2021 the 6th International Conference on Control, Robotics and Cybernetics

A Novel Low-Cost Quadruped Robot with Joint

Fault-Tolerant Control
2021 6th International Conference on Control, Robotics and Cybernetics (CRC) | 978-1-6654-2437-0/21/$31.00 ©2021 IEEE | DOI: 10.1109/CRC52766.2021.9620112

Yixin Zhang Xingjian Wang* Shaoping Wang Jian Zhao

School of Automation Science School of Automation Science School of Automation Science School of Automation Science
and Electrical Engineering and Electrical Engineering and Electrical Engineering and Electrical Engineering
Beihang University Beihang University Beihang University Beihang University
Beijing, China Beijing, China Beijing, China Beijing, China
[email protected] [email protected] [email protected] [email protected]

Abstract—The quadruped robots among legged robots have the pedal and joystick.
been developed for scientific research and human life due to the
better-complicated terrain adaptability. A novel quadruped Meanwhile, the operator can feel the forces of robot foot
robot designed with low cost and low energy consumption is ends by mechanical structure, which gives the operator the
proposed in this work, and each leg is actuated by direct drive basis to judge whether the foot ends touch the ground. Recently,
servos. The three degrees of freedom leg kinematics are plenty of advanced quadruped robots have been designed. In
calculated, and the trotting gait algorithm is adopted to make the 2011, a quadruped robot called StarlETH was designed, which
robot capable of walking. Under complex tasks, the multi-joint was driven by series elastic actuators [3][4]. It is the first
and multi-degree-of-freedom drive of the robot mean potential quadruped actuated by series elastic actuation. The advantage
failures. A fault-tolerant control strategy for the joint stick is of the robot is the excellent terrain adaption ability and low
adopted, maintaining part of workability when the fault occurs. power loss. Boston Dynamic has developed a hydraulic
The walking experiment has confirmed that when the quadruped quadruped robot called BigDog [5]. The robot has 16 active
robot joint faults, it can still work with the performance joints actuated by the hydraulic cylinder. Therefore, it can
degradation. Therefore, the robot has higher reliability and fault support the heavy load and deal with tough terrain. Afterward,
tolerance to complete complex tasks and explore extraterrestrial Boston Dynamic developed a few new productions such as
planets instead of wheeled vehicles. Spot, SpotMini, and LittleDog [6]-[8]. Unfortunately, the
company never published more details about the construction
Keywords—quadruped robot, kinematic, trotting gait, fault-
tolerant control
design and the control algorithm of the quadrupeds.
The development of quadruped robots has been continuing.
I. INTRODUCTION For the robot with complex construction, it will fail without
fault tolerance ability when faults happen. Thus, it is important
Mobile robotic platforms have been developed for to design the fault tolerance framework for the quadruped robot.
improving the life and production of humanity as well as The fault tolerance process can be divided into two parts: fault
scientific research. The platforms are mainly divided into detection and fault tolerance control. There are mainly two
wheeled vehicles and legged robots. The wheeled vehicles can kinds of faults for quadruped robots: damaged leg structure or
move on the solid flat surface with high speed and stiff stability. failure of actuator [9]. When one leg is broken, the quadruped
The vehicles have already played important roles in daily life robot most likely transfers to a three-legged robot. In 2002,
because of its advantages. However, the major limitation of Yun-Jung Lee presented a fault tolerance strategy for
wheeled vehicles is the terrible terrain adaptive capacity. quadruped one leg broken fault [10]. He named the strategy the
On the contrary, the legged animals can reach most of the kick-and-swing gait. However, it is difficult to achieve the
terrain on the surface of the earth [1]. This fact has encouraged locomotion of the three-legged robot with small torque joint.
researchers to focus on the exploitation of bionic-legged robots. There is the fact that the contact between legs and ground is
The common legged robots include biped robots, quadruped approximate to point contact. According to the COG (center of
robots, and hexapod robots. The robots have the potential to gravity) principle, when a three-legged robot needs to lift one
surpass the traditional wheeled vehicles. Compared with other leg to move, it will get into a dynamic instability state
constructions, the quadruped robots have acquired more according to the COG (center of gravity) principle. Therefore,
attention because they have better all-around performance. The the locomotion of the three-legged robot needs at least one
first substantial quadruped robot, Walking Truck, was large torque and high-speed joint actuator. For actuator fault,
produced in the 1960s [2]. The manned robot is driven by the the actuator stick fault has the largest influence on robot
hydraulic actuator, and each leg of the robot is controlled by motion, which will be analyzed in this paper.
The main attribution of the study is to design a novel
This research is supported by the National Natural Science Foundation of twelve degrees of freedom (DOF) directly electrically actuated
China (51675019, 51620105010), Open Foundation of the State Key
Laboratory of Fluid Power and Mechatronic Systems (GZKF-201710) and the
quadruped robot and propose the joint fault tolerance method
Fundamental Research Funds for the Central Universities (YWF-17-BJ-Y- of the robot. The design of the robot follows the principles of
105). portability and low power consumption. Then we adopt a

978-1-6654-2437-0/21/$31.00 ©2021 IEEE 19

Authorized licensed use limited to: Trial User - Universidad Santiago de Cali. Downloaded on May 02,2022 at 17:03:35 UTC from IEEE Xplore. Restrictions apply.
trotting gait to experiment on the robot. Finally, the joint fault The motor system consists of direct drive servo, secondary
tolerance method with fault detection is proposed. MCU, position feedback, and torque feedback. The secondary
MCU system communicates with the main controller by the
This paper is organized as follows. Section II describes the serial peripheral interface (SPI), reaching a high speed. More
design and calculation of the kinematic model of the quadruped system parameters about the quadruped robot are displayed in
robot. Section III proposes the implementation process of the Table I, and the assembly of the hardware and mechanical
gait and conducts the gait experiment. In Section IV, the joint component is shown in Fig. 2.
fault-tolerant method for single-leg failure is proposed. Finally,
conclusions are drawn in Section V.
TABLE I. PARAMETERS OF THE QUADRUPED ROBOT
II. DESIGN AND KINEMATIC MODEL CALCULATION OF THE Parameter Value
QUADRUPED ROBOT Mass of One Leg (kg) 2.1
Mass of the Robot (kg) 10.2
A. The Construction Design and Hardware Length of Base (cm) 60
In order to design the structure of the quadruped robot, the Width of Base (cm) 23
previous research of other scholars is referred to [11], and the Length of the First Leg (cm) 19.1
applicability of different types of actuators is studied. The
Length of the Second Leg (cm) 20.5
novel construction of the quadruped is shown in Fig. 1. The
robot has one base and four legs. Each leg of the robot contains Width of Hip (cm) 5.7
three DOF, include two hip joints and one knee joint. The three Height of Hip (cm) 3.5
DOF legs have simple structure and are capable of reaching for Max Torque of Actuator ( N ˜ m ) 8
demand area. The ultra-light fuselage composed of aluminum Walking Power (W) 30~80
alloy and carbon fiber has high structural rigidity, and the
whole actuator adopts a serial bus digital high-torque servo B. The Kinematic Model of the Robot
(SM80).
The robot mechanical assembly is actually an open-chain
linkage mechanism consisting of the rigid body, the rotate joint
and the rigid connecting rod. The control of the robot system is
the coupling control of the joint actuator. The connection
between the joint angle is analyzed, and the leg endpoint
position is the key kinematics of the robot. Therefore, the
forward kinematic of the quadruped robot and the inverse
kinematic of the legs are calculated. As the four legs of the
robot have similar construction, one leg kinematic model is an
important basis for the whole-body control.

Fig. 1. The construction diagram of the quadruped robot.

Fig. 2. The prototype of the quadruped robot.

The hardware system of the quadruped robot consists of the

microcontroller unit (MCU), inertial measurement unit (IMU),
actuator system, and angular transducer. The STM32F407 is
adopted as the core of the main MCU. The IMU (icm20602) is
used to calculate the attitude of the rigid body. The actuator
system is applied as SM80, the product of the Feetech Robot. Fig. 3. Five coordinate systems of one leg.澳

20

Authorized licensed use limited to: Trial User - Universidad Santiago de Cali. Downloaded on May 02,2022 at 17:03:35 UTC from IEEE Xplore. Restrictions apply.
 The leg endpoint position is considered relative to the base z n ˜ cosT1  m ˜ sin T1  L1 ˜ cosT1 ˜ cosT2
(rigid body). The one-leg model of the quadruped robot can be (8)
approximated as a 3-DOF arm model. The forward kinematic  l2 ˜ cos T1 ˜ cosT2 ˜ cosT3  sin T3 ˜ sin T2
problem is to calculate the position by the joint angle. On the where ( x , y , z ) represents the spatial coordinate of the
contrary, the calculation of angle to reach the end position is an
inverse kinematic problem. To calculate the forward kinematic endpoint of the leg and refers to the base hip coordinate system.
model, first, we need to establish a coordinate system of the In addition, according to geometric principle, the inverse
static position. The coordinate system is shown in Fig. 3. kinematic of one leg can be calculated by
Then calculate the homogeneous transformation matrix m y
between every two adjacent coordinate systems in (1)-(4). T1 = arcsin  arctan (9)
y z
2 2 z
ª1 0 0 0º
«0 cos T  sin T1 0 »» x
Base
T « 1
(1) T2 arcsin
H1
«0 sin T1 cos T1 0» L1  L2 ˜ cos T3
2
 L2 ˜ sin T3
2

« » (10)
¬0 0 0 1¼ L2 ˜ sin T3
 arctan
ª cos T 2 0 sin T 2 0 º L1  L2 ˜ cos T3
« 0 1 0 m »» 2
H1
T « (2) O y  n ˜ sin T1 (11)
H2
«  sin T 2 0 cos T 2 n »
« » P z  n ˜ cosT1
2
(12)
¬ 0 0 0 1 ¼

ª cos T3 0 sin T3 0 º x 2  O 2  P 2  m2  L12  L2 2

« 0 T3 arccos (13)
H2 « 1 0 0 »» 2 ˜ L1 ˜ L2
T (3)
K
«  sin T3 0 cos T3  L1 »
« » III. TROTTING GAIT AND EXPERIMENT
¬ 0 0 0 1 ¼
After kinematic calculation, the entire control framework
ª1 0 0 0 º can be established as shown in Fig. 4. In this section, a new
«0 1 0 0 »» method is proposed to implement a trotting gait and an
K
T « (4)
End
«0 0 1  L2 » experiment was carried out on the prototype.
« »
¬0 0 0 1 ¼ A. Trotting Gait
where Base
T , H1
T , H2
T , K
T represent the homogeneous According to the quadruped animals, the typical gaits of
H1 H2 K End
quadruped robots include crawl gait, trotting gait, bound gait,
transformation matrixes respectively between the base and first and pace gait. The crawl gait is a static gait, which means at
hip joint, between the first hip joint and second hip joint, least three legs are steady-state during walking. The other three
between second hip joint and knee joint, between the knee joint kinds of gait are dynamic gait and trotting gait, which possess
and leg end. T1 , T 2 , T 3 denote the first hip joint angle, the better stability. Compared with the crawl gait, the trotting gait
second hip joint angle and the knee joint angle respectively. m has the advantage of a higher moving speed and lower power
and n is the distance between the first hip joint and second hip consumption [12].
joint. L1 is the length of the first leg and L2 is the length of the Here we only consider the situation that the duty is 0.5 (the
second leg. stance phase of each leg occupies 50% of the gait cycle).
According to the advantages, we adopt the trotting gait as our
In order to obtain the homogeneous transformation matrix
robot gait.
between the base and the leg end, it is necessary to multiply the
homogeneous transformation matrix of each joint.
Base
End T Base
H1 T ˜ HH12T ˜ HK2 T ˜ EndKT (5)
Fig. 4. Control framework of the quadruped robot.
Then, the forward kinematics model of the single-leg is
obtained by Base
EndT and can be denoted by
B. Leg Ends Trajectory
x l1 ˜ sin T2  l2 ˜ cosT2 ˜ sin T3  cos T3 ˜ sin T2 (6) To reach the demand of higher speed and lower
consumption, a cycloid is chosen as the leg ends trajectory.
y n ˜ sin T1  m ˜ cosT1  L1 ˜ sin T1 ˜ cosT2 The cycloid trajectory is shown in Fig. 5. The trajectory
(7) equation can be expressed as
 l2 ˜ sin T1 ˜ cos T2 ˜ sin T3  cosT3 ˜ sin T2

21

Authorized licensed use limited to: Trial User - Universidad Santiago de Cali. Downloaded on May 02,2022 at 17:03:35 UTC from IEEE Xplore. Restrictions apply.
 ­x
° r t  sin t x Initial state: Leg 1, 2, 3, and 4 are all in the stance phase.
® 0 d t d 2S (14)
°̄ y r 1  cos t x Leg 1 and 3 are in the swing phase, and Leg 2 and 4 are
in the stance phase. During the stance, Leg 2 and 4
cause the base moving one step size. Meanwhile, Leg 1
and 3 swing a length of one step size. Additively, the
steady dot of Leg 1 and 3 move two step sizes.
x Similarly, Leg 2 and 4 are in the swing phase, while
Leg 1 and 3 are in the stance phase. During the stance,
Leg 1 and 3 cause the base moving one step size.
Meanwhile, Leg 2 and 4 swing a length of one step size.
Fig. 5. Cycloid trajectory. This continuous cycle of excitation produces walking
motion.
C. Locomotion Method The change process of the stance leg ends is shown in Fig. 7.
In this work, the second hip joints and knee joints are D. Experiment
responsible for the locomotion, and the first hip joints are in
Prototype experiment is implemented with the trotting gait
charge of attitude and direction adjustment. For direction and
loaded to the quadruped robot, as shown in Fig. 8. The
attitude control, the outer loop PID control is adopted. For
experimental trotting gait method performs well, and the robot
trotting gait, we focus on the coordination between legs and
is tested to be able to walk at a maximum speed of 0.26m/s.
base motion.

Fig. 6. Moving method of one leg in each phase: (a) Swing phase; (b) Stance
phase.

Our trotting gait can be divided into four parts in one period.
The unique parameter that needs to be input is the step size of
澳
one leg, displayed as s in Fig. 6, where the solid lines denote
Fig. 8. Trotting gait on robot. Each state corresponds the same label in Fig. 7.
the state before moving, and the dotted lines represent the state
after moving. More details about the gait have been explained At the same time, the limit load state of the robot quickly
as follows: switching from squatting to standing is tested, as shown in Fig.
9. The prototype experiment uses a high-speed camera to
capture the motion pictures of the robot from the front view
and track the motion trajectory of the COG and the knee joint.
The finite difference method is used to calculate the tracking
data to obtain the corresponding speed and acceleration, which
can analyze the performance of the robot's kinematics.

Fig. 7. The stance leg ends change process. Fig. 9. Kinematics test of prototype through high-speed camera.

22

Authorized licensed use limited to: Trial User - Universidad Santiago de Cali. Downloaded on May 02,2022 at 17:03:35 UTC from IEEE Xplore. Restrictions apply.
 The tracking results of COG and knee joints are shown in B. Fault-Tolerant Method
Fig. 10. The quadruped robot can achieve stable state switching, After one joint stick, the leg can be equivalent to a 2-DOF
and the output torque of each joint can meet the needs of leg in Fig. 12. Furthermore, we need only to pay attention to
motion. During the two processes of squatting and standing up, the leg end position on height direction and forward direction.
the robot can switch between states and reach a stable state The procedure of the fault tolerance method is as follows:
within 2s. The vertical velocity of the COG of the robot can
reach up to 0.4m/s, indicating that the response speed of the x After detection of the knee joint stick fault, record the
actuator meets the design. joint position by the sensor.
x The leg endpoint position with forward kinematic is
calculated, and the level distance l between the
second hip joint and the leg end is obtained. Here, a
minimum value lmin is set. If l d lmin , the height of the
COG will be adjusted according to l .
x Reframe the gait. The step size is changed to 2l . The
fault leg swings as shown in Fig. 13. The stance
position of the opposite side leg of the fault leg is
changed to maintain the COG on the connection of the
stance point.
x During the stance phase, the opposite leg should follow
the height of the fault leg to avoid the overturn of the
base.

Fig. 10. Tracking of COG and knee joints.

Fig. 11. The velocity of COG in the x direction and y direction. Fig. 12. The trajectory of fault leg endpoint position.

IV. JOINT FAULT TOLERANCE

Compared with broken leg fault, joint fault can be
responded to more conveniently, while leg damage usually
causes the leg to lose the stance ability. In this section, the
fault-tolerant method for knee joint stick fault is designed in
the control system. The knee joints are actuated by direct drive
servos, which bear more loads because of the construction. The
most common and damaged fault of the drive module is motor
jam. Thus, the joint stick is considered an experimental fault
here.

A. Fault Detection
The angle sensor can detect the fault easily. By evaluating
the error between the input angle and the detection angle by the
sensor, we can judge whether the fault happened. To exclude
the sensor fault, we can compare two oblique leg detection
results. Because under the trotting gait, the oblique leg knee
joints should act synchronously. Fig. 13. The stance leg ends changing procedure on fault situation (Red point
represents fault leg).

23

Authorized licensed use limited to: Trial User - Universidad Santiago de Cali. Downloaded on May 02,2022 at 17:03:35 UTC from IEEE Xplore. Restrictions apply.
 Finally, a performance degradation walking test is carried under single-leg failure. So, the robot can walk as changing
out on a smooth marble floor. As shown in in Fig. 14, the robot trotting gait with the fault-tolerant method rather than falling.
advances by 0.6m in 7.5s, resulting in a velocity of
approximately 0.08m/s. Therefore, the fault-tolerant control In addition, it is necessary to establish an accurate dynamic
algorithm has been well verified. model of a quadruped robot to achieve fast walking. Although
actuator sensor failures and motor drive circuit failures can also
cause the abnormal performance of the robot system, these
situations are relatively rare and have not been considered here.
In the future, more detailed analyses will be carried out on the
failure modes and failure factors of the quadruped robot.

REFERENCES
[1] P. Gregorio, M. Ahmadi, and M. Buehler, “Design, control, and
energetics of an electrically actuated legged robot,” IEEE Transactions
on Systems, Man, and Cybernetics, Part B (Cybernetics), 1997, vol. 27,
no. 4, pp: 626-634.
[2] R. A. Liston, and R. S. Mosher, "A versatile walking truck,"
Transportation Engineering Conference. 1968.
[3] M. Hutter, C. Gehring, M. Bloesch, M. A. Horpflinger, C. D. Remy and
R. Siegwart, “StarlETH: A compliant quadrupedal robot for fast,
efficient, and versatile locomotion,” Adaptive Mobile Robotics. 2012,
pp: 483-490.
[4] M. Hutter, C. Gehring, M. Bloesch, and M. A. Horpflinger, “Walking
and running with StarlETH,” The 6th International Symposium on
Adaptive Motion of Animals and Machines (AMAM), Darmstadt,
Germany, March 11-14, 2013
[5] M. Raibert, K. Blankespoor, G. Nelson, and R. Playter, “Bigdog, the
rough-terrain quadruped robot,” IFAC Proceedings Volumes, 2008, vol.
41, no. 2, pp: 10822-10825.
[6] M. Kalakrishnan, J. Buchli, P. Pastor, and M. Mistry, “Learning,
Fig. 14. Performance degradation walking test on smooth marble floor. planning, and control for quadruped locomotion over challenging
terrain,” The International Journal of Robotics Research, 2011, vol. 30,
no. 2, pp: 236-258.
V. CONCLUSION [7] M. Kalakrishnan, J. Buchli, P. Pastor, and M. Mistry, “Fast, robust
quadruped locomotion over challenging terrain,” Robotics and
A novel low-cost and low power consumption electrical Automation (ICRA), 2010 IEEE International Conference on. IEEE,
actuated quadruped robot is developed in this work. The robot Anchorage, Alaska, USA, 2010, pp: 2665-2670.
can be supported to carry out conventional jobs with less [8] J. R. Rebula, P. D. Neuhaus, B. V. Bonnlander, M. J. Johnson, and J. E.
energy consumption and long battery life. The construction of Pratt, “A controller for the LittleDog quadruped walking onrough
the quadruped has been shown with some important parameters. terrain,”The IEEE Int. Conf. on Robotics and Automation, 2007, pp.
Then, the kinematic model of the robot leg is calculated. The 1467–1473.
gait of normal quadruped walking and single foot failure state [9] J. M. Yang, “Fault-tolerant gaits of quadruped robots for locked joint
walking is adopted and implemented in the walking test with failures,” IEEE Transactions on Systems, Man, and Cybernetics, Part C
(Applications and Reviews), 2002, vol. 32, no. 4, pp: 507-516.
the prototype.
[10] Y. J. Lee, and S. Hirose, “Three-legged walking for fault-tolerant
We dispose of the problem about the coupling of swing locomotion of demining quadruped robots,” Advanced Robotics, 2002,
legs and the base position moving. The knee joint stick fault- vol. 16, no. 5, pp: 415-426.
tolerant method for single-leg failure is proposed. The [11] C. Semini, N. G. Tsagarakis, E. Guglielmino, M. Focchi, F. Cannella,
and D. G. Caldwell, “Design of HyQ – a hydraulically and electrically
performance of the robot walking with this method is worse actuated quadruped robot,” Journal of Systems and Control Engineering,
than that of the fault-free state, but it can still complete the 2011, vol. 225, no. 6, pp: 831-849.
walking under performance degradation. After experimental [12] G. Hawker, and M. Buehler, “Quadruped trotting with passive knees:
measurement, the walking speed can reach 0.26m/s under design, control, and experiments,” IEEE International Conference on.
normal conditions, and the walking speed can reach 0.08m/s IEEE, San Francisco, CA, USA, 2000, pp: 3046-3051.

24

Authorized licensed use limited to: Trial User - Universidad Santiago de Cali. Downloaded on May 02,2022 at 17:03:35 UTC from IEEE Xplore. Restrictions apply.