2017 International Conference on Computer Technology, Electronics and Communication (ICCTEC)

Adaptive Sliding Mode Control for Omnidirectional Mobile Robot Based on A New
Friction Modeling

Xiangnan Fu a, Shuoyu Wang, Juyou Yang, Yina Wang, Ziyuan Liu

School of Electrical Engineering,
Shenyang University of Technology
Shenyang, China
a
[email protected]

Abstract—Due to the special structure of omni-directional nonlinear function was used to compensate the nonlinear
wheels, friction forces among hubs, wheels and ground have a effects of the parameters in the friction force model. In [7]
great influence on the motion accuracy of four-wheeled the state of four wheels longitudinal, lateral skidding or not
omnidirectional mobile robot. In order to reduce the effect of skidding was introduced into the modeling, and the internal
friction on motion control system, this paper proposes an force state and motion state of the system were analyzed by
adaptive sliding mode controller (ASMC) based on the Newton mechanics and Lagrange equation modeling
dynamic model considering the friction on omni-directional respectively.
wheels. The research on the omni-directional wheels friction However, there were few studies on the friction force
force is described in detail, which lays a foundation for the
modeling of omnidirectional wheeled mobile robots. As in
accurate control of omnidirectional mobile robot. Firstly, the
paper analyzes the Coulomb's friction on the driving wheel
[8], a predictive control algorithm with friction force
and the driven wheel respectively which is introduced to the compensation was proposed to solve the trajectory tracking
friction model of the omni-directional wheel. Secondly, a new problem for a three-wheeled omnidirectional mobile robot.
adaptive sliding mode control method is proposed to address In [9], the trajectory tracking problem was solved for a four-
the friction problem. Finally, the validity and performance of wheeled omnidirectional robot model using nonlinear MPC
the control method are verified by simulations. considering a static friction force model. Most of these
researches either directly used the existing friction modeling
Keywords-omni-directional wheels; friction force; dynamic or take friction force as a constant that can be counteracted
model; adaptive sliding mode controller (ASMC). during modeling. There is no detailed analysis of the friction
force on the omni-directional wheels, which affect the
I. INTRODUCTION accurate of the dynamic model for the omnidirectional
With the rapid development of science and technology, mobile robot.
various robots with different functions are developed and The dynamic model of mobile robot is very important to
introduced to the daily life of human beings. Part of these design the controller, especially that when the robot needs to
robots can replace the heavy work for human, therefore, the travel at a higher speed and carry out heavier work. For
motion accuracy of the robots are important. However, the example, the importance of dynamic models in the control
motion performance of robots is seriously affected by the strategy of mobile robots is discussed in [10] [11] [12].
external unknown disturbance and the robot's unknown Omnidirectional mobile robots have the ability to move
friction. Thus, it is important and meaningful to deal with independently in both translation and rotation [13], while
the influence of the external unknown disturbance and nonlinearity, such as the dynamic constraints of the motor or
friction for the robots. To design a proper controller for the the friction force associated with the speed of the robot. It is
robots aimed at the friction, a friction modeling is necessary. greatly affect the behavior of the robot which is accelerated
Many models of friction have been proposed in previous and decelerated.
researches [1] [2]. However, some of the parameters in these Therefore, the paper gives a detailed analysis of the
models [3] were difficult to determine, increasing the friction force of the omni-directional wheels, especially the
difficulty and complexity of dynamic modeling. There were driving wheel and the driven wheel, then established the
many researches on the compensation control for friction dynamic equation of the omnidirectional mobile robot, and
force of robots. Reference [4] developed a new proposed an adaptive sliding mode controller to reduce
compensation technique for a friction force model, the friction force and external disturbances, thereby greatly
proposed control utilized a PD control structure and is improving the mobility of mobile robots. Finally, the
adaptive to estimate the friction force. Specifically, a radial simulation results show the effectiveness and accuracy of the
basis function (RBF) was used to compensate the effects of controller.
the unknown nonlinearly Stribeck in the friction model. In
[5], a robust controller was presented to compensate the II. OMNI-DIRECTIONAL WHEELS
dynamic characteristics of friction force in servo-driven Omni-directional wheels or poly wheels, similar to
system by using the industrial manipulator. And in [6], a Mecanum wheels, are wheels with small discs around the

978-1-5386-5784-3/17/31.00 ©2017 IEEE 1042

DOI 10.1109/ICCTEC.2017.00229
circumference which are perpendicular to the turning
direction. A photo of an omni-directional wheel discussed in
this paper is shown in Fig.1. It comprises a hub and a driven
wheel that the outer circumference of the hub is uniformly
provided with 5 hub teeth. The driven wheel is arranged
between each 2 hub teeth and the radial direction of the
driven wheel is perpendicular to the tangent direction of the Fig 3. Projection of the wheel motion curve
outer circumference of the hub. The wheel not only has the
ability to roll forward around its axis but can move around The curves generated by the Omni-directional wheels
with 2 degrees of freedom under the action of external force. structure and the roll contour [14] can be known that the
The movement is more stable and the load capacity is friction direction of the driven gear is opposite to the
improved correspondingly of this kind of omni-directional direction of motion.
wheels since that it can be alternately contacted with the In order to prove the conclusion of above, the detailed
ground force analysis of the driven wheel is carried out as follows˖

Fig 1. Omnidirectional Wheel

A. Force analysis
To facilitate the analysis and modeling, the friction
between the wheel and axle is neglected in this paper.
Furthermore, the deformation and wear of wheel are Fig 4. Force analysis of the driven wheel
neglected. Fig.2 is the force analysis of a single
omnidirectional wheel. A Force analysis of the driven wheel is shown in Fig. 4.
In Fig. 4, v is the motion velocity of the robot,  is the
angle between the motion direction of the robot and the x-
axis, and  is the angle between the first omnidirectional
wheel center and the robot geometry center.
According to the force analysis of the driven gear shown
in Fig.4, we can draw:

 =  =  ( )
 (1)
 =    =  ( )

Then combine the mathematical theorems that if the

alternate interior angles are equal the lines are parallel. It
can be proved that the direction of friction from the driven
Fig 2. Force analysis of omni-directional wheel wheel is opposite to the overall motion of the robot.
Therefore, it can be got a single omnidirectional wheel
In Fig. 2, F is the driving force of the motor on Omni- in the x-axis and the y-axis of the combined external force
directional wheels, F is the friction force for four Omni- component:
directional wheels driving wheel, f is the friction force for
four Omni-directional wheels driven wheels, and l is the  =  sin   sin   cos 
 (2)
distance from the center of the Omni-directional wheels to  =  cos  +  cos   sin 
the center of the robot geometry.
Then we take the driven wheel separately and analyze it: III. DYNAMIC MODELING
the wheel surface is curved surface and the 2 row of driving The force analysis of the whole robot can be deduced
wheels with 10 driven wheels are rotated round for a week from the detailed analysis of the single omnidirectional
[14]. Therefore, the curve of the entire wheel during motion wheel.
is projected on the ground in a straight line as shown in Fig Dynamic coordinate system of four-wheeled omni-
3.

1043
directional mobile robot is shown in Fig.5. In Fig. 5, the IV. TRACKING CONTROL OF OMNI-DIRECTIONAL MOBILE
XOY is the stationary coordinate system of the robot, the xoy ROBOT
is the motion coordinate system of the robot, F (i = 1,2,3,4)
is the driving force of the motor on four Omni-directional A. Controller Design
wheels, F (i = 1,2,3,4) is the friction force for four Omni- Considering the uncertainty of the dynamic model of the
directional wheels' driving wheel, and f (i = 1,2,3,4) is the four-wheeled omnidirectional mobile robot and the effect of
friction force for four Omni-directional wheels’ driven additional disturbances, the formula (4) is added to the
wheels. Suppose that the posture of the robot in the parameter E(t) containing the additional unknown
stationary coordinate system is q = (X, Y, ) . interference.

-. = /0 + C(D)(5)

Assuming that E(t) is bounded, it satisfies:

G C(D) GH I(6)

In order to facilitate the calculation, the proper

deformation treatment of equation (5) will be obtained:

-. = K + K +  + C(D)(7)

Fig 5. Dynamic coordinate system of mobile robot where:

If the robot's uncertainty is neglected, the dynamic sin  cos  sin  cos 
model of four-wheeled omnidirectional mobile robot can be P = 9 cos  sin  cos  sin  ;,
obtained according to Newton's second law: l l l l

 = ! + ! "   + # + # "  F = [F! F# F$ F% ] , F = [F! F# F$ F% ] ,

 $ + $ "   + % + % " 
 f = [ L Mgcos  LMgsin  0]
 ! + # + $ + % " 

& = ! + ! "  + # + # "   Suppose that the trajectory of the four-wheeled
(3)
 +$ + $ "  + % + % "   omnidirectional mobile robot in the stationary coordinate
 ! + # + $ + % "   system is a smooth curve q(t) = [X Y ] and has a
 second derivative; the expected trajectory of a four-wheeled
 ' = *! + ! " *# + # "
 omni-directional mobile robot is qN (t) = [XN YN N ] .
 *$ + $ " + *% + % " Thus, the tracking error can be expressed as e(t) = qN (t)
q(t).
where M is the total weight of the mobile robot, J is the
According to the basic principle of sliding mode control,
rotational inertia of the robot.
the selection of sliding surface is usually based on linear
It can be written as follows:
sliding surface. The method is simple and the parameter
design is easy. However, the linear sliding surface cannot
-. = /0 (4) guarantee the system's state tracking error converges to zero
in a limited time. Therefore, we introduce nonlinear term to
Where solve this problem. Then the system sliding surface can be
8
designed:
- = 567{, , '}, . = [ & ] ,
Z
(D) = QR + ST Q + SU V\ Q(W)5W(8)
sin  cos  sin  cos  cos 
/ = 9 cos  sin  cos  sin  sin  ;
* * * * 0 where k ^ = diag{a! , a# , a$ }, k  =
diag{b! , b# , b$ }, a , b > 0, i = 1,2,3.
! + ! Formula (8) derivation operation descendants into
? B formula (7) can be obtained:
>  # + # A
0 = > $ + $ A
> A - ! K` = - ! K +  " .h + ST QR + SU Q + C(D)(9)
>  % + % A
<! + # + $ + % @

1044
 ! !
where u indicates the output of the sliding mode  ~„ ~(D)
R = ~„ €| | = ~„ | | = (~ ~p )| |(16)
ƒ ƒ
controller.
According to the basic theory of sliding mode control,
Then
the sliding mode control law of trajectory tracking control
system of four-wheeled omnidirectional mobile robot can be
designed: ‚R = C ~ | | + (~ ~p )| |

` = Kj -(.h - ! K - !  ST QR SU Q +  j 7( ))(10) = C ~p | | H 0(17)

where Pj = (P  P)! P is the system model of the Therefore, we can get limZ‹Œ (D) = 0. 7KHUHIRUHˈ
Ž
conjugate transpose array, m j sgn(s) is a switch function when = 0 , there are Q =  V\ Q(W)5W , and when
and m > 0
Q(0)  0, limZ‹Œ Q(D) = 0 It is shown that the adaptive
Assuming that the parameters in the above equation are sliding mode trajectory tracking control system is stable.
known, it is difficult to obtain the upper bound of the
uncertain
E(t) due to the uncertainty of the robot dynamics V. SIMULATIONS
model and the influence of external disturbances. Therefore, In order to verify the feasibility and effectiveness of
adaptive control method is needed to realize the adaptive trajectory tracking method of four-wheeled omnidirectional
control of the upper bound of E(t). mobile robot based on adaptive sliding mode control, we
Suppose rp is the upper bound of E(t), |E(t)| < rp ,and rz is use MATLAB software to simulate the tracking effect of the
the estimate of the upper bound of E(t). The sliding film curve trajectory by the controller.
control rate can be designed as: It is obvious that the trajectory tracking of the robot is
closely related to the angle of the motion of the robot which
` = Kj -(.h - ! K - !  ST QR SU Q + ~ j 7( ))(11) can be validated by a simple PD controller, Select the
desired trajectory equation
Similarly, the adaptive rate can be designed to
qN! = (3cos t 3 sin t ‘/6) .
R = €| (D)|(12)
~ (D)
qN# = (3cos t 3 sin t ‘/4) .
where  is an adaptive gain,  > 0.
The result is shown in Fig.6:
B. Stability Analysis Fig.6 are clearly shows the effect of moving direction on
For the trajectory tracking control system of the four- the motion of the robot in the case of friction force and the
wheeled omnidirectional mobile robot described in formula PD controller has no exact effect on the robot system
(7), the control system is asymptotically stable if the sliding established in this paper.
mode control rate (11) and the parameter adaptive rate (12) 4
are selected. The certification process is as follows: expect

Defining a Lyapunov function 
3

! # !
‚= + ~„ (D)# (13)
# #ƒ
2

where
1
Y-Position/m

r(t) = rz(t) rp .
0
Then
!
‚R = R (14)
R + ~„ ~(D) -1
ƒ

 = Q + ST QR + SU Q = . .h + ST QR + SU Q -2

= - ! K + K +  " .h + ST QR + SU Q + C

= - ! K †Kj - ‡.h - ! K - !  ST QR SU Q ~ -3
-4 -3 -2 -1 0 1 2 3 4
X-Position/m

j 7( )ˆ‰ + - ! K + - !  .h (a) Trajectory tracking at different angles

+ ST QR + SU Q + C
=C ~ j 7( )(15)

1045
 0.5 friction force respectively firstly and the simulation results
0 areas shown in Fig.7:
X-Axis position tracking of Difference

4
-0.5
expect
-1 PD( no f )
3 PD( f )
-1.5

-2
2
-2.5

-3

1
-3.5

Y-Position/m
-4
0 1 2 3 4 5 6 7
0
time/s
0.25
Y-Axis position tracking of Difference

0.2 
-1

0.15

0.1 -2

0.05
-3
0

-0.05
-4
-4 -3 -2 -1 0 1 2 3 4
-0.1 X-Position/m
0 1 2 3 4 5 6 7
time/s (a) Trajectory tracking with friction and no friction
0.5
(b) X-axis, y-axis error at different motion angles
Fig 6. Influence of robot motion angle on trajectory tracking 0
X-Axis position tracking of Difference

-0.5
The results of the adaptive sliding mode control method
presented in this paper are as follows: -1

Parameter setting of trajectory tracking controller: -1.5

k ^ = diag{60 60 60}, k  = diag{70 70 70}, -2

-2.5 Xe PD( no f )
Xe PD( f )
and the adaptive parameter is set to  = 10.
-3
The parameters of the four-wheeled omnidirectional 0 1 2 3
time/s
4 5 6 7

mobile robot are shown in table 1. Assuming the robot's (b) Error of x-axis under friction force
starting posture is q(t \ ) = (3 0 0) , the unknown 0.25
Ye PD( no f )
perturbation in the tracking process is 0.2 Ye PD( f )
Y-Axis position tracking of Difference

0.15

E(t) = [2 sin(–‘t) 1.5 cos(–‘t) 3 sin(–‘t)] . 0.1

0.05

TABLE I. PARAMETERS OF MOBILE ROBOT 0

-0.05
Description Symbol Value Unity
-0.1

Overall mass of the robot M 58 kg -0.15

-0.2
0 1 2 3 4 5 6 7
Moment of Inertia J 27.7 kg·m² time/s

(c) Error of y-axis under friction force

Robot radius l 0.4 m
Fig 7. Influence of friction force on trajectory tracking
Rolling friction coefficient  0~0.5
From the Fig.7, it is obvious that the friction force has a
Gravitational acceleration g 9.8 m/s²
great influence on the precise motion of the omnidirectional
mobile robot, and the traditional PD controller cannot
In this paper, the accuracy of the model and the
effectively reduce this effect.
effectiveness of the control method are validated by
In order to verify the validity of the controller designed
simulating a circular trajectory. Select the desired trajectory
by the paper, the adaptive Sliding mode controller designed
as qN = (3cos t 3 sin t t) . by the PD controller and the paper is used to compare and
Now, in order to verify the impact of friction on the verify the robot system with friction force as shown in
omnidirectional mobile robot system, the traditional PD Fig.8:
controller is used to simulate the system with and without

1046
 4
expect
VI. CONCLUSION
PD ( f )
SMC ( f )
The new adaptive Sliding mode control method
3
proposed in this paper can weaken the effect of the full
wheel friction on the precise motion of the mobile robot.
2 Additionally, the Lyapunov stability theory further proves
the stability of the system and the convergence of the
tracking error. However, the paper does not consider the
1
deformation caused by the contact between omni-directional
Y-Position/m

wheel and ground. The resistance of these deformations can

0 also affect the precise movement of mobile robots, which
we will continue to study in future work.
-1 REFERENCES
[1] C. Canudas de Wit, H. Olsson, K. J. Astrom, and P. Lischinsky, “A
-2
new model for control of systems with friction,” IEEE Trans. Autom.
Control, vol. 40, no. 3, pp. 419–425, Mar. 1995.
[2] B. Bona and M. Indri, “Friction compensation in robotics: An
-3 overview,” in Proc. 44th IEEE Conf. Decis. Control Eur. Control
-4 -3 -2 -1 0 1 2 3 4
X-Position/m
Conf., Dec. 2005, pp. 4360–4367.
(a) Trajectory tracking of different control methods [3] L. E. Ray, “Estimation of terrain forces and parameters for rigid-
0.5 wheeled vehicles,” IEEE Trans. Robot., vol. 25, no. 3, pp. 717–726,
Jun. 2009.
0
X-Axis position tracking of Difference

[4] S. Huang, K. K. Tan, T. H. Lee, “Adaptive friction compensation

-0.5 using neural network approximations.” IEEE Transactions on
Systems Man & Cybernetics Part C, 30(30), 551-557. 2000.
-1
[5] L. Mostefai, M. Denai, and Y. Hori, “Robust tracking controller
-1.5 design with uncertain friction compensation based on a local
modeling approach,” IEEE/ASME Trans. Mechatronics, vol. 15, no. 5,
-2 pp. 746–756, Oct. 2010
-2.5 Xe PD ( f )
[6] T. H. Lee, K. K. Tan, and S. Huang, “Adaptive friction compensation
Xe SMC ( f ) with a dynamical friction model,” IEEE/ASME Trans. Mechatronics,
-3 vol. 16, no. 1, pp. 133–140, Feb. 2011.
0 1 2 3 4 5 6 7
time/s
[7] Chao Yingli. Research on modeling and motion control of four-
(b) Error of x-axis under different control methods wheel-drive mobile robot [D]. Zhejiang University, 2017.
0.25
Ye PD ( f ) [8] Julio, C. L. B. S., Conceicao, A. G. S., Dorea, C. E. T., Martinez, L.&
0.2 Ye SMC ( f )
Pieri, E. R. D. “Design and implementation of model predictive
Y-Axis position tracking of Difference

0.15
control with friction compensation on an omnidirectional mobile
0.1 robot.” IEEE/ASME Transactions on Mechatronics, 19(2), 467-476.
0.05 2014.
0 [9] A. G. S. Conceicao, H. Oliveira, A. Sousa e Silva, D. Oliveira, and A.
-0.05
Moreira, “A nonlinear model predictive control of an omnidirectional
mobile robot,” in Proc. IEEE Int. Symp. Ind. Electron, Jun. 2007, pp.
-0.1
2161–2166.
-0.15
[10] A. Albagul and Wahyudi, “Dynamic modeling and adaptive traction
-0.2 control for mobile robots,” Int. J. Adv. Robot. Syst., vol. 1, pp. 149–
0 1 2 3 4 5 6 7
time/s 154,2004.
(c) Error of y-axis under different control methods
[11] K.J. Waldron and M.E. Abdallah, “An optimal traction control
Fig 8. Effect of effective control method on trajectory tracking scheme for off-road operation of robotic vehicles,” IEEE/ASME
Trans. Mechatronics, vol. 12, no. 2, pp. 126–133, Apr. 2007.
As can be seen from Fig. 8, based on the dynamic model [12] K.L. Moore and N.S. Flann, “A six-wheeled omnidirectional
and the adaptive sliding mode trajectory tracking controller, autonomous mobile robot,” IEEE Control Syst. Mag., vol. 20, no. 6,
pp. 53–66, Dec.2000.
the four-wheeled omnidirectional mobile robot has better
tracking effect on the circular trajectory, which can ensure [13] T. Jun, K. Watanabe, and Y. Shiraishi, “Design and traveling
experiment of an omnidirectional holonomic mobile robot,” in Proc.
the complete tracking of the desired trajectory in a short IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), 1996, vol. 1, pp. 66–
time under the condition of large initial error. At the same 73.
time, the chattering phenomenon of the system has been [14] Gfrerrer A. Geometry and kinematics of the Mecanum wheel [J].
greatly weakened and the desired effect has been achieved. Computer Aided Geometric Design, 2008, 25(9):784-791.

1047