l2 2 (mb + 2m1 + m2 )
+ K (l0 − l2 ) + bl˙2 
= Fi cos(θ2 − θ1 + θ + θp ) + µFi cos(θ2 − θ1 + θ + θp ). − l1 l2 (mb + m1 + m2 ) cos(θ2 ) θ̈2

The equation of motion for generalised Coordinate θ becomes: − 2(mb + 2m1 + m2 )l2 l̇2 (θ̇ + θ̇2 − θ̇1 )
[(mb + 2m1 + 2m2 )L cos(θ1 − θ2 ) + (mb + m1 + m2 )l1 sin(θ2 ) + (mb + 2m1 + 2m2 )
× −Ll1 θ̇ 2 cos(θ1 ) − Ll2 θ̇ 2 cos(θ1 − θ2 )
 
+ (m1 + m2 )l1 sin(θ3 + θ1 − θ2 )
− m2 l0 sin(θ2 − θ1 + θ4 − θ3 )] + (mb + m1 + m2 )l1
× l̇2 (2θ̇2 − 2θ̇1 + 2θ̇ ) cos(θ2 )

× l̈2 + [(mb + 4m1 + 4m2 )L2 + (mb + m1 + m2 )l1 2
+ l2 θ̇2 (θ̇2 − 2θ̇1 + 2˙θ ) sin(θ2 )

+ (mb + 2m1 + m2 )l2 2 + (m1 + m2 )l1 2 + m2 l0 2
+ 2(mb + 2m1 + 2m2 )Ll1 sin(θ1 ) + 2(mb + 2m1 + 2m2 )

+ (m1 + m2 ) −l1 2 θ̇ 2 sin(θ1 + θ3 ) − l1 l2 θ̇ 2
× Ll2 sin(θ1 − θ2 ) + 2(mb + m1 + m2 )l1 l2 cos(θ2 ) 
− 4(m1 + m2 )Ll1 sin(θ3 ) − 2(m1 + m2 )l1 2 cos(θ1 +θ 3 ) × sin(θ1 − θ2 + θ3 )
− 2(m1 + m2 )l1 l2 cos(θ1 − θ2 + θ3 ) − 4m2 Ll0 sin(θ4 − θ3 ) 
+ m2 −l1 l0 θ̇ 2 sin(θ1 + θ3 − θ4 ) + l0 l2 θ̇ 2
− 2m2 l0 l1 cos(θ1 + θ 3 −θ4 ) − 2m2 l0 l2 cos(θ2 − θ1 + θ4 − θ3 ) 
+ 2m2 l0 l1 cos(θ4 ) + mb rb 2 ] θ̈ + [−(mb + m1 + m2 )l1 2 × sin(θ2 − θ1 + θ4 − θ3 )
− (mb + 2m1 + m2 )l2 2 − (mb + 2m1 + 2m2 )Ll1 sin(θ1 ) 
− (mb + 2m1 + 2m2 )Ll2 sin(θ1 − θ2 ) + g −(mb + m1 + m2 )l1 sin(θ1 − (θ + θp ))
− 2(mb + m1 + m2 )l1 l2 cos(θ2 ) + (m1 + m2 )l1 2 cos(θ1 +θ 3 )

+ (mb + 2m1 + m2 )l2 sin(θ2 − θ1 + θ + θp ) = τ1 .
+ (m1 + m2 )l1 l2 cos(θ1 − θ2 + θ3 ) + m2 l0 l1 cos(θ1 + θ 3 −θ4 )
+ m2 l0 l2 cos(θ2 − θ1 + θ4 − θ3 )] θ̈1 + [(mb + 2m1 + m2 )l2 2 The equation of motion for generalised Coordinate θ2 becomes:
+ (mb + 2m1 + 2m2 )Ll2 sin(θ1 − θ2 ) + (mb + m1 + m2 ) [(mb + 2m1 + m2 )l2 2 + (mb + 2m1 + 2m2 )Ll2 sin(θ1 − θ2 )
× l1 l2 cos(θ2 ) − (m1 + m2 )l1 l2 cos(θ1 − θ2 + θ3 ) + (mb + m1 + m2 )l1 l2 cos(θ2 )
− m2 l0 l2 cos(θ2 − θ1 + θ4 − θ3 )]θ̈2 + 2(mb + 2m1 + m2 ) − (m1 + m2 )l1 l2 cos(θ1 − θ2 + θ3 ) − m2 l0 l2
× l2 l̇2 (θ̇ − θ̇1 + θ̇2 ) + (mb + 2m1 + 2m2 )Ll1 θ̇1 (2θ̇ − θ̇1 )

× cos(θ2 − θ1 + θ4 − θ3 )] θ̈ + −l2 2 (mb + 2m1 + m2 )
× cos(θ1 ) + (mb + 2m1 + 2m2 )L[l̇2 (2θ̇ − θ̇1 + θ̇2 )  
× sin(θ1 − θ2 ) + ( θ̇1 − θ̇2 )(2θ̇ − θ̇1 + θ̇2 ) cos(θ1 − θ2 ) − l1 l2 (mb + m1 + m2 ) cos(θ2 ) θ̈1 + l2 2 (mb + 2m1 + m2 )
− l̇2 (θ̇1 − θ̇2 ) sin(θ1 − θ2 )] + (mb + m1 + m2 )l1 [−l2 θ̇2 
+ m2 r2 2 θ̈2 + (mb + 2m1 + 2m2 )Ll2 θ̇ 2 cos(θ1 − θ2 )
× (2θ̇ − 2θ̇1 + θ̇2 ) sin(θ2 ) + l̇2 (2θ̇ − 2θ̇1 + θ̇2 ) cos(θ2 )
+ l̇2 θ̇2 cos(θ2 )] + (m1 + m2 )l1 2 θ̇1 (2θ̇ − θ̇1 ) sin(θ1 +θ 3 ) + (mb + m1 + m2 )l1 l2 (θ̇ − θ̇1 )2 sin(θ2 )
228 K. Ganesh Kumar, P.M. Pathak / Robotics and Autonomous Systems 61 (2013) 221–228

+ (m1 + m2 ) l1 l2 θ̇ 2 sin(θ1 − θ2 + θ4 )
 
[9] Panagiotis Chatzakos, Evangelos Papadopoulos, Bio-inspired design of
electrically-driven bounding quadrupeds via parametric analysis, Mechanism
+ m2 l0 −l2 θ̇ 2 sin(θ2 − θ1 + θ4 − θ3 )
 
and Machine Theory (2009) 559–579.
[10] Dimitris Pongas, Michael Mistry, Stefan Schaal, A robust quadruped walking
− g (mb + 2m1 + m2 )l2 sin(θ2 − θ1 + θ + θp )
 
gait for traversing rough terrain, in: IEEE International Conference on Robotics
and Automation, Rome, Italy, 10–14 April 2007, pp. 1474–1479.
+ 2(mb + 2m1 + m2 )l2 l̇2 (θ̇ + θ̇2 − θ̇1 ) = τ2 . [11] Sooyeong Yi, Reliable gait planning and control for miniaturized quadruped
robot pet, Mechatronics (2010) 485–495.
Similarly, equations of motion can be derived for the front [12] Katayon Radkhah, Stefan Kurowski, Oskar von Stryk, Design considerations
stance phase. for a biologically inspired compliant four-legged robot, in: IEEE International
Conference on Robotics and Biomimetics, Guilin, China, 19–23 December,
2009.
References

Ganesh Kumar K. was a M. Tech Student with CAD, CAM

[1] H. Miura, I. Shimoyama, Dynamic walk of a biped, International Journal of
and Robotics specialisation at the Mechanical & Industrial
Robotics Research 3 (2) (1984) 60–74. Summer. Engineering Department, IIT Roorkee, India.
[2] M.H. Raibert, Legged Robots That Balance, MIT Press, Cambridge, MA, 1986.
[3] M. Ahmadi, M. Buehler, Stable control of a simulated one—legged running
robot with hip and leg compliance, IEEE Transactions on Robotics and
Automation 13 (1) (1997).
[4] D. Papadopoulos, M. Buehler, Stable running in a quadruped robot with
compliant legs, in: International Conference of Robotics and Automation, vol.
1, San Francisco, 2000, pp. 444–449.
[5] N. Cherouvim, E. Papadopoulos, Single actuator control analysis of a planar
3DOF hopping robot, in: Robotics: Science and Systems, Massachusetts
Institute of Technology, Cambridge, Massachusetts, U.S.A, 8–11 June, 2005.
[6] M.D. Berkemeier, Approximate return maps for quadrupedal running, in: IEEE
International Conference on Robotics and Automation, Albuquerque, New Pushparaj Mani Pathak is the Associate Professor of
Mexico, April 1997, pp. 805–810. Mechanical & Industrial Engineering at the Mechanical &
[7] M. Raibert, K. Blankespoor, G. Nelson, R. Playter and the BigDog Team, Industrial Engineering Department, IIT Roorkee, India. His
BigDog, the rough-terrain quadruped robot, in: Proceedings of the 17th World area of interest are space robots, walking robots, in-vivo
Congress, The International Federation of Automatic Control, Seoul, Korea, robots modeling and control.
6–11 July, 2008, pp. 10822–10825.
[8] Z. Zhang, T. Masuda, H. Kimura, K. Takase, Towards realization of adaptive
running of a quadruped robot using delayed feedback control, in: IEEE
International Conference on Robotics and Automation, Rome, Italy, 10–14
April 2007.