This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS 1

Motion Tracking Control Design for a Class of

Nonholonomic Mobile Robot Systems
Jun Fu , Senior Member, IEEE, Fangyin Tian, Tianyou Chai, Fellow, IEEE,
Yuanwei Jing, Zhijun Li , Senior Member, IEEE, and Chun-Yi Su , Senior Member, IEEE

Abstract—Motion tracking control design of nonholonomic this paper concerns the design of tracking controller for the
mobile robot systems considering actuator dynamics is addressed nonholonomic mobile robot at actuator level.
in this paper. A trajectory tracking controller is designed at Anupoju et al. [21] proposed a robust adaptive tracking
actuator level, which guarantees that the nonholonomic mobile
robot tracks a given trajectory. A numerical example is shown to method at actuator level, but to the authors’ best knowledge,
demonstrate and validate the proposed approach in this paper. the drawback of which is that the complicated controller struc-
ture may cause heavy computing burden when applied to
Index Terms—Actuator dynamics, backstepping, cascaded
systems, motion control, nonholonomic mobile robot. high order nonholonomic systems. Because that this method
involves calculating the sum of high order derivative of the
given signal ud when designing the virtual controllers for kine-
I. I NTRODUCTION matic subsystem, which can be a heavy burden when it comes
ONHOLONOMIC constraints abound in many important to high order nonholonomic systems. Motivated by this and
N applications, for example, wheeled vehicles, mobile
robots, multifingered robotic hands, and space robots [1]–[9].
based on our work [19], we proposed a constructive method
for motion control at actuator level. This method adopts the
The control of nonholonomic robot systems has received cascaded design method at kinematic level to gain two cas-
much attention of the control field, because nonholonomic caded subsystems and then design virtual controllers for the
systems cannot be stabilized by using any time-invariant two subsystems, respectively. By doing so, we can avoid cal-
smooth state-feedback control laws [10], [11]. Therefore, culating high order derivative and make the virtual controllers
many efforts are given on design of stabilizers of non- more simple. Then we extend the simple virtual controllers to
holonomic systems [10], [12]. The tracking of nonholonomic dynamic and actuator level with backstepping. The main con-
robot system, as a more practical problem in control engi- tributions includes two aspects. First, we use cascaded design
neering, however, has gained relatively less emphasis [1]. method to simplify the virtual controllers for the kinematic
The tracking problems are usually classified into kinematic subsystem. Second, having the simplified virtual controllers
(such as [13]–[15]) and dynamic problems(such as [16]–[18]). we then obtained the controller at actuator level using back-
The controller designed for the kinematic system, which stepping technique. The method proposed in this paper consists
have the velocity as input, needs an almost perfect veloc- of two stages. Stage one is to design the controllers at kine-
ity tracking assumption (the velocity needed can be gained matic and dynamic level, respectively, using the cascaded
immediately) [19]. The controller at dynamic level, comput- design method and backstepping technique [19], [22]. Stage
ing required torque for the robot, has torque of the driven two is to design the tracking controller at actuator level
wheels as the control input. However, the actuator dynamics, through combining the above subcontrollers with backstepping
as an essential part of the whole dynamics of electrically driven technique.
robots [20], plays an important role in the control of mobile The structure of this paper is organized as follows. Section II
robots, especially when high-velocity movement and highly gives the dynamic model of nonholonomic mobile robot
varying loads exist in the robot system [21]. Along this line, systems containing actuators. The controller design procedure
is presented in Section III, which also includes several nec-
Manuscript received June 7, 2017; revised October 25, 2017; accepted essary remarks. Section IV gives the illustrative simulation
February 4, 2018. This paper was presented in part at the 2017 IEEE
Conference on Advanced Robotics and Mechatronics (ICARM), August 29– results validating the proposed method. Section V draws some
31, 2017, Hefei, China. This paper was recommended by Associate Editor conclusions.
R. Roberts. (Corresponding author: Jun Fu.)
J. Fu, F. Tian, T. Chai, and Y. Jing are with the State Key Laboratory
of Synthetical Automation for Process Industries, Northeastern University, II. M ODEL OF N ONHOLONOMIC M OBILE ROBOT S YSTEMS
Shenyang 110819, China (e-mail: [email protected]). Consider the nonholonomic system which includes actuator
Z. Li is with the College of Automation, South China University of
Technology, Guangzhou 510061, China. dynamics as one part of it [19], [21]
C.-Y. Su is with the Department of Mechanical, Industrial and Aerospace J(q)q̇ = 0 (1)
Engineering, Concordia University, Montreal QC H3G 1M8, Canada (e-mail:
[email protected]). D(q)q̈ + C(q, q̇)q̇ + G(q) = J (q)λ + B(q)KN I
T
(2)
Color versions of one or more of the figures in this paper are available dI
online at http://ieeexplore.ieee.org. L + RI + Ka ω = V (3)
Digital Object Identifier 10.1109/TSMC.2018.2804948 dt
2168-2216 c 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

where q denotes a n-vector of generalized coordinates; I form. In Section III-B, the controller is constructed at the kine-
denotes a r-vector of mature current; λ ∈ Rm is Lagrangian matic and dynamic levels under the framework of cascaded
multiplier representing the contact force; D(·), C(·, ·), and design and by using backstepping techniques. In Section III-C,
G(·)q̇ are the n × n positive definite inertia matrix, which the controller is then constructively derived at the actuator
is symmetric and bounded, centripetal and Coriolis torques, level.
and the gravitational torques, respectively; B(·) and J(·) are
the n × r input transformation matrix, which is assumed
known and m × n constraint matrix, respectively; and KN A. Model Transformation of the Nonholonomic System
is a positive definite diagonal matrix, which gives the elec- We obtain the two-input nonholonomic system in
tromechanical relationship between armature current and chained-form, which is transformed from (8) by coordi-
torque nate change x = (q), and a state feedback control
v = 1 (q)u [19], [23]
L = diag[L1 , L2 , L3 , . . . , Lr ]
⎧
R = diag[R1 , R2 , R3 , . . . , Rr ] ⎪
⎪ ẋ1 = u1
⎪
⎪
Ka = diag[Ka1 , Ka2 , Ka3 , . . . , Kar ] ⎪
⎨ ẋ2 = u2
ẋ3 = x2 u1 (12)
ω = [ω1 , ω2 , . . . , ωr ]T . (4) ⎪ ..
⎪
⎪
⎪
⎪ .
⎩
L and R represent inductances and resistances, respectively. ẋn = xn−1 u1
Ka is EMF constants and ω represents angular velocities of
driving motors and V is the input voltage of the motor. We where u = (u1 , u2 )T and x are the system input and system
choose v as a vector of generalized velocities q̇, and define state, respectively. Equations (9) and (11) of the system model
R(q) so that q̇ satisfies the constraint (1) are accordingly converted into
q̇ = R(q)v. (5)
D2 (x)R2 (x)u̇ + C2 (x, ẋ)u + G2 (x) = B2 (x)KN I + J2T (x)λ (13)
Given (1) and (5), we have dI
L + RI + Ka Q(u, μ, 1 ) = V (14)
dt
RT (q)J T (q) = 0. (6)
where
Differentiating (5), one has
D2 (x) = D(q)|q= −1 (x)
q̈ = Rv̇ + Ṙv. (7)
R2 (x) = R(q)1 (q)|q= −1 (x)
 
Therefore, system (2) which satisfies (1) could be transformed ˙ 1 (q)+C(q, q̇)1 (q) |q= −1 (x)
C2 (x, ẋ) = D(q)R(q)
into G2 (x) = G(q)|q= −1 (x)
q̇ = R(q)v (8) J2 (x) = J(q)|q= −1 (x)
D(q)R(q)v̇ + C1 (q, q̇)v + G(q) = B(q)KN I + J (q)λ (9)
T B2 (x) = B(q)|q= −1 (x)
Q = μ1 (q)u|q= −1 (y) .
where C1 (·, ·) = D(·)Ṙ(·)+C(·, ·)R(·). The relation between ω
and V in the actuator dynamics (3), is solely dependent on the The motion tracking for (12) is to construct a tracking con-
type of nonholonomic robot systems, and is given as follows: troller in order to guarantee the trajectory of (12) follows a
reference path xd = (x1d , . . . , xnd )T , which is generated by a
ω = μV (10)
same system driven with ud = (u1d , u2d )T , and we need the
where μ is a invariable vector parameter, which depends on following assumptions [19].
the given mechanical system. Combining (3) with (10) gives Assumption 1: Suppose there exist nonzero constants T and
D such that |D| > T > 0, and u1d = d(t) where d(t) is
dI bounded continuous and satisfies |d(t) − D| < T.
L + RI + Ka μV = V. (11)
dt Assumption 2: The reference trajectory xd satisfies that its
up to (N − 1)th derivative are all bounded.
III. M AIN R ESULTS We obtain tracking errors yi = xi − xid , i = 1, . . . , n, and
have the dynamics of the tracking error
We outline the controller design strategy in this section. We
keep our attention on a class of two-input chained form non- ⎧
⎪
⎪ ẏ1 = u1 = u1d
holonomic mobile robot just for simplicity, but this method can ⎪
⎪
⎪
⎨ ẏ2 = u2 − u2d
be easily extended to more general cases. The main results of
ẏ3 = u1d y2 + x2 (u1 − u1d ) (15)
this paper are as follows. In Section III-A, the chained-form ⎪ ..
⎪
⎪
system is derived with x = (q) and v = 1 (q)u and in turn, ⎪
⎪ .
⎩
the dynamics of tracking error are derived in the cascaded ẏn = u1d yn−1 + xn−1 (u1 − u1d )
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

FU et al.: MOTION TRACKING CONTROL DESIGN FOR CLASS OF NONHOLONOMIC MOBILE ROBOT SYSTEMS 3

system (15) can be converted into to guarantee quadratical stabilization [19] of the closed-loop
⎡ ⎤ ⎡ ⎤⎡ ⎤ subsystem
ẏ2 0 0 ··· 0 0 y2
⎢ ẏ3 ⎥ ⎢ d(t) 0 · · · 0 0 ⎥⎢ y3 ⎥ Ẏ = (A + BK + E(t)F)Y. (21)
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ .. ⎥ ⎢ .. .. . . .. .. ⎥⎢ .. ⎥
⎢. ⎥ = ⎢. . . . . ⎥⎢ .⎥ ⎢ ⎥ Assume that: 1) (t) ∈  = {(t)T (t)
≤ I ∀t} and 2) A +
⎢ ⎥ ⎢ ⎥
⎣ ẏn−1 ⎦ ⎣ 0 0 · · · 0 0 ⎦⎣ yn−1 ⎦ BK is stable and F(SI − A − BK)−1 E∞ ≤ 1. Condition 1) is
ẏn 0 0 · · · d(t) 0 yn easily satisfied given Assumption 1. Condition 2) is equivalent
⎡ ⎤ ⎡ ⎤
1 0 that there exists a positive definite P such that
⎢0⎥ ⎢ x2 ⎥
⎢ ⎥ ⎢ ⎥ (A + BK)T P + P(A + BK) + PE(PE)T + F T F < 0. (22)
⎢ .. ⎥ ⎢ ⎥
+ ⎢ . ⎥(u2 − u2d ) + ⎢ ... ⎥(u1 − u1d ) (16)
⎢ ⎥ ⎢ ⎥ Thus inequality (22) is then transformed to
⎣0⎦ ⎣ xn−2 ⎦
⎡  T ⎤
0 xn−1 P−1 (A + BK)T + (A + BK)P−1 E FP−1
⎣ ET −I 0 ⎦ < 0. (23)
and −1
FP 0 −I
ẏ1 = u1 − u1d . (17) Finally, we can use the LMI techniques to solve inequality (23)
By defining Y = (y2 , y3 , . . . , yn ) and (t) = [(d(t) − D)/(T)], so as to determine the matrix K. To see more details, please
then (16) can be arranged as refer to our previous work [19].
Remark 2: Since (21) is of quadratical stability by choosing
Ẏ = (A + E(t)F)Y + B(u2 − u2d ) (18) a Lyapunov function V1 = γ Y T PY, where γ is a positive
constant [25], we have V̇1 < −α1 Y2 where α1 > 0 is a
where constant [19].
⎡ ⎤
0 0 ... 0 0 Remark 3: As [(dV1 )/(dY)] = 2γ PY ≤ PY, there
⎢D 0 ... 0 0⎥ exists η such that [(dV1 )/(dY)] ≤ ηY [24] and η is a
⎢ ⎥
⎢ .. ⎥
A = ⎢ ... ..
.
..
.
..
. .⎥
positive constant.
⎢ ⎥ For subsystem (17), we choose the linear stabilization
⎣0 0 ... 0 0⎦ controller
0 0 ... D 0
⎡ ⎤ u1 = u1d − k1 y1 , k1 > 0 (24)
0 0 ... 0 0
⎢1 0 ... 0 0⎥
⎢ ⎥ to achieve globally exponential stabilization of the correspond-
⎢ .. ⎥
E = ⎢ ... ..
.
..
.
..
. .⎥ ing closed-loop system by choosing the Lyapunov function
⎢ ⎥
⎣0 0 ... 0 0⎦ V2 = (1/2)y21 [19].
0 0 ... 1 0 Remark 4: According to the boundedness of the
⎡ ⎤ solution of (16) of [24], (0, x2 , . . . , xn−2 , xn−1 )T  is
⎡ ⎤ 1
T 0 ... 0 0 ⎢0⎥ upper-bounded by some positive constant, say α2 , i.e.,
⎢0 T ... 0 0⎥ ⎢ ⎥ (0, x2 , . . . , xn−2 , xn−1 )T  ≤ α2 [19].
⎢ ⎥ ⎢. ⎥
F = ⎢. .. .. .. .. ⎥, B = ⎢ .. ⎥
⎣ .. . . . .⎦ ⎢ ⎥ Remark 5: According to [19] and [24], a Lyapunov func-
⎣0⎦ tion is constructed to guarantee exponentially stability of the
0 0 ... 0 T
0 cascaded system (15) as follows:
interconnected with V3 = V1 + δV2 (25)

fint (x, (u1 − u1d )) = g(y1 , Y)(u1 − u1d ) (19) where δ = [(η2 α22 )/(2α1 )].
Remark 6: From [24, Lemma 8], we know that V̇3 < 0.
where g(y1 , Y) = (0, x2 , . . . , xn−2 , xn−1 )T . Therefore, the error dynamics is stabilized with (20) and (24).
Remark 1: Error dynamics (15) can be viewed as a To this end, we have completed the design of the sub-
cascaded system with (18) cascaded by (17) and its controllers u1 and u2 at the kinematic level under the frame-
interconnection term is fint (x, (u1 , u1d )) [19]. Then under the work of the cascaded design method, then we will design the
framework of cascaded design method [22], [24], a kinematic controller at dynamic level by using backstepping techniques
controller is then derived. and we will use u1 , u2 as virtual inputs u1 and u2 in the con-
trol design at dynamic level. To further design, we need the
B. Controller Design for Kinematic and Dynamic Subsystems Assumption 3, Property 1 and Property 2 [19].
In this section, we will design subcontrollers for subsys- Assumption 3: The determinant of RT2 R2 is not zero for all
tems (17) and (18) of the kinematic systems gained above y ∈ Rn and t ∈ R+ .
at kinematic level, and then, a controller at dynamic level is Property 1: D3 = RT2 D2 (q)R2 is a positive definite matrix.
derived. For subsystem (18), a controller is given as Property 2: An appropriate definition of C(·, ·) in (2) ren-
ders (Ḋ − 2C) skew-symmetric [11], [19]. Thus Ḋ3 − 2RT2 C2
u2 = u2d + KY. (20) is skew-symmetric.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

 
Let u = [u1 , u2 ]T and u = [u1 , u2 ]T , the error is gained as = εT N T + RT2 B2 KN I + J2T λ − G2
  
ε = [ε1 , ε2 ]T = u1 − u1 , u2 − u2 (26) − C2 (x, ẋ)u − D2 (x)R2 (x)u̇ (35)

where u = [u1 , u2 ]T was designed by (20) and (24). The error where N = [(y1 +2γ Y T Pg(y1 , Y), 2γ Y T PBε2 ]. Then, we have
dynamics is gained by differentiating (26). the desired controller at dynamic level
D2 (x)R2 (x)ε̇ = −C2 (x, ẋ)ε + B2 KN I + J2T λ − G2  −1  
B2 KN I = RT2 − ε − N T − J2T λ + G2
− C2 (x, ẋ)u − D2 (x)R2 (x)u̇ . (27)
+ C2 (x, ẋ)u + D2 (X)R2 (x)u̇ (36)
A Lyapunov function is constructed for (15) and (27)
where is a positive definite matrix and RT2 is nonsigular.
V5 = V3 + V4 (28)
Then, we have
where V3 is defined by (25), and we know that the δ in (25) is
positive constant, so we choose δ value as 1 just for simplicity V̇5 < −εT ε ≤ 0. (37)
in the following design procedure. Thus, we take V3 = V1 +V2 .
Differentiating V3 one has Remark 7: Noticing the form of the controller designed
above, no matter whether the matrix RT2 is invertible or not,
dV2 dV1 we can modify the form of (36) by multiplying the term RT2
V̇3 = +
dt dt
  on both sides of Equation (36). Thus, noting that RT2 J2T = 0
= y1 ẏ1 + γ Ẏ T PY + Y T PẎ we obtain that
 
= −k1 y21 + ε1 y1 + γ Ẏ T PY + Y T PẎ (29)
   RT2 B2 KN I = − ε − N T + RT2 G2 + RT2 C2 u + RT2 D2 R2 u̇
Y T PẎ = Y T P (A + EF)Y + B u2 + ε2 − u2d
 (38)
+ g(y1 , Y)(−k1 y1 + ε1 )

= Y T P (A + BK + EF)Y − g(y1 , Y)k1 y1 and in turn, we have

+ Bε2 + g(y1 , Y)ε1 . (30)  −1 
I = RT2 B2 KN − ε − N T + RT2 G2 + RT2 C2 u
Combining (29) and (30) and noting Ẏ T PY = Y T PẎ yield 
+ RT2 D2 R2 u̇ . (39)
V̇3 = −k1 y21 + ε1 y1
 
+ 2γ Y T P ((A + BK + EF)Y − g(y1 , Y)k1 y1 ) C. Controller Design at the Actuator Level
+ 2γ Y T PBε2 + 2γ Y T Pg(y1 , Y)ε1 . (31) To this end, we have accomplished the stage of the tracking
From Remark 6, we know that V̇3 < 0 by the virtual con- control for the kinematic system and the dynamic one. The
trollers u1 , u2 . Therefore, it is clear that we only need to cancel tracking controller at the actuator level will be constructed in
the second, and the last two terms [19]. Then we have the time in the following stage, as similarly did in Fu and Chai [28].
derivative of V Let I = [I1 , I2 ]T and I = [I1 , I2 ]T , we have the error
dV3 dV4  T
V̇5 = +  = [1 , 2 ]T = I1 − I1 , I2 − I2 . (40)
dt dt
dV4
< ε1 y1 + 2γ Y T PBε2 + 2γ Y T Pg(y1 , Y)ε1 + . (32) Noting (40), one has
dt
Noting that V˙4 = (1/2)(ε̇T D3 ε) + (1/2)(εT D3 ε̇) + I =I + (41)
(1/2)(εT Ḋ3 ε), we have ˙
İ = I˙ + . (42)
−1
ε D3 ε̇ = ε D3 (D2 R2 ) (−C2 (x, ẋ)ε + B2 KN I
T T
 Noticing (14), we obtain
+ J2T λ − G2 − C2 (x, ẋ)u − D2 (x)R2 (x)u̇
(33) Lİ + RI + Ka Q = V. (43)
and noting Property 1, we have Substituting I and İ in (43), with (41) and (42), one has
1       
V̇4 = εT Ḋ3 − 2RT2 C2 ε + εT RT2 B2 KN I + J2T λ L İ +  ˙ + R I +  + Ka Q = V.
2 
− G2 − C2 (x, ẋ)u − D2 (x)R2 (x)u̇ . (34)
Then we can get the error dynamics
From (32) and (34) and Property 2, we have
˙ = V − Lİ − RI − R − Ka Q.
L (44)
V̇5 < ε1 y1 + 2γ Y T PBε2 + 2γ Y T Pg(y1 , Y)ε1
1   Choose a Lyapunov function candidate for (15), (27), and (44)
+ εT Ḋ3 − 2RT2 C2 ε + εT RT2 B2 KN I + J2T λ
2
− G2 − C2 (x, ẋ)u − D2 (x)R2 (x)u̇ Vf = V3 + V4 + V6 = V5 + V6 (45)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

FU et al.: MOTION TRACKING CONTROL DESIGN FOR CLASS OF NONHOLONOMIC MOBILE ROBOT SYSTEMS 5

where V5 is defined by (28) and V6 = (1/2)T L in which

L = diag[L1 , L2 , L3 , . . . , Lr ] is symmetric and positive definite
defined in Section II. Differentiating (45) yields
V˙f = V̇5 + V̇6 . (46)
Noting (35), we have
 
V˙f < εT N T + RT2 B2 KN I + J2T λ − G2 − C2 (x, ẋ)u

− D2 (x)R2 (x)u̇ + V̇6
   
= εT N T + RT2 B2 KN I +  + J2T λ − G2

− C2 (x, ẋ)u − D2 (x)R2 (x)u̇ + V̇6
< εT RT2 B2 KN  + V̇6
˙
= εT RT2 B2 KN  + T L
 T T 
=  KN B2 R2 ε + V − Lİ − RI − R − Ka Q .
T
(47)
Design the state feed-back control law
V = −KNT BT2 R2 ε + Lİ + RI +R + Ka Q −  (48)
where  is a constant matrix, which is positive definite. Fig. 1. Simplified nonholonomic mobile robot.
Substituting V in (47) with (48), we have
V˙f < −T  < 0. (49)
Remark 8: The methods proposed in this paper and [21]
have two similar stages. Although Stage II needs similar com-
putations for both methods, to implement the controller in [21]
in Stage I, the virtual controller z shown in (19) needs calcula-
tion of first up to (n − 1)th derivatives of the given ud1 , while
our method even does not need its first derivative in Stage I.
Therefore, the method in [21] may result in heavy computing
burden, especially when applied to high order nonholonomic
systems.

IV. S IMULATION R ESULTS

To verify the proposed control method, we choose a mobile
wheeled robot in [26], shown in Fig. 1. The robot is constituted
by a rigid trolley equipped with three wheels. There are two
rear driving wheels with one motor for each wheel and a front
wheel only serving as the supporting wheel. More details may
be found in [21] Fig. 2. Tracking errors.
1
mẍ = λ cos θ − (KN1 I1 + KN2 I2 ) sin θ The nonholonomic constraint is characterized as
P
1
mÿ = λ sin θ + (KN1 I1 + KN2 I2 ) cos θ ẋ cos θ + ẏ sin θ = 0. (52)
P
E Choose a circular path on the plane as the desired trajectory,
I0 θ̈ = (KN1 I1 − KN2 ). (50) which is qd = [2 cos θ, 2 sin θ, t]T . Our objective is to deter-
P      mine a control law V so that the trajectory q = [x, y, θ]T
L1 0 İ1 R1 0 I1
+ follows qd .
0 L2 İ2 0 R2 I2
     We choose R(q) defined in (6) as
Ka1 0 ω1 V1 ⎡ ⎤
+ = (51) − sin θ 0
0 Ka2 ω2 V2
R = ⎣ cos θ 0 ⎦. (53)
where x and y are the coordinates O which is the reference
0 1
point in the inertia frame, θ is the orientation of the system
basis to the inertia frame. m is the mass and I0 is inertia Using a transformation x = (q) and a feedback control
moment for the vertical axis at O. P is the radius of the v = (q)u, we can convert the kinematic system into the
two rear wheels and 2E is the length of the axis of the rear chained-form system. More details are contained in [1]. For
wheels. V1 and V2 are the voltage for the motors of the two R(q), the desired trajectory qd = [2 cos θ, 2 sin θ, t]T satisfies
rear wheels [21]. q̇d = R(qd )vd , with vd1 = 2 and vd2 = 1. By utilizing the same
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS

The initial values for the positions and velocities of the

wheeled robot are q(0) = [3, 0, 0.5] and q̇(0) = [0, 0, 0].
The simulation results are presented in Figs. 2–4. Fig. 2
shows the tracking trajectory errors of x, y, and θ with time
as e1, e2, and e3, respectively. Fig. 3 represents the phase
trajectory of the reference signal and the followed one. And
Fig. 4 characterizes the geometric trajectory of x and y. The
trajectory followed by the nonholonomic robot and the track-
ing errors has verified the control method proposed in this
paper.
V. C ONCLUSION
The trajectory tracking has been addressed for the nonholo-
nomic robot system which includes actuator dynamics. A new
motion tracking controller is constructively designed at actua-
tor level which is our main goal in this paper, meanwhile, we
use the cascaded design method in the process of the controller
design at the kinematic level in order to simplify the controller
structure. The proposed method guarantees that the states of
Fig. 3. Phase trajectories. the system asymptotically tracks the desired trajectory. The
ongoing work is about combining the cascaded design and the
adding a power integrator technique [27] to cope with track-
ing problem of high-order nonholonomic systems in power
chained form. To validate the method in this paper more accu-
rately, the experimental comparison between this method and
other methods, especially the method in [21] on a complex
physical system is our focus in the future work.
R EFERENCES
[1] M. Oya, C.-Y. Su, and R. Katoh, “Robust adaptive motion/force tracking
control of uncertain nonholonomic mechanical systems,” IEEE Trans.
Robot. Autom., vol. 19, no. 1, pp. 175–181, Feb. 2003.
[2] A. De Luca, G. Oriolo, and C. Samson, “Feedback control of a non-
holonomic car-like robot,” in Robot Motion Planning and Control.
Heidelberg, Germany: Springer, 1998, pp. 171–253.
[3] C. Samson, “Control of chained systems application to path follow-
ing and time-varying point-stabilization of mobile robots,” IEEE Trans.
Autom. Control, vol. 40, no. 1, pp. 64–77, Jan. 1995.
[4] J. Fu, R. Ma, and T. Chai, “Global finite-time stabilization of a class
of switched nonlinear systems with the powers of positive odd rational
numbers,” Automatica, vol. 54, pp. 360–373, Apr. 2015.
Fig. 4. Geometric trajectory of x via y. [5] J. Rubio-Hervas and M. Reyhanoglu, “Nonlinear control of a robot
manipulator with a nonholonomic jerk constraint,” Asian J. Control,
transformation x = (q), the kinematic system q̇d = Rqd vd vol. 18, no. 4, pp. 1566–1573, 2016.
[6] W. Sun, S. Tang, H. Gao, and J. Zhao, “Two time-scale tracking control
can be converted to of nonholonomic wheeled mobile robots,” IEEE Trans. Control Syst.
xd1 = t Technol., vol. 24, no. 6, pp. 2059–2069, Nov. 2016.
[7] M. Li, Y. Li, S. S. Ge, and T. H. Lee, “Adaptive control of robotic
xd2 = 0 manipulators with unified motion constraints,” IEEE Trans. Syst., Man,
Cybern., Syst., vol. 47, no. 1, pp. 184–194, Jan. 2017.
xd3 = 2 (54) [8] Z. Li et al., “Trajectory-tracking control of mobile robot systems incor-
with ud1 = 1 and ud2 = 0. It is easy to know that ud and xd porating neural-dynamic optimized model predictive approach,” IEEE
Trans. Syst., Man, Cybern., Syst., vol. 46, no. 6, pp. 740–749, Jun. 2016.
satisfy Assumptions 1 and 2, respectively. The controller (48) [9] C. Chen, Z. Liu, Y. Zhang, and S. Xie, “Coordinated motion/force
is used and the physical parameters for this simulation are control of multiarm robot with unknown sensor nonlinearity and manip-
chosen as m = 0.5, I0 = 0.5, E = P = 1 L1 = L2 = 2.03, ulated object’s uncertainty,” IEEE Trans. Syst., Man, Cybern., Syst.,
vol. 47, no. 7, pp. 1123–1134, Jul. 2017.
R1 = R2 = 2, KN1 = KN2 = 1.5, and Ka1 = Ka2 = 1. And [10] I. Kolmanovsky and N. H. Mcclamroch, “Developments in nonholo-
the Q is taken as nomic control problems,” IEEE Control Syst., vol. 15, no. 6, pp. 20–36,
  1  Dec. 1996.
Q1 (u (x cos θ + y sin θ ) + u2 (1 + E)) [11] Z. Li, S. S. Ge, M. Adams, and W. S. Wijesoma, “Robust adap-
= P1 1 . tive control of uncertain force/motion constrained nonholonomic mobile
P (u1 (x cos θ + y sin θ ) + u2 (1 − E))
Q2
manipulators,” Automatica, vol. 44, no. 3, pp. 776–784, 2008.
Using the above diffeomorphism transformation, Q can be [12] S. S. Ge, Autonomous Mobile Robots: Sensing, Control, Decision
Making and Applications, vol. 22. Boca Raton, FL, USA: CRC Press,
expressed as 2006.
  1  [13] R. V. Cowlagi and Z. Zhang, “Motion-planning with linear temporal
Q1 (u1 x3 + u2 (1 + E))
= 1 P . logic specifications for a nonholonomic vehicle kinematic model,” in
P (u1 x3 + u2 (1 − E))
Q2 Proc. Amer. Control Conf., Boston, MA, USA, 2016, pp. 6411–6416.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

FU et al.: MOTION TRACKING CONTROL DESIGN FOR CLASS OF NONHOLONOMIC MOBILE ROBOT SYSTEMS 7

[14] J. Jarom, “Kinematic model of nonholonomic mobile robots,” Appl. Tianyou Chai received the Ph.D. degree in con-
Mech. Mater., vol. 611, pp. 107–114, Aug. 2014. trol theory and engineering from Northeastern
[15] M. B. Linan and M. C. Munoz-Lecanda, “Strict abnormal extremals University, Shenyang, China, in 1985.
in nonholonomic and kinematic control systems,” in Discrete and He became a Professor with Northeastern
Continuous Dynamical Systems—Series S. Springfield, MO, USA: Amer. University in 1988. He is the Founder and the
Inst. Math. Sci., 2008, pp. 1–17. Director of the Center of Automation, which became
[16] A. M. Bloch, M. Reyhanoglu, and N. H. Mcclamroch, “Control and a National Engineering and Technology Research
stabilization of nonholonomic dynamic systems,” IEEE Trans. Autom. Center and a State Key Laboratory. He is the
Control, vol. 37, no. 11, pp. 1746–1757, Nov. 1992. Director of the Department of Information Science,
[17] Z. Zhang and Y. Wu, “Adaptive motion and force tracking control for National Natural Science Foundation of China. He
nonholonomic dynamic systems subject to affine constraints,” Trans. has published 170 peer reviewed international jour-
Inst. Meas. Control, vol. 38, no. 4, pp. 482–491, 2015. nal papers. He has developed control technologies with applications to various
[18] Y.-Q. Wu, C.-L. Zhu, and Z.-C. Zhang, “Finite-time stabilization of a industrial processes. His current research interests include modeling, control,
general class of nonholonomic dynamic systems via terminal sliding optimization, and integrated automation of complex industrial processes.
mode,” Int. J. Autom. Comput., vol. 13, no. 6, pp. 585–595, 2015. Dr. Chai was a recipient of 4 prestigious awards of National Science and
[19] J. Fu, T. Chai, C.-Y. Su, and Y. Jin, “Motion/force tracking control of Technology Progress and National Technological Innovation and the 2007
nonholonomic mechanical systems via combining cascaded design and Industry Award for Excellence in Transitional Control Research from IEEE
backstepping,” Automatica, vol. 49, no. 12, pp. 3682–3686, 2013. Multiple-conference on Systems and Control. His paper titled “Hybrid intel-
[20] M. C. Good, L. M. Sweet, and K. L. Strobel, “Dynamic models for ligent control for optimal operation of shaft furnace roasting process” was
control system design of integrated robot and drive systems,” J. Dyn. selected as one of three best papers for the Control Engineering Practice
Syst. Meas. Control, vol. 107, no. 1, pp. 53–59, 1985. Paper Prize for 2011 C2013. He is a member of the Chinese Academy of
[21] C. M. Anupoju, C.-Y. Su, and M. Oya, “Adaptive motion tracking
Engineering, IFAC Fellow.
control of uncertain nonholonomic mechanical systems including actu-
ator dynamics,” IEE Proc. Control Theory Appl., vol. 152, no. 5, Yuanwei Jing received the B.S. degree in math-
pp. 575–580, Sep. 2005. ematics from Liaoning University, Shenyang,
[22] Y.-P. Tian and K.-C. Cao, “An LMI design of tracking controllers for China, in 1981, and the M.S. and Ph.D. degrees
nonholonomic chained-form system,” in Proc. Amer. Control Conf., in automatic control from Northeastern University,
New York, NY, USA, 2007, pp. 4512–4517. Shenyang, in 1984 and 1988, respectively.
[23] P. Martin and P. Rouchon, “Feedback linearization and driftless From 1998 to 1999, he was a Senior
systems,” Math. Control Signals Syst., vol. 7, no. 3, pp. 235–254, 1994. Visiting Scholar with the Computer Science
[24] E. Panteley and A. Loria, “On global uniform asymptotic stability of Telecommunication Program, University Missouri–
nonlinear time-varying systems in cascade,” Syst. Control Lett., vol. 33, Kansas City, Kansas City, MO, USA. He is
no. 2, pp. 131–138, 1998. currently with the School of Information Science
[25] P. P. Khargonekar, I. R. Petersen, and K. Zhou, “Robust stabilization and Engineering, Northeastern University. His
of uncertain linear systems: Quadratic stabilizability and H∞ control
current research interests include complex control systems and control
theory,” IEEE Trans. Autom. Control, vol. 35, no. 3, pp. 356–361,
problems in modern communication network systems.
Mar. 1990.
[26] G. Campion, B. D’Andrea-Novel, and G. Bastin, “Controllability and Zhijun Li (M’07–SM’09) received the Ph.D.
state feedback stabilizability of non holonomic mechanical systems,” degree in mechatronics from Shanghai Jiao Tong
in Advanced Robot Control. Heidelberg, Germany: Springer, 1991, University, Shanghai, China, in 2002.
pp. 106–124. From 2003 to 2005, he was a Post-Doctoral
[27] W. Lin and C. Qian, “Adding one power integrator: A tool for global Fellow with the Department of Mechanical
stabilization of high-order lower-triangular systems,” Syst. Control Lett., Engineering and Intelligent Systems, University
vol. 39, no. 5, pp. 339–351, 2000. of Electro-Communications, Tokyo, Japan. From
[28] J. Fu and T. Chai, “Motion tracking control of nonholonomic systems 2005 to 2006, he was a Research Fellow with
including actuator dynamics,” in Proc. IEEE Conf. Adv. Robot. the Department of Electrical and Computer
Mechatronics, Hefei, China, 2017, pp. 25–30. Engineering, National University of Singapore,
Jun Fu received the Ph.D. degree in mechanical Singapore, and Nanyang Technological University,
engineering from Concordia University, Montreal, Singapore. Since 2012, he has been a Professor with the College of
QC, Canada, in 2009. Automation Science and Engineering, South China University of Technology,
He was a Post-Doctoral Fellow/Associate Guangzhou, China. His current research interests include service robotics,
with the Department of Mechanical Engineering, teleoperation systems, nonlinear control, and neural network optimization.
Massachusetts Institute of Technology, Cambridge, Prof. Li has been the Chair of the Technical Committee on Biomechatronics
MA, USA, from 2010 to 2014. He is a Full and Biorobotics Systems of the IEEE Systems, Man and Cybernetics Society
Professor with Northeastern University, Shenyang, since 2016. He is an Editor-at-Large of the Journal of Intelligent and Robotic
China. He has authored/coauthored over 70 Systems and an Associate Editor of several IEEE T RANSACTIONS. He was
publications which appeared in journals, conference the General Chair and the Program Chair of the 2016 and 2017 IEEE
proceedings, and book chapters. His current Conference on Advanced Robotics and Mechatronics, respectively.
research interests include switched systems, robust control, mathematical
Chun-Yi Su (M’93–SM’99) received the Ph.D.
programming, and dynamic optimization.
degree in control engineering from the South China
Dr. Fu was a recipient of the 2016 Young Scientist Award from Chinese
University of Technology, Guangzhou, China, in
Association of Automation. He is currently an Associate Editor of the
1990.
IEEE T RANSACTIONS ON S YSTEMS , M AN , AND C YBERNETICS : S YSTEMS,
He was with the University of Victoria, Victoria,
IEEE T RANSACTIONS ON N EURAL N ETWORKS AND L EARNING S YSTEMS ,
BC, USA, for seven years. He joined Concordia
C ONTROL E NGINEERING P RACTICE, and the Journal of Industrial and
University, Montreal, QC, Canada, in 1998, where he
Management Optimization.
is a Professor and the Concordia University Research
Fangyin Tian received the B.S. degree in measure- Chair. He has authored or coauthored over 300 pub-
ment and control technology and instrument from lications in journals, book chapters, and conference
Shenyang Ligong University, Shenyang, China, in proceedings. His current research interests include
2015 and the M.S. degree in control theory and the application of automatic control theory to mechanical systems. He is par-
control engineering from Northeastern University, ticularly interested in control of systems involving hysteresis nonlinearities.
Shenyang, in 2018. Dr. Su was a recipient of several best conference paper awards. He has
His current research interests include centered on served as an Associate Editor of the IEEE T RANSACTIONS ON AUTOMATIC
the nonholonomic mobile robot control, dynamic C ONTROL, IEEE T RANSACTIONS ON C ONTROL S YSTEMS T ECHNOLOGY,
optimization, and nonlinear control. and Journal of Control Theory and Applications. He has been on the Editorial
Mr. Tian was a recipient of the First Prize of Board of 18 journals, including the IFAC Journal of Control Engineering
2012 Mathematics Competition of Chinese College Practice and Mechatronics. He served for many conferences as an Organizing
Students. Committee Member.