382 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 5, NO.

1, JANUARY 2018

Bionic Quadruped Robot Dynamic Gait Control

Strategy Based on Twenty Degrees of Freedom
Dawei Gong, Peng Wang, Shuangyu Zhao, Li Du, and Yu Duan

Abstract—Quadruped robot dynamic gaits have much more gait has more advantages on speed and efficiency that makes
advantages than static gaits on speed and efficiency, however it become a hot topic for research [3].
high speed and efficiency calls for more complex mechanical Nowadays, scholars have obtained some achievements in
structure and complicated control algorithm. It becomes even
more challenging when the robot has more degrees of freedom. quadruped robot research. In 2004, Hornby et al. realized
As a result, most of the present researches focused on simple pace gait and trot gait in a straight line on Sony’s two small
robot, while the researches on dynamic gaits for complex robot quadruped robot ERS-110 and OPEN-R using evolutionary
with more degrees of freedom are relatively limited. The paper algorithm. Both of them have three degrees of freedom in
is focusing on the dynamic gaits control for complex robot with leg, two in hip and one in knee [4]. In 2011, Wang et al.
twenty degrees of freedom for the first time. Firstly, we build
a relatively complete 3D model for quadruped robot based on realized bound gait in a simple quadruped robot which has
spring loaded inverted pendulum (SLIP) model, analyze the four passive legs and a spine connecting front and rear body
inverse kinematics of the model, plan the trajectory of the swing [5]. Each hip is equipped with only one drive. Quadruped
foot and analyze the hydraulic drive. Secondly, we promote the robot models used in the above researches are simplified in
control algorithm of one-legged to the quadruped robot based on leg structure and degrees of freedom compared to tetrapods.
the virtual leg and plan the state variables of pace gait and bound
gait. Lastly, we realize the above two kinds of dynamic gaits in Due to this reason, this model-based control algorithms have
ADAMS-MATLAB joint simulation platform which testify the few limitations to complex road. In 2008, Ma et al. proved
validity of above method. that quadruped robot is able to achieve bound gait under the
Index Terms—Degree of freedom, dynamic gait, foot trajectory, initial condition of a fixed point by numerical calculation [6].
quadruped robot, spring loaded inverted pendulum (SLIP). However, this conclusion has not been realized in simulation
platform.
In summary, the paper presents a complex quadruped robot
I. I NTRODUCTION model with 20 degrees of freedom which could contain all

M ORE and more researchers have been giving great the above models. Each leg in the model has 5 degrees of
attention to the robot research [1]. Quadruped robot freedom, a yaw and a pitching freedom of hip, a pitching
which can be used to simulate the common animals becomes freedom of knee and ankle. And each leg also has an elastic
a hot topic in recent time [2]. It has higher carrying capacity component simulating leg tendon in motion as a passive degree
and stability compared with single-legged and biped robot. of freedom. The model retains basic motion characteristics
And it also reduces redundancy in structure design compared of quadruped leg movement that can simulate four-legged
to hexapod and more legs robot. So the quadruped robot has animals movement very well. Finally, we realize a better
been studied widely and developed rapidly in the recent years. pace gait and bound gait in the ADAMS-MATLAB simulation
Quadruped robot is inspired by four-legged animals not only platform that is proving the feasibility of technical route.
in structure but also in the ways of moving. The common Although the technical route making pace gait and bound gait
gaits include crawl, trot, pace, bound and so on. In which, is an example, it also applies to dynamic gait such as trot
crawl is static gait, and others are dynamic gaits. Static gait gait. The research of trot gait is so mature that it is not being
is the most stable gait in quadruped movement. But dynamic focused on in this paper.
In the current research, scholars focus more on flexible
Manuscript received May 27, 2016; accepted August 8, 2016. This work contact and gait control of single-legged robot. But, it lacks
was supported by the National Science Fund for Distinguished Young Scholars attention of the effect on high speed contact, with feet and
of China (51225503), the National Natural Science Foundation of China (616
03076), and the Fundamental Research Funds for the Central Universities ground, in quadruped robot dynamic gait. And we also ignore
(ZYGX2016J116). Recommended by Associate Editor Chengdong Li. (Cor- how to achieve stabilizing control. Due to this reason, we build
responding author: Dawei Gong.) a relatively complete 3D model for quadruped robot based
Citation: D. W. Gong, P. Wang, S. Y. Zhao, L. Du, and Y. Duan, “Bionic
quadruped robot dynamic gait control strategy based on twenty degrees of on SLIP model, analyze the inverse kinematics of the model,
freedom,” IEEE/CAA J. of Autom. Sinica, vol. 5, no. 1, pp. 382−388, Jan. plan the trajectory of the dynamic gait with 0 velocity and
2018. acceleration when touch the ground and analyze the hydraulic
The authors are with the University of Electronic Science and Technology,
Chengdu 611731, China (e-mail: [email protected]; peng wang uestc@out drive. At last, we achieve a better simulation of pace gait and
look.com; [email protected]; [email protected]; derekduan jy@163. bound gait. The paper is focusing on the dynamic gaits control
com). in such a complex situation for the first time. This can provide
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. reference for quadruped robot design.
Digital Object Identifier 10.1109/JAS.2017.7510790
GONG et al.: BIONIC QUADRUPED ROBOT DYNAMIC GAIT CONTROL STRATEGY BASED ON TWENTY DEGREES OF FREEDOM 383

II. SLIP M ODEL III. PATH P LANNING FOR Q UADRUPED ROBOT

Spring loaded inverted pendulum (SLIP) is a very common DYNAMIC G AIT
model in research of biological running and jumping, with A. Gait Description
few degrees of freedom, simple structure, less coupling and
simple control, as shown in Fig. 1 [7]. The motion of planar The gait is a coordinative relationship of animals limb be-
SLIP model can be decomposed into the forward direction (X tween time and space. It includes regularly repeating sequence
direction) and the vertical direction (Y direction). Its motion and method of moving legs.
has the standing and vacating phase. Running or jumping could Quadruped robot gait can be divided into static and dynamic
be considered as alternation between two states. The 3D-SLIP gait. Static gait means projection of robot gravity on the
model also has speed in lateral direction (Y direction) except ground is always located within the supporting triangle. It has
in forward direction (X direction) and vertical direction (Z the slowest speed and the best stability. Crawl is a kind of static
direction). Compared to the planar SLIP model, it is a more gait. Common dynamic gait includes trot, pace, bound and
appropriate choice for bionic robot research. gallop, in which, gallop is asymmetric gait with time difference
of any two legs movement. The other three are symmetrical
gaits.
Trot gait means that the two legs on diagonal have consis-
tently movement. And they are as standing and vacating legs
by turns.
Pace gait means same-side feet having synchronous motion.
The body during exercise has significantly tumbling motion
which causes less stability. But it avoids the mutual inter-
ference between front and rear legs like trot. So, it is more
suitable for camels, giraffes and other long leg quadrupeds.
Bound gait means front and rear legs are paired respectively.
It suits for smaller flexible quadruped with poor walking
ability. The body has an obvious pitching during motion.
In this paper, we focus on control strategy of symmetrical
gait based on pace and bound gait.

B. Virtual Leg
Virtual leg is an important tool in research of multi-legged
robot, proposed by Sutherland in 1989. The virtual leg means
one virtual leg could instead of two real legs when the two real
legs are exerting equal force on the ground, producing same
horizontal displacement and equal torque on the hips. Different
Fig. 1. Planar SLIP model and 3D-SLIP model. quadruped robot gaits can be equivalent to the corresponding
biped robot gait by the principle of virtual leg, as shown in
A. SLIP-based Single-leg Robot Control Algorithm Fig. 3.
Single-legged robot control algorithm is the basis of legged
robot control which includes forward speed, body attitude and
jumping height control [8]−[10].

B. Quadruped Robot Modeling

Quadruped robot model from this paper is based on 3D-
SLIP. The 3D-model is provided by the Harbin Institute of
Technology, as shown in Fig. 2.

Fig. 2. Quadruped robot’s 3D model. Fig. 3. Application of the virtual leg in quadruped robot.
384 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 5, NO. 1, JANUARY 2018

And biped robot motion can be seen as two single-leg robots as sine curve learning from forward acceleration curve. We
movement whose each leg could as standing and vacating leg have:
by turns. So, we can establish connections between quadruped t
robot and single-leg robot by virtual leg to extend of control z̈ = A2 sin 4π . (7)
T
strategy from single-leg robot to quadruped robot. When t = 0 or t = T, Ż = 0, we can obtain:
µ ¶
A2 T t
C. Foot Trajectory Settings ż = 1 − cos 4π . (8)
4π T
During the forward movement, the trajectory of the swing
Boundary conditions are as follows:
foot needs to satisfy some conditions. 
1) Leg lifting must be enough high to prevent collision with z = 0, t = 0
ground and obstacles. T (9)
z = H, t = .
2) Velocity and acceleration of the swing foot are set to 0 2
to reduce the impact of collision with ground when starting The foot trajectory curve in the vertical direction is as
and hitting the ground. follows:
Using the simple curve (such as sine curve, elliptic curves) µ ¶
t 1 t
as the swinging foot trajectory curve, the foot will have an z = 2H − sin 4π (10)
T 4π T
additional acceleration at the moment it touching ground. So
we consider using composite cycloid instead of simple curve H is the height of the swing leg lift in the vertical direction.
in this paper. It is easy to validate the foot trajectory curve in the vertical
First, consider the robot forward foot trajectory. direction also satisfying the formula in the second half of the
According to the condition 2), the acceleration is set to 0 cycle.
when getting off and hitting the ground. The acceleration curve If we set T = 0.7 s, step length S = 450 mm and foot left
can be set to a sine curve. height H = 100 mm.
Fig. 4 shows the swinging foot trajectory based on com-
t
ẍ = A1 sin 2π . (1) posite cycloid planning. The x-axis represents the forward
T direction. And z-axis represents the vertical direction. And
The velocity is also 0 when leaving and touching the ground. Figs. 5 and 6 are displacement time series plots of forward
Integral the formula at t = 0 or t = T , ẋ = 0 . The following and vertical direction. The first and second derivatives of the
equation will be obtained: curve are both 0 at the start and the end that matches the foot
µ ¶ trajectory condition correctly.
A1 T t
ẋ = 1 − cos 2π . (2)
2π T
D. Inverse Kinematic Analysis of Quadruped Robot Leg
By the velocity curve, we can obtain foot trajectory curve
Kinematic analysis is the basis for robot path planning. We
as follows:
µ ¶ establish the relationship between the trajectory of foot and the
A1 T T t joint angle through it. And we could implement the expected
x= t− sin 2π + C1 . (3)
2π 2π T foot motion by controlling the joint angle [11], [12].
Boundary conditions are as follows: Structure of the single leg is shown in Fig. 7, where ΣN is
( global coordinate system; ΣB is moving coordinate system at
x = 0, t = 0 the center of the body. H, M, K, E, F represent yaw motion,
(4)
x = S, t = T. pitching motion of hip joint, knee, ankle and foot, respectively.
The lengths of the leg between joints are, in order, l0 , l1 , l2 ,
Substitute this into (3), we have: l3 . θ1 represents Yaw angle of hip. Angles of leg are θi =
2πS (θi1 , θi2 , θi3 ). Yaw, Pitch, Roll are named α, β, γ.
C1 = 0, A1 = . (5)
T2 The rotation matrix from ΣB to ΣN is
 
Therefore, we can obtain: cα cβ cα sβ sγ − sα cγ cα sβ cγ + sα sγ
µ ¶ N
RB =  sα cβ sα sβ sγ + cα cγ sα sβ cγ − cα sγ  .
t 1 t −sβ cβ sγ cβ cγ
x=S − sin 2π . (6)
T 2π T (11)
S is the robot step length. Assume the position of B and F are known as (xb , yb , zb )
Then, focus on the robot foot trajectory in the vertical and (xf , yf , zf ). In the XB -YB plane, the angle between foot
direction. and XB is ϕ. It is not difficult to get the conclusion according
Swinging leg movement in the vertical direction is symmet- to mechanism sketch
ric about the highest point. So only consider the case 0 ≤ t B−−→ −1 N − −→ −1 −−→ −−→ −−→
≤ T /2. HF = N RB · HF = N RB · (N OB + N BH − N OF ).
The velocity and acceleration of swinging leg in the vertical (12)
direction are both 0 at start of movement in every cycle and at Substituting each vector into the equation, inverse kinemat-
the highest point. So, we can describe vertical foot trajectory ics solution can be obtained as follows:
GONG et al.: BIONIC QUADRUPED ROBOT DYNAMIC GAIT CONTROL STRATEGY BASED ON TWENTY DEGREES OF FREEDOM 385

Fig. 4. The foot trajectory curve.

Fig. 7. Model of frame and structure of the single leg.

h21 − l12 − l22 − l32 + (h2 − l0 cos θ1 )2

θ12 = arccos
2l1 l2
π
θ13 = − φ − θ11 − θ12 (13)
2
where
     d 
h1 xb − xf 2
 h2  = N −1 
RB yb − yf  +  0  (14)
w
h3 zb − zf 2
d and w are the length and width of the body.
In the same way, we can obtain inverse kinematics solution
Fig. 5. Displacement time series plots in forward direction. of other 3 legs. In the case of foot trajectory is known, corre-
sponding joint angles can be calculated by inverse kinematics.

E. Analysis the Hydraulic Drive System

Quadruped robot model used in this paper is driven by
hydraulic cylinders. We can control the hydraulic position to
achieve the control of robot leg joints [13].
Each leg has four hydraulic cylinders for driving in this
model. The first one drives yawing motion of the hip. The
second one drives pitching motion of the hip. The third one
drives knee pitching movement. And the fourth one drives
ankle pitching motion.
First, calculate the first hydraulic cylinder elongation. We
intends to use cosine law to calculate the corresponding angle
which hydraulic cylinder in. It can be calculated by cosine
law through calculating the corresponding angle in the triangle
Fig. 6. Displacement time series plots in vertical direction. which includes hydraulic cylinder.
h3 According to Fig. 8, it is easy to obtain:
θ1 = arctan y
h2 ∠1 = arctan . (15)
h1 z
θ11 = arctan
h2 − l0 cos θ1 According to the relation between each angle in graph, we
have:
l1 + l2 cos θ12
− arccos p ∠ABC = π + ∠1 − ∠CBD − ∠ABZ. (16)
l12 + l22 + 2l1 l2 cos θ12
So
l2 + l2 + 2l1 l2 cos θ12 + h21 + (h2 − l0 cos θ1 )2 − l32 q
− arccos 1p 22 LAC = L2BC + L2AB − 2LBC LAB cos ∠ABC. (17)
2 (l1 + l22 + 2l1 l2 cos θ12 )(h21 + (h2 − l0 cos θ1 )2 )
386 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 5, NO. 1, JANUARY 2018

This is the extended length of second, third and fourth

hydraulic cylinder. Their elongations are respectively equal to
the result of extended length minus original length.

F. Planning of State Variables

The above analysis can fully describe the motion of the
robot’s single leg. The quadruped robot gait can be planned
based on it. The implementation process of quadruped robot
gait is actually the coordination process of four legs movement
along established path. In the control process, it is often useful
to define the movement in one stage as a state variable. By
controlling the state variable, the robot could move along the
established path.
In the example of pace and bound gait, we define the state
variable for every process of paired legs rise and fall. It can
be expressed as Fig. 10.

Fig. 8. Hydraulic cylinder of the first joint.

This is the extended length of first hydraulic cylinder.

Minusing original length, we can get hydraulic cylinder’s
elongation.
Fig. 9 shows hydraulic cylinder of the second, third and
fourth joints. From Fig. 9, it is easy to obtain:
∠1 = π − θ11 − ∠y1 Az0 − ∠y2 AB

∠2 = θ11 − ∠BAy4 − ∠y3 AE

∠3 = θ12 − ∠y5 BF − ∠y6 BC. (18) Fig. 10. Transition between the state variables.
According to the cosine law, we have: In this paper, the left front leg and left hind leg are the first
q
Ly1 y2 = L2Ay1 + L2Ay2 − 2LAy1 LAy2 cos ∠1 pair of legs in pace gait. The right front leg and right hind leg
are the second pair. In bound gait, the left front leg and right
q
front leg are the first pair of legs. And the right hind leg and
Ly3 y4 = L2Ay3 + L2Ay4 − 2LAy3 LAy4 cos ∠2
left hind leg are for the second pair. Four-leg support state is
q the initial state of every cycle. Cycles of four states form the
Ly5 y6 = L2By5 + L2By6 − 2LBy5 LBy6 cos ∠3. (19) forward gait.

IV. S IMULATION R ESULTS AND A NALYSIS OF G AIT

According to the above method, joint angles of each leg
can be figured out by foot trajectory. Then, the hydraulic
cylinder elongation as driving function can be obtained. At-
titude angles (yaw, pitch, roll) are adjusted by PID control
based on SLIP. Import the 3D-model of quadruped robot
into Adams and establish control variables and constraints.
Adopting the ADAMS and MATLAB joint simulation method,
we can obtain simulation graphics and parameters of robot
motion in ADAMS post-processing module.

A. Result of Pace Gait

As shown in Fig. 9, we begin from the state of four leg
on land. Then robot first raises left front leg and left hind leg
simultaneously. Then, the left front and left hind legs touch the
ground, and raise right front leg and right hind leg up. It fits in
with the characteristics of pace, reciprocating cycle. (Fig. 11
is in right view. The actual forward direction in simulation of
Fig. 9. Hydraulic cylinder of the second, third and fourth joints. the robot is in left in Fig. 11. So that the right front and right
GONG et al.: BIONIC QUADRUPED ROBOT DYNAMIC GAIT CONTROL STRATEGY BASED ON TWENTY DEGREES OF FREEDOM 387

hind leg in this right view is actually left front and left hind cylinder strobe, then we can adjust the joint angle through
leg of the robot, and vice versa). control of hydraulic cylinder elongation.
4) By combining with the pace and bound gait characteris-
tics, we can plan state variables and achieve two gaits cyclical
motion.
5) Based on the above analysis, we realize a better pace and
bound gait of quadruped robot in simulation platform proving
the feasibility of technical route.

Fig. 13. Four-leg state of bound gait.

Fig. 11. Four-leg state of pace gait.

As Fig. 12 shows, at the beginning of the movement, the

body’s pitch, roll and yaw have greater volatility. And then
the movement becomes smooth.

Fig. 14. Curve of attitude angle in bound gait.

R EFERENCES
[1] W. He, Y. T. Dong, and C. Y. Sun, “Adaptive neural impedance control
of a robotic manipulator with input saturation,” IEEE Trans. Syst. Man
Fig. 12. Curve of attitude angle in pace gait. Cybern. Syst., vol. 46, no. 3, pp. 334−344, Mar. 2016.
[2] Z. Y. Liu, D. D. Liu, and T. L. Shi, “Gait planning and movement
simulation of biped robots,” J. Wuyi Univ. (Nat. Sci. Ed.), vol. 30, no. 1,
B. Result of Bound Gait pp. 29−33, Mar. 2016.
Similar as Fig. 11, Bound Gait can be seen in Fig. 13. As [3] Z. W. Chen, B. Jin, S. Q. Zhu, H. L. Huang, and G. Chen, “Design and
experiment of single leg of hydraulically actuated bionic multi-legged
seen in Fig. 14, the body pitch and roll is better in convergence. robot,” Trans. Chinese Soc. Agric. Eng., vol. 32, no. 5, pp. 36−42, Mar.
However, due to control error, the yaw fluctuated slightly with 2016.
periodical change as same as the single-leg robot movement. [4] G. S. Hornby, S. Takamura, T. Yamamoto, and M. Fujita, “Autonomous
evolution of dynamic gaits with two quadruped robots,” IEEE Trans.
Robot., vol. 21, no. 3, pp. 402−410, Jun. 2005.
V. C ONCLUSION [5] X. Wang, M. T. Li, P. F. Wang, and L. N. Sun, “Running and turning
control of a quadruped robot with compliant legs in bounding gait,” in
1) Using composite cycloid to plan the swing foot trajectory Proc. 2011 IEEE Int. Conf. Robotics and Automation, Shanghai, China,
curve makes the velocity and acceleration to be 0 which can 2011, pp. 511−518.
reduce the impact of collision with ground and energy loss. [6] S. P. Ma, G. H. Miao, and L. Q. Yu, “Research on the leaping gait of
2) After setting the swing foot trajectory curve, we can four feet robot with articulated legs,” J. Machine Des., vol. 25, no. 8,
pp. 32−36, Oct. 2008.
get relationship between trajectory and joint angle by inverse
kinematic analysis for single leg model of quadruped robot. [7] B. B. Xu, Y. J. Li, and Z. Wang, “Control method of spring loaded
inverted pendulum model for legged robots,” Mech. Eng. Automat., no. 4,
This is the foundation of control for single leg motion. pp. 141−143, Aug. 2014.
3) Establishing the relationship between the joint angle and [8] M. T. Li, Z. Y. Jiang, W. Guo, and L. M. Sun, “Leg prototype of a bio-
the hydraulic cylinder elongation by analysing of hydraulic inspired quadruped robot,” Robot, vol. 36, no. 1, pp. 21−28, Jan. 2014.
388 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 5, NO. 1, JANUARY 2018

[9] M. Raibert, M. Chepponis, and H. Brown, “Running on four legs as Shuangyu Zhao is currently an Engineer at
though they were one,” IEEE J. Robot. Automat., vol. 2, no. 2, pp. 70− Shenyang Aircraft Design and Research Institute.
82, Jun. 1986. She received the M.S. degree in mechanical engi-
neering from the University of Electronic Science
[10] M. H. Raibert, H. B. Brown Jr, and C. Michael, “Experiments in balance
with a 3D one-legged hopping machine,” Int. J. Robot. Res., vol. 3, no. 2, and Technology of China in 2014. Her research
pp. 75−92, Jun. 1984. interests include biomimetic robotics.

[11] G. B. Xu and L. Z. Zeng, “Robot and it’s simulation,” Comput. Simulat.,

vol. 33, no. 2, pp. 1−16, May 2016.
[12] J. H. Zheng, J. C. Niu, M. S. Jiang, M. Li, and X. W. Rong, “Dynamic
analysis and simulation of spring legs in quadruped robot based on trot
gait,” J. Cent. South Univ. (Sci. Technol.), vol. 46, no. 8, pp. 2877−2883,
Aug. 2015.
[13] M. Zhuang, Z. W. Yu, D. P. Gong, M. L. Xu, and Z. D. Dai, “Gait
planning and simulation of quadruped robot with hydraulic drive based
on ADAMS,” Mach. Des. Manufact., no. 7, pp. 100−102, Jul. 2012.
Li Du is currently a Lecturer with the University
of Electronic Science and Technology of China,
Chengdu, China. She received the Ph.D. degree in
mechanical engineering from the University of Elec-
Dawei Gong is currently a Lecturer with the Univer-
tronic Science and Technology of China, Chengdu,
sity of Electronic Science and Technology of China,
China, in 2010. Her current research interests in-
Chengdu, China. He received the Ph.D. degree in
clude mechatronics and precision engineering.
control theory and control engineering from North-
eastern University, Shenyang, China, in 2012. His
current research interests include neural networks,
complex networks, fuzzy modelling and control,
optimization in process industries, and intelligent
optimization algorithms.

Peng Wang is currently a graduate student and will Yu Duan is currently an Engineer. He received the
receive the M.S. degree in mechanical engineering M.S. degree in mechanical engineering from the
from the University of Electronic Science and Tech- University of Electronic Science and Technology
nology of China. He received the B.S. degree in of China, in 2017. His main research interest is
mechanical engineering from Shandong University electrical engineering.
of Science and Technology, Qingdao, China, in
2014. His research interests include parallel robotics.