Modern Physics Letters B

(2025) 2550119 (26 pages)

.c World Scienti¯c Publishing Company
#
DOI: 10.1142/S0217984925501192
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Adaptive fast terminal sliding mode trajectory tracking control

of Mecanum-wheeled omnidirectional mobile robots using
barrier function

Long Chen *, Zeyuan Lu †, Bin Yan ‡, Peipei Jin § and Guangyi Wang ¶

Institute of Modern Circuit and Intelligent Information,

Hangzhou Dianzi University,
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

Hangzhou, Zhejiang, China

*[email protected]
†
[email protected]
‡[email protected]
§
[email protected]
¶
[email protected]

Received 14 October 2024

Revised 7 November 2024
Accepted 15 November 2024
Published 4 February 2025

This paper presents a barrier function-based adaptive nonsingular fast terminal sliding mode
control (BFANFTSMC) algorithm for the trajectory tracking of a Mecanum-wheeled omnidi-
rectional mobile robot (MWOMR). A four-inputs-three-outputs kinematic-and-dynamic model
of the MWOMR is constructed with considerations of parameter uncertainties and unknow
external disturbances. To achieve a stable and robust performance, a BFANFTSMC scheme is
designed for the MWOMR, while a modi¯ed barrier function (MBF) is further proposed with
particular consideration of actuator saturation. Stability and robustness of the proposed control
methods are analyzed using the Lyapunov theory. Also, real-time experimental results on a
MWOMR platform show superior performance on tracking precision and control robustness
compared with three existing control schemes.

Keywords: Nonsingular fast terminal sliding mode control (NFTSMC); trajectory-tracking;

barrier function (BF); Mecanum-wheels omnidirectional mobile robot (MWOMR); actuator
saturation.

PACS Nos.: 02.30.Yy, 07.05.Dz

1. Introduction
Nowadays, mobile robots are widely used in agriculture, planetary exploration and
other domains.1,2 Traditional mobile robots mostly utilize rubber wheels, such as
four-wheeled or six-wheeled mobile robots,3,4 which require the information of turn

¶
Corresponding author.

2550119-1
 L. Chen et al.

radius and are restricted by the terrain. The Mecanum-wheeled omnidirectional

mobile robot (MWOMR) utilizes Mecanum wheels, which are conventional wheels
equipped with passive rollers along their circumference. Compared with traditional
mobile robots, omnidirectional mobile robots (OMRs) have been extensively inves-
tigated because of their enhanced mobility, omnidirectional movement and zero
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

turning radius,5{8 for instance, spherical robots and snake-like robots.5,6 This design
allows MWOMR to move in any direction without the need for reorientation.9
Due to these distinct advantages, the MWOMR has been widely used in narrow
environments, such as factories, warehouses, and greenhouses.10
In practice, the MWOMR may encounter uneven roads and even bumps from
external objects, which will bring a lot of uncertainties and nonlinearities to the
system. Therefore, the robustness and adaptability of the controller is crucial. The
accuracy of the MWOMR trajectory tracking is also the key to ensuring the quality
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

of mobile operation, so it is necessary to design an e®ective trajectory-tracking

control algorithm for MWOMR.11 To address the aforementioned control di±culties,
many researchers have utilized the Lagrange method to describe the kinematics and
dynamics of the MWOMR, i.e. a second-order state-space equation with four inputs
and four outputs,12 and a second-order state-space equation with four inputs and
three outputs.13,14 On the basis of kinematic and dynamic models, various control
algorithms have been developed, including distributed model predictive control,15
adaptive control,16 neural network control and sliding mode control (SMC), etc.17,18
As one of the best nonlinear controllers, SMC is favored by many scholars for its
robust control capabilities and reliable convergence properties.19,20 Typically, con-
ventional SMC exhibits in¯nite convergence time. To achieve ¯nite-time conver-
gence, a terminal sliding mode control (TSMC) method was introduced in Ref. 21.
This method enables rapid state convergence through the use of a nonlinear sliding
surface. However, the TSMC often encounters to the singularity problem due to the
presence of terms with negative fractional powers. Hence, a nonsingular terminal
sliding mode control (NTSMC) scheme was presented in Ref. 22. Further, a non-
singular fast terminal sliding mode control (NFTSMC) method was designed in
Ref. 23, known for its ¯nite-time convergence, singularity avoidance, and chattering
reduction properties, has been applied in mechatronic systems, including non-
holonomic wheeled mobile robot and Polaris-I cable-driven manipulators.24,25
Moreover, the NFTSMC has a faster convergence speed compared to the NTSMC
when the states are far from the origin.
On the other hand, for designing the SMC algorithm, it is crucial to account for
the upper bound of disturbances. However, obtaining precise disturbance informa-
tion in practical applications is often challenging.26 To mitigate the disturbance's
e®ects, an overestimation control gain is commonly employed, which can lead to
signi¯cant chattering issues. Consequently, di®erent approaches were developed for
the elimination or alleviation of chattering, i.e. disturbance observer,27 neural net-
work control,28 and adaptive SMC,29 etc. Xi et al.30 proposed a novel sliding mode
disturbance observer coupled with a new variable structure scheme, which e®ectively

2550119-2
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

compensates for interference while simultaneously mitigating the chattering phe-

nomenon. Han et al.31 developed an adaptive tuning approach that adjusts the
control law without frequency switching, e®ectively mitigating the undesired chat-
tering phenomenon. Among these approaches, the barrier function is a smooth
function de¯ned in the vicinity of zero. As the variable approaches zero, the output
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

also approaches zero. Conversely, when the variable nears the boundary of this
region, the output tends to in¯nity.32,33 In Ref. 34, the barrier function is combined
with the NFTSMC, demonstrating that this approach e®ectively reduces the chat-
tering phenomenon while constraining the sliding mode variable within a prede¯ned
region in ¯nite time. In Ref. 35, a barrier function-based nonsingular terminal sliding
mode (BFNTSM) control is proposed, incorporating a barrier function into the
controller to simultaneously eliminate chattering and prevent gain overestimation.
Furthermore, in many control systems, achieving faster convergence often necessi-
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

tates larger control inputs.36,37 Nevertheless, these substantial control inputs can
result in actuator saturation. To address this issue, we have introduced the inno-
vative concept of a modi¯ed barrier function (MBF) designed to restrict the maxi-
mum control gain and guarantee system stability.
In this paper, we propose a barrier function-based adaptive nonsingular fast
terminal sliding mode control (BFANFTSMC) scheme for the trajectory-tracking
task of MWOMR. The main contributions are summarized as follows:

(1) The kinematic and dynamic models of the MWOMR incorporate considerations
for the nominal model, parameter uncertainties, and external disturbances.
Leveraging the rapid convergence and nonsingular features of the NFTSMC, we
have developed NFTSMC for trajectory tracking in MWOMR. This expansion in
the application of the NFTSMC scheme widens its utility.
(2) To reduce control chattering, this paper introduces a combination of barrier
functions and SMC to create the BFANFTSMC scheme for MWOMR. This
controller e®ectively constrains the sliding mode variable and tracking error
within a prede¯ned small region in ¯nite time, even without prior knowledge of
the upper bound of disturbances.
(3) Considering the MWOMR is driven by DC motors with speeds related to pulse
width modulation (PWM) duty ratio, there will be the potential problem of
actuator saturation. To tackle this problem, an MBF-based adaptive NFTSMC
control (MBFANFTSMC) for MWOMR is designed with the adaptive control
gain to further reduce the control input while maintaining the small tracking
error bound, such that the actuation saturation can be avoided.

The rest of this paper is structured as follows. Section 2 provides the dynamic and
kinematic models of MWOMR. In Sec. 3, we design the BFANFTSMC and
MBFANFTSMC schemes for robot trajectory tracking, and the stability analysis is
presented in detail. In Sec. 4, we showcase real-time experimental results from an
MWOMR experimental platform to highlight the superiority of the proposed control

2550119-3
 L. Chen et al.

strategy compared to the conventional sliding mode control (CSMC), NTSMC and
NFTSMC algorithms. Finally, Sec. 5 serves as the conclusion of this paper.

2. Model Description
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

2.1. Kinematics of MWOMR

In the kinematic models of the MWOMR, we consider three coordinate frames, as
depicted in Fig. 1.
In Fig. 1, we de¯ne Or Xr Yr as the global coordinate frame (GCF). Ob Xb Yb
represents the body coordinate frame (BCF). Herein, P r ¼ ½xr yr
r T and P b ¼
½xb yb
b T are de¯ned as the poses of the MWOMR in the GCF and BCF, respec-
tively. The kinematic model can be expressed as follows16:
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

2: 3 2 3
xb 1 1 1 1
6 y: 7 R 6 1 1 1 1 7 :
4 b5¼ 6 7 µ; ð1Þ
: 44 1 1 1 1 5

b
LþW LþW LþW LþW
: : : : : :
where R is the radius of the Mecanum wheels, i in ! ¼ µ ¼ ½1 2 3 4 T is the
angular velocity of the ith wheel, and L and W are lengths as shown in Fig. 1. The
relationship between the GCF and BCF is expressed as
: :
½x_ r y_ r
r T ¼ ­ð
r Þ½x_ b y_ b
b T ; ð2Þ

where
2 3
cosð
r Þ  sinð
r Þ 0
­ð
r Þ ¼ 4 sinð
r Þ cosð
r Þ 0 5: ð3Þ
0 0 1

Fig. 1. (Color online) Top view of MWOMR.

2550119-4
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

Combining Eqs. (1) and (2), we obtain

: R :
½x_ r y_ r
r T ¼ hð
Þµ; ð4Þ
4
where
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

2 3
 sinð
Þ cosð
Þ  sinð
Þ cosð
Þ
pffiffiffi6 7
hð
Þ ¼ 24 cosð
Þ sinð
Þ cosð
Þ sinð
Þ 5;
pffiffiffi pffiffiffi pffiffiffi pffiffiffi
1= 2ðL þ W Þ 1= 2ðL þ W Þ 1= 2ðL þ W Þ 1= 2ðL þ W Þ
ð5Þ
2 pffiffiffi pffiffiffi 3
 2 sinð
Þ 2 cosð
Þ LþW
6 pffiffiffi pffiffiffi 7
1 6 2 cosð
Þ 2 sinð
Þ ðL þ W Þ 7
¼ 6 7;
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

hð
Þ1 pffiffiffi pffiffiffi ð6Þ

464  2 sinð
Þ 2 cosð
Þ
7
ðL þ W Þ 5
pffiffiffi pffiffiffi
2 cosð
Þ 2 sinð
Þ LþW
where


¼
r þ : ð7Þ
4
According to Eq. (4), we can obtain the second-order derivative of Pr as follows:
:: R : R ::
Pr ¼ hð
Þ0 µ þ hð
Þµ; ð8Þ
4 4
where
2 3
 cosð
Þ  sinð
Þ  cosð
Þ  sinð
Þ
pffiffiffi :
hð
Þ0 ¼ 24  sinð
Þ cosð
Þ  sinð
Þ cosð
Þ 5
: ð9Þ
0 0 0 0

2.2. Dynamic model of MWOMR

The dynamic model of the MWOMR is described by42
:: :
Jµ þ bµ þ ± ¼ Km u; ð10Þ

where J represents the nominal values of rotational inertia of each motor-and-

Mecanum-wheel system; b denotes the viscous friction of each motor-and-Mecanum-
wheel system; ± ¼ ½1 2 3 4 T are the lumped parametric uncertainties existing in
the system; u ¼ ½u1 u2 u3 u4 T corresponds to the input voltages for the respective
four motors and Km is the nominal motor constant. The parametric uncertainties are
de¯ned as follows:
J ¼ J0 þ J; ð11Þ
b ¼ b0 þ b: ð12Þ

2550119-5
 L. Chen et al.

By substituting Eqs. (11) and (12) into Eq. (10), the dynamics can be rewritten as
:: :
J0 µ þ b0 µ þ d ¼ Km u; ð13Þ
where J0 and b0 are the normal value corresponding parametric, J:: and b
:
are the
uncertainties corresponding parametric, d ¼ ½d1 d2 d3 d4 T ¼ Jµ þ bµ þ ± are
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

lumped uncertainties existing in the system.

Equation (8) can be rewritten as in Ref. 42
:: R :: ::
Pr ¼ hð
ÞðG
µ þ J0 µÞ; ð14Þ
4J0
where
2 3
0 J0 =2 0 J0 =2
6 J =2 0 J0 =2 0 7
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

6 0 7
G¼6 7: ð15Þ
4 0 J0 =2 0 J0 =2 5
J0 =2 0 J0 =2 0
Di®erent from Ref. 42, by combining Eqs. (13) and (15), we can derive the ¯nal
equivalent plant model, expressed as follows:
:: R : :
Pr ¼ hð
ÞðGµ  b0 µ  d þ Km uÞ
4J0
 
R : 4 1
: 4 1
:
¼ hð
Þ G
hð
Þ P r  d þ Km u  b0 hð
Þ P r ;
4J0 R R
: R
¼ AP r þ gð
Þu  hð
Þd; ð16Þ
4J0
where
2 : 3
b0 =J0 
0
6 : 7
A¼4
b0 =J0 0 5 ð17Þ
0 0 b0 =J0
and
Km R
gð
Þ ¼ hð
Þ: ð18Þ
4J0
Then the dynamic model equation of the MWOMR can be rewritten as
:: : R
P r ¼ AP r þ gð
Þu  hð
Þd: ð19Þ
4J0
From Eq. (19), we de¯ne
2 pffiffiffi 3
2ðsinð
Þðd1 þ d3 Þ  cosð
Þðd2 þ d4 ÞÞ
6 pffiffiffi 7
D ¼ ½D1 D2 D3 T ¼ hð
Þd ¼ 4  2ðcosð
Þðd1 þ d3 Þ þ sinð
Þðd2 þ d4 ÞÞ 5; ð20Þ
1=ðL þ W Þðd1 þ d2 þ d3  d4 Þ

2550119-6
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

where di ði ¼ 1; 2; 3) are the components of disturbances along the X-axis, Y-axis

and orientation.
jDi j < ki ; i ¼ 1; 2; 3; ð21Þ
where ki ði ¼ 1; 2; 3) represent the bounds of disturbances' components along the
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

X-axis, Y-axis and orientation, respectively.

Remark 1. In contrast to the conventional dynamic modeling method for Mecanum
wheel robots, the proposed modeling framework presents two signi¯cant bene¯ts.
First, the acquisition of system parameters is simpli¯ed. Unlike the challenging-to-
measure parameters such as the robot's rotational inertia matrix and Coriolis
matrices, the motor's rotational inertia and other data can be directly provided by
the manufacturer. Second, the system's control inputs are converted from torque or
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

force to motor voltage inputs through some appropriate methods, which is much
more suitable for the control of the robot system.

3. Design of the Controller

In this section, we design a BFANFTSMC to facilitate the convergence of sliding
variables and tracking errors to prede¯ned regions within a ¯nite time. Furthermore,
to address the issue of actuator saturation, we introduce an MBF.

3.1. Design of BFANFTM

In this work, we formulate the system tracking error as
2 3 2 3
e1 xr  xd
e ¼ 4 e2 5 ¼ 4 yr  yd 5; ð22Þ
e3
r 
d
where P d ¼ ½xd yd
d T is the reference trajectory to be tracked by the
MWOMR. The sliding variable is introduced as
2 a p 3
e1 þ 1 sigðe1 Þ b þ 1 sigðe_ 1 Þ q
6 p 7
s ¼ ½s1 s2 s3 T ¼ 6 7
4 e2 þ 2 sigðe2 Þ b þ 2 sigðe_ 3 Þ q 5;
a
ð23Þ
a p
e3 þ 3 sigðe3 Þ b þ 3 sigðe_ 3 Þ q

where a, b and p, q are positive odd integers which satisfy the conditions 1 < p=q < 2,
p=q < a=b and sigðxÞk ¼ jxjk signðxÞ. i and i for i ¼ 1; 2; 3 are positive constants.
According to Ref. 38, when ei is far from the equilibrium point, the term i ðjei jÞa=b
plays a primordial role in the fast convergence rate of ei . Once ei approaches
the origin, then Eq. (23) provides the same convergence rate similar to that of
NTSMC. Therefore, compared to CSMC and NTSMC, NFTSMC exhibits a faster
convergence rate. Furthermore, unlike integral terminal sliding mode control (ITSMC),
NFTSMC is not dependent on the robot's initial state.16 Upon di®erentiating the

2550119-7
 L. Chen et al.

sliding mode variables mentioned above, we derive the following equation:

: : : ::
s ¼ e þ Ne þ Me ; ð24Þ
where
2 3
1 p
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

6 q je_ 1 j 0 0
p=q1
7
6 7
6 2 p 7
6 7
M ¼6 0 je_ jp=q1 0 7 ð25Þ
6 q 2 7
6 7
4 3 p 5
0 0 je_ jp=q1
q 3
and
2 a 3
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

1
je1 ja=b1 0 0
6 b 7
6 2 a 7
6 0 je2 ja=b1 0 7
N ¼6 7: ð26Þ
6 b 7
4 3 a 5
0 0 je3 ja=b1
b
:
Assuming s ¼ ½0 0 0T and neglecting all disturbances, the equivalent control
input, denoted as u eq , is
: : : ::
u eq ¼ gð
Þ1 ½M 1 ðe  N eÞ  AP r þ P d ; ð27Þ
where
2 q 3
je_ 1 j1p=q 0 0
6 1 p 7
6 q 7
6 0 je_ j1p=q 0 7
M 1 ¼6 7: ð28Þ
6 2 p 2 7
4 q 5
0 0 je_ j1p=q
3 p 3
To enhance the system's robustness, we introduce a reaching control input,
denoted as u bf :
R
u bf ¼  gð
Þ1 BsignðsÞ; ð29Þ
4J0
where B is the adaptive gain matrix, designed as follows:
^ 1; k
B ¼ diagfk ^2; k
^ 3 g: ð30Þ
Initially, we need to design three small positive numbers "i ði ¼ 1; 2; 3Þ which are
the eventual regions of the three sliding mode variables converge into, respectively.
Next, t i ði ¼ 1; 2; 3Þ represents the time when the corresponding sliding mode
variables reach the corresponding regions ½"i =2; "i =2 for the ¯rst time from any
initial state outside the region, jsi ð0Þj > "i =2. Otherwise, t i ¼ 0 if jsi ð0Þj  "i =2.

2550119-8
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

Then, the adaptive parameters are designed as follows:

Z t
^ i ¼ i
k exp ð i jsi jÞjsi jdt; for 0  t  t i ; i ¼ 1; 2; 3; ð31Þ
0
^ i ¼ fbi ðsi Þ;
k for t > t i ; i ¼ 1; 2; 3; ð32Þ
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

where
jsi j
fbi ðsi Þ ¼ ; for si 2 ð"i ; "i Þ; i ¼ 1; 2; 3: ð33Þ
"i  jsi j
The whole control input of the BFANFTSMC scheme can be constructed as
u ¼ u eq þ u bf : ð34Þ
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

Remark 2. Due to the integral term, the sliding mode variable si can quickly
converge to the predesigned region ½"i =2; "i =2, which means we can get smaller t i .
For t > t i , when the sliding mode variable s goes to zero, fb ðsÞ will also go to zero. In
addition, fb ðsÞ approaches in¯nity as the variable approaches the boundary of the
previously de¯ned region, which e®ectively inhibits the occurrence of chattering.

3.2. Stability analysis

The stability analysis for the proposed controller is provided below.
Lemma 3.1. Under the control law Eq. (34) for 0  t  t i (i ¼ 1; 2; 3), there exist
three positive numbers k1 , k2 , k3 in Eq. (21) and the adaptive control gain k^ i in
^ ^
Eqs. (31) and (32) that satisfy k i < k i for i ¼ 1; 2; 3.
The proof of Lemma 1 has been given in Ref. 39.
Theorem 3.1. Under the BFANFTSMC scheme in Eq. (34), the sliding variables
si ði ¼ 1; 2; 3Þ and tracking errors ei ði ¼ 1; 2; 3Þ will converge to the predesigned
regions in a ¯nite time for the arbitrary initial state:
jsi j < "i ; i ¼ 1; 2; 3; ð35Þ
jei j < 2"i ; i ¼ 1; 2; 3; ð36Þ
where "i are the prede¯ned small positive numbers.
Proof. The ¯rst derivative of the sliding functions can be obtained as follows:
: : :: ::
s ¼ e þ Ne þ Me
: :: :: R R
¼ e þ Ne  MP d  Mhð
Þd  MB signðsÞ
4J0 4J0
: :: : :: :
þ M ½ðM 1 ðe  Ne Þ  AP r þ P d Þ þ AP r 
R
¼ M ðhð
Þd  B signðsÞÞ: ð37Þ
4J0

2550119-9
 L. Chen et al.

The proof will be completed in the following steps based on the switching sequence
of the adaptive law in Eqs. (31) and (32).
Step 1. As 0  t  t ; t ¼ maxft 1 ; t 2 ; t 3 g.
We have jsi ð0Þj > "i =2 ði ¼ 1; 2; 3Þ. Consequently, choose a Lyapunov function V1 as
1 1 ~2 1 ~2 1 ~2
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

V1 ¼ sT s þ k þ k þ k ; ð38Þ
2 2 1 1 2 2 2 2 3 3
where k~i ¼ k
^ i  ki , taking the time derivative of V1 with respect to time, we obtain
: : : :
V 1 ¼ sT s_ þ ~^
1k1k1 þ
~^
2k2k2 þ
~^
3k3k3: ð39Þ
Substituting Eqs. (31) and (37) into Eq. (39), we have
  : : :
: R
V 1 ¼ sT M ðhð
Þd  BsignðsÞÞ þ 1 k~1 k ^ 1 þ 2 k~2 k
^ 2 þ 3 k~3 k
^3
4J0
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

R ^ 1 js1 jÞ þ R M2 ðA2 s2  k ^ 2 js2 jÞ þ R M3 ðA3 s3  k

¼ M1 ðA1 s1  k ^ 3 js3 jÞ
4J0 4J0 4J0
þ 1 ðk ^ 1  k1 Þ1 exp ð 1 js1 jÞjs1 j þ 2 ðk
^ 2  k2 Þ2 exp ð 2 js2 jÞjs2 j
þ ^  k3 Þ3 exp ð 3 js3 jÞjs3 j;
3 ðk 3 ð40Þ
where
i p p
Mi ¼ je_ i j q 1 ; i ¼ 1; 2; 3 ð41Þ
q
and
8 pffiffiffi
< A1 ¼ p
> 2ðsinð
Þðd1 þ d3 Þ  cosð
Þðd2 þ d4 ÞÞ;
ffiffiffi
A ¼  2ðcosð
Þðd1 þ d3 Þ þ sinð
Þðd2 þ d4 ÞÞ; ð42Þ
>
:
2
A3 ¼ 1=ðL þ W Þðd1 þ d2 þ d3  d4 Þ:

For clear representation, 1 1 exp ð 1 js1 jÞjs1 j, 2 2 exp ð 2 js2 jÞjs2 j and 3 3 exp ð 3
js3 jÞ js3 j can be rewritten as ex1 , ex2 and ex3 , respectively. The following inequality can be
obtained:
: RM1 RM2 RM3
V1  ðjA1 jjs1 j  k1 js1 jÞ þ ðjA2 jjs2 j  k2 js2 jÞ þ ðjA3 jjs3 j  k3 js3 jÞ
4J0 4J0 4J0
^ 1  k1 Þ þ ex2 ðk
þ ex1 ðk ^ 2  k2 Þ þ ex3 ðk^ 3  k3 Þ  RM1 ðk ^ 1  k1 Þjs1 j
4J0
RM2 ^ RM3 ^
 ðk 2  k2 Þjs2 j  ðk 3  k3 Þjs3 j
4J0 4J0
RM1 RM2 RM3
¼ js1 jðjA1 j  k1 Þ þ js2 jðjA2 j  k2 Þ þ js jðjA3 j  k3 Þ
4J0 4J0 4J0 3
   
RM1 ^ RM2 ^
þ ex1  js1 j ðk 1  k1 Þ þ ex2  js2 j ðk 2  k2 Þ þ ðex3  js3 j
4J0 4J0

RM3 ^
 ðk 3  k3 Þ: ð43Þ
4J0

2550119-10
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

In order to prove that the sliding mode surface can converge to the speci¯ed region
in ¯nite time under the BFANFTSMC scheme, we de¯ne
RMi
si ¼ ðk  jAi jÞ; i ¼ 1; 2; 3 ð44Þ
4J0 i
and
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

RMi
i ¼ exi  jsi j ; i ¼ 1; 2; 3: ð45Þ
4J0
According to Eqs. (21) and (41), s1 > 0, s2 > 0 and s3 > 0 for e_ i 6¼ 0. More-
over, there will always exist three positive constants i satisfy ex1 > js1 jRM1 /4J0 ,
:
ex2 > js2 jRM2 /4J0 , ex3 > js3 jRM3 /4J0 , thus leading to i > 0 ði ¼ 1; 2; 3Þ. V 1 can be
calculated as
: pffiffiffi js j pffiffiffi js j js j
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

V 1   s1 2 p1ffiffiffi  s2 2 p2ffiffiffi  s3 p3ffiffiffi

2 2 2
qffiffiffiffiffiffiffiffiffiffiffirffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffirffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffirffiffiffiffiffiffi
1 1 ~ 1 2 ~ 3 ~
 1 2 1 jk j  2 2 2 jk j  3 2 1 jk j;
2 1 2 2 3
2 3
!
X 3 rffiffiffiffiffi
3
jsi j X 1
  ’1 pffiffiffi þ i ~
jk i j  ’1 V 12 ; ð46Þ
i¼1 2 i¼1 2

where
 qffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffi
pffiffiffi
1 1 1
’1 ¼ 2 min s1 ; s2 ; s3 ; 1 1 ; 2 2 ; 3 3 : ð47Þ

For ’1 > 0, there exists a positive constant 1 such that

"
1 < ’1 ðtÞ; 8 jsi ðtÞj < i ; i ¼ 1; 2; 3: ð48Þ
2
Consequently, the following inequality can be obtained:
: 1
V 1   1 V 12 : ð49Þ
According to Ref. 40, for t < t, the sliding mode surfaces converge to the interval
½"i =2; "i =2, and the ¯nite time is bounded by
1 1

t  V 1 ð0Þ  V 1 ðt Þ :
2 2 

ð50Þ
0:5 1
For e_ i ¼ 0, substituting Eq. (34) into Eq. (19), we have
:: :: R R
Pr ¼ Pd  BsignðsÞ  hð
Þd: ð51Þ
4J0 4J0
According to Eqs. (22) and (51)
^ 1 signðs1 Þ 3 2
A1  k
R 6
:: ^ 2 signðs2 Þ 7
R
e¼ ðBsignðsÞ  hð
ÞdÞ ¼ 4 A2  k 5: ð52Þ
4J0 4J0
^ 3 signðs3 Þ
A3  k

2550119-11
 L. Chen et al.

For the sliding variables with initial value jsi ð0Þj > "i =2 (i ¼ 1; 2; 3) at t ¼ 0, it can
be observed that k ^ t ðtÞ will keep increasing due to the integral of a positive function of
jsi j, as described in Eq. (30). Then, there exists a time ti such that A1 < k ^ 1 ; A2 < k
^2
::
^
and A3 < k 3 hold. From Eq. (52), we have e i 6¼ 0 for any jsi j > "i =2, which implies
e_ i ¼ 0 is not an attractor in the reaching phase. Combining Eqs. (41) and (44), we
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

can conclude that si will not remain at 0. As a result, si ði ¼ 1; 2; 3Þ can converge to

½"i =2; "i =2 ði ¼ 1; 2; 3Þ under the adaptive law Eq. (30) in the ¯nite time t , re-
spectively. Due to t ¼ maxft 1 ; t 2 ; t 3 g, the sliding mode variables enter the prede¯ned
region successively.
Step 2. As t  t.
We introduce the following virtual variables i ði ¼ 1; 2; 3Þ as

jAi j
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

i ¼ " < "i ; i ¼ 1; 2; 3: ð53Þ

1 þ jAi j i

Then, there exists the following Lyapunov function:

1 T 1 ^2 1 ^2 1 ^2
V2 ¼ s sþ k þ k þ k : ð54Þ
2 2 1 2 2 2 3

Based on Eq. (32), the time derivative of V2 is given as

: : : :
^1k
V 2 ¼ sT s_ þ k ^ 2k
^1 þ k ^ 3k
^2 þ k ^3: ð55Þ

Substituting Eqs. (32) and (37) into Eq. (55), we have

  : : :
: R
V2 ¼s T
M ðhð
Þd  BsignðsÞÞ þ k ^1k^1 þ k ^ 2k
^2 þ k^3k
^3
4J0
RM1 ^ 1 js1 jÞ þ RM2 ðA2 s2  k ^ 2 js2 jÞ þ RM3 ðA3 s3  k
¼ ðA1 s1  k ^ 3 js3 jÞ
4J0 4J0 4J0
 
þ k ^ 1 "1 signðs1 Þ k^ 2 "2 signðs2 Þ k
^ 3 "3 signðs3 Þ s: _ ð56Þ
ð"1  js1 jÞ2 ð"2  js2 jÞ2 ð"3  js3 jÞ2

Similarly, by substituting Eq. (37) into Eq. (56), we have

  : : :
: R
V 2 ¼ sT M ðhð
Þd  BsignðsÞÞ þ k ^1k ^1 þ k ^ 2k
^2 þ k ^3k
^3
4J0
RM1 ^ 1 js1 jÞ þ RM2 ðA2 s2  k ^ 2 js2 jÞ þ RM3 ðA3 s3  k
¼ ðA1 s1  k ^ 3 js3 jÞ
4J0 4J0 4J0
^ 1 "1
k RM1 ^ 2 "2
k RM2
þ ^1Þ þ
ðA1 signðs1 Þ  k ^2Þ
ðA2 signðs2 Þ  k
2 4J 2 4J
ð"1  js1 jÞ 0 ð"2  js2 jÞ 0
^ 3 "3
k RM3
þ ^ 3 Þ:
ðA3 signðs3 Þ  k ð57Þ
ð"3  js3 jÞ2 4J0

2550119-12
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

Taking Ai si and Ai signðsi Þ, we can get the following inequality:

: RM1 ^ RM2 ^ RM3 ^
V2   ðk 1  jA1 jÞjs1 j  ðk 2  jA2 jÞjs2 j  ðk 3  jA3 jÞjs3 j
4J0 4J0 4J0
^ 1 "1
k RM1 ^ ^ 2 "2
k RM2 ^
 ð k  jA jÞ  ðk 2  jA2 jÞ
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

1 1
ð"1  js1 jÞ2 4J0 ð"2  js2 jÞ2 4J0
^ 3 "3
k RM3 ^
 2 4J
ðk 3  jA3 jÞ: ð58Þ
ð"3  js3 jÞ 0

Similar to the previous step, we de¯ne

R
sbi ¼ ^  jAi jÞ;
M ðk i ¼ 1; 2; 3 ð59Þ
4J0 i i
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

and
"i R
bi ¼ 2 4J
^ i  jAi jÞ;
M i ðk i ¼ 1; 2; 3: ð60Þ
ð"i  jsi jÞ 0

According to the expressions of Eqs. (33) and (53), it can be found that kðs ^ iÞ >
^
kð i Þ ¼ Ai for any jsi ðtÞj > i ; i ¼ 1; 2; 3. As a result, we have sb1 ; sb3 ; sb3 ; b1 ; b2 ;
b3 > 0 for e_ i 6¼ 0. Hence, it follows from Eq. (58) that
 
: pffiffiffi js j js j js j 1 ^ 1 ^ 1 ^
V 2  2  sb1 p1ffiffiffi  sb2 p2ffiffiffi  sb3 p3ffiffiffi  b1 pffiffiffi jk 1 j  b2 p ffiffi
ffi j k 2 j  b3 p ffiffi
ffi j k 3 j
2 2 2 2 2 2
!
X 3
js j X 3
1 ^ 1
  ’2 piffiffiffi þ pffiffiffi jk ij  ’2 V 22 ; ð61Þ
i¼1 2 i¼1 2

where
pffiffiffi
’2 ¼ 2 minf sb1 ; sb2 ; sb3 ; b1 ; b2 ; b3 g: ð62Þ

Since ’2 > 0, there exists a positive constant 2 such that

"i
2 < ’2 ðtÞ; 8 i < jsi ðtÞj  ; i ¼ 1; 2; 3: ð63Þ
2
Then, from Eq. (61), we have
: 1
V 2   2 V 22 : ð64Þ

In a ¯nite time period tm , si can converge to the corresponding regions ½ i ; i

from any initial values sðt i Þ, which are bounded by
1 1
V 2 ðt Þ  V 22 ðt þ tm Þ
tm  2 : ð65Þ
0:5 2
It can be inferred from the de¯nitions bi ; sbi and Mi that bi ; sbi will not remain
at 0. Therefore, the sliding variables will remain inside the corresponding regions

2550119-13
 L. Chen et al.

jsi ðtÞj  i < "i for all t > t þ tm . Equation (23) can be rewritten as
!
a si p
ei þ i sigðei Þ b þ i  p sigðe_ i Þ q ¼ 0: ð66Þ
sigðe_ i Þ q

Then, letting je_ i j > ð"i =i Þq=p implies i  si =sigðe_ i Þp=q > 0 since jsi j  "i . Under
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

the conditions speci¯ed above, Eq. (66) maintains the same property of ¯nite-time
stability as that in Ref. 41, which implies, in reverse, that the velocity of the tracking
error converges to the region
je_ i j  ð"i =i Þq=p ð67Þ
in a ¯nite time. Thus, we can conclude that
a "i
ei þ i sigðei Þ b  jsi j þ i  2"i : ð68Þ
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

i

Since sign(ei Þ = sign(sig(ei Þa=b Þ, Eq. (68) can be expressed as

"
jei j þ ji sigðei Þa=b j  jsi j þ i i  2"i : ð69Þ
i
Therefore, we can deduce that the tracking error converges to the region jei j  2"i
in a ¯nite time as well. This completes the proof of Theorem 3.1.

Remark 3. In comparison to the conventional CSMC and NTSM, the proposed

BFANFTSMC scheme can adapt the control gain based on the tracking error, which
e®ectively mitigates the severe chattering problem. If the robot's initial tracking
error is relatively small, the ¯xed-gain control mode can respond more quickly.
However, if the initial error is signi¯cant, the proposed method's control gain can be
progressively increased via integration, resulting in faster response time. Addition-
ally, the greater the tracking error at the initiation of robot operation, the more
prominent the advantages of the proposed control become evident.

3.3. Control parameter selection

In practical implementations, the MWOMR's ability to successfully accomplish
trajectory tracking tasks is also contingent upon the appropriate parameter choices.

(1) Selections of i , i , a, b, p, q: The parameters i , i , a, b, p and q play a signi¯cant

role in determining the convergence rate of the sliding mode function. A larger
value of i and i or a smaller ratio of a=b and p=q results in a faster convergence
rate. However, this may exacerbate the chattering issue. To strike an optimal
balance between the convergence rate and chattering suppression, we set i ¼ 5,
i ¼ 1, a ¼ 7, b ¼ 5, p ¼ 5, and q ¼ 3.
(2) Selections of i , i : The parameter i and i de¯ne the rate of sliding mode
surface converging from the initial state to the design interval. However, when

2550119-14
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

their values are too large, the control gain will easily exceed the interference
boundary, resulting in chattering problem. In our experiment, chose
1 ¼ 2 ¼ 0:3, 3 ¼ 0:0001, 1 ¼ 2 ¼ 0:2, 3 ¼ 0:001.
(3) Selections of "i : A smaller convergence boundary value, determined by the pa-
rameter, results in a reduced tracking error in the system. However, this also
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

leads to a higher control gain, which may potentially saturate the controller. In
experiments, we discovered that setting "1;2 to 0.015 and "3 to 0.07 strikes an
ideal balance.

3.4. MBF to tackle actuator saturation

Due to the use of DC geared motors in this platform, the motor speed is directly
proportional to the PWM duty cycle. However, there is a maximum limit for the
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

duty cycle, which causes actuator saturation in this platform. To address this issue,
we further improve the adaptive rate as follows:
 Z t 
^ 
k i ¼ min k i ; i exp ð i jsi Þjsi jdt ; for 0  t  t i ; ð70Þ
0

^ i ¼ minfki ; fb ðsi Þg;

k for t > t i ; ð71Þ
where i ¼ 1; 2; 3 and

jsi j
fbi ðsÞ ¼ ; for si 2 ð"i ; "i Þ; i ¼ 1; 2; 3: ð72Þ
"i  jsi j

As shown in Fig. 2, k1 , k2 and k3 represent the values of k ^ 1; k

^ 2 and k^ 3 , respec-
tively, when the actuator is saturated. It is straightforward to observe that as si
increases to the point where ki ¼ ki , the value of jsi j, denoted as  i , can be expressed
as  i ¼ "i ðki =ki þ 1Þ.

Fig. 2. (Color online) The illustration of MBF.

2550119-15
 L. Chen et al.

To demonstrate the e®ectiveness of the proposed MBFANFTSMC, we make the

following assumption:
Assumption 1. The amplitude of disturbance satis¯es that jA1 j < k1 , jA2 j < k2
and jA3 j < k3 . In addition, jui j  jueqi j þ jubfi j  umaxi for any t  0, where umaxi is
the maximum output of the actuator, and B ¼ diagfk1 ; k2 ; k3 g.
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

First, under the adaptive law Eq. (70), the sliding mode variables si will converge
to the corresponding regions ½"i =2; "i =2 in ¯nite time. Due to integral items, k ^i
^ 
continue to increase until si ¼ 0. Note that k i ðtÞ are bounded by k i . According to
Eq. (21) and Assumption 1, there exist three constants ki that satisfy k ^ i  ki  ki .
As a result, it is also guaranteed that si converges to the corresponding regions
½"i =2; "i =2 in ¯nite time based on Eq. (70).
Then, we will establish that ¯nite-time convergence is assured with the adaptive
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

law Eq. (71).

Proof. Case 1. jsi j <  i for i ¼ 1; 2; 3.
According to Fig. 2 and Eq. (72), we obtain

jAi j k
i ¼ "i   i "i ¼  i < "i ; i ¼ 1; 2; 3: ð73Þ
1 þ jAi j ki þ 1

Due to i <  i , si will converge to ½ i ; i  in ¯nite time under the adaptive law
Eq. (71) whose proof steps refer to Step 2 of Theorem 3.1.
Case 2. jsi j   i for i ¼ 1; 2; 3.
^ i ¼ ki , the Lyapunov equation is chosen as V3 ¼ 0:5s2 , we have
As k
: :RM1 RM2 RM3
V 3 ¼ ss  ðjA1 js1  k1 js1 jÞ þ ðjA2 js2  k2 js2 jÞ þ ðjA3 js3  k3 js3 jÞ
4J0 4J0 4J0
RM1  RM2  RM3 
¼ ðk 1 js1 j  jA1 js1 Þ  ðk 2 js2 j  jA2 js2 Þ  ðk 3 js3 j  jA3 js3 Þ:
4J0 4J0 4J0
ð74Þ

We can deduce that si complies with the ¯nite-time stability condition stated in
Assumption 1.
Remark 4. The MBFANFTSMC scheme resolves actuator saturation issues
while retaining most of the advantages of the BFANFTSMC scheme. However,
due to the maximum gain limitation, the MBFANFTSMC scheme can only handle
bounded or short-time disturbances that exceed the limit at low frequencies, in
contrast to the BFANFTSMC. When the disturbances exceed the limit for a brief
period, the MBFANFTSMC requires more time to correct the trajectory than the
BFANFTSMC. Nonetheless, given its ability to resolve actuator saturation, this
trade-o® is justi¯able, and in practical applications, the gain limitation of the
MBFANFTSMC results in smaller °uctuations in tracking error.

2550119-16
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

4. Experimental Studies
To validate the tracking performance of the proposed BFANFTSMC and assess the
e®ectiveness of the MBFANFTSMC scheme in addressing input saturation, we
conduct two experimental scenarios involving lemniscate and circular movements on
an experimental platform.
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

4.1. Experimental con¯gurations

Figure 3 depicts the MWOMR experimental platform. This robotic platform consists
of four Mecanum wheels driven by four DC motors, a laser radar, an STM32, and a
Jetson TX1 utilizing ROS. To measure the (X{Y)'s velocity and rotation informa-
tion, the platform is equipped with four giant gagnetoresistive encoders and an IMU
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

sensor (MPU6050). From Fig. 4, the distance between the car and the wall is mea-
sured using the laser radar, which serves as the coordinates for the car's motion. The
values of the robot model parameters are presented in Table 1.

Fig. 3. (Color online) Experimental platform.

2550119-17
 L. Chen et al.

Table 1. Parameters of MWOMR.

Parameter Symbol Value

Radius of wheel R 0.03 m

Length of platform 2L 0.18 m
Breadth of platform 2W 0.20 m
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Nominal rotational inertia J0 0.095 km2

Nominal viscous friction b0 0.09 Nms/rad
Nominal motor constant Km 0.81 Nm/V
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

Fig. 4. (Color online) Experimental site.

The NFTSMC in Eq. (23) (i.e., the BFANFTSMC without barrier function),
NTSMC in Ref. 42, and the CSMC were selected as control groups for comparative
experiments.

4.2. Experiment case 1: Lemniscate movement with starting

point outside trajectory
In reality, mobile robots are generally located in relatively narrow spaces, such as
workshops and warehouses. Under these circumstances, the excellent omnidirec-
tional mobility of MWOMR is very crucial. In this scenario, the lemniscate tracking
trajectory has been selected as the test path, a classical route frequently employed to
evaluate the omnidirectional mobility capabilities of mobile robots.
Additionally, to assess the adaptability of the control system, we intentionally
place the starting point outside the desired trajectory, speci¯cally at coordinates.
The experimental results for the three control schemes are presented in Figs. 5{7.
From Fig. 5, it is clear that all the controllers successfully guide the MWOMR to
accomplish the path-following task. It can be observed that the tracking performance
of the BFANFTSMC is best, with both the magnitude and frequency of deviations
from the trajectory being less than the other three control methods. Figure 6 shows
the tracking errors of all four control schemes. It is evident that the BFANFTSMC

2550119-18
 Adaptive fast terminal sliding mode control on MWOMR using barrier function
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

Fig. 5. (Color online) Trajectory tracking result of the lemniscate trajectory.

(a)

(b)

(c)

Fig. 6. (Color online) Tracking errors in lemniscate trajectory tracking. (a) Tracking errors of X-axis
displacement. (b) Tracking errors of Y-axis displacement. (c) Tracking errors of orientation.

2550119-19
 L. Chen et al.
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(a)

(b)
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

(c)

(d)

Fig. 7. (Color online) Control inputs in lemniscate trajectory tracking. (a) Control law u1 . (b) Control
law u2 . (c) Control law u3 . (d) Control law u4 .

achieves the smallest steady-state errors. When the tracking error is large, the
control output increases rapidly due to the characteristics of the barrier function,
resulting in a fast convergence of the error and thus a smaller peak error. In contrast,
traditional control methods with ¯xed gains struggle to stabilize the system when the
error exceeds the control gain. Therefore, it is necessary to predict the peak error.
However, predicting the peak error is often challenging, and setting the control gain
too small would result in system instability when facing signi¯cant disturbances. On
the other hand, setting the control gain too large would cause severe vehicle oscil-
lations. The BFANFTSMC's peak X-axis error is 11.3 mm, and its peak Y-axis error
is 13.4 mm, signi¯cantly smaller than the NFTSMC (22.1 mm, 24.2 mm), NTSMC
(18.4 mm, 27.1 mm) and the CSMC (19.5 mm, 27.5 mm), it illustrates that the
tracking performance of the proposed control surpasses that of the NFTSMC,
NTSMC and CSMC schemes. This superiority is evident in notably smaller peak
errors and steady-state errors. In the aspect of angular tracking, all four control
methods successfully maintain the error within a range of 0{5. The performance of
the proposed controller is nearly optimal, yet the di®erences with the other three
controllers are not particularly signi¯cant. This is largely related to the precision of

2550119-20
 Adaptive fast terminal sliding mode control on MWOMR using barrier function
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

Fig. 8. (Color online) Trajectory tracking result of the circular trajectory.

the sensors and indicates that all four controllers have almost achieved the optimal
angular control that this platform is capable of reaching. In Fig. 7, we can readily
observe that the proposed control signi¯cantly mitigates chattering e®ects in the
control input, in contrast to the other three schemes. This will greatly enhance the
applicability of the proposed control for practical applications.

4.3. Experiment case 2: Circular movement with actuator saturation

In order to verify the proposed MBFANFTSMC's performance for preventing ac-
tuator saturation, we deployed the MBF controller utilizing identical control para-
meters to those used in the BFANFTSMC. The umaxi is 7, and we set k1 ¼ 0:45,
k2 ¼ 0:4, k3 ¼ 0:3 for satisfying the inequality condition as stated in Assumption 1.
Figures 8 and 9 show the experimental results of the tracking performance. Both
controllers can successfully accomplish the tracking task.
It can be observed from the graph that due to the limited value of the adaptive
control gain, the steady tracking error of the MBFANFTSMC is a bit larger than
that of the BFANFTSMC. Furthermore, except for the x-axis, the MBFANFTSMC
requires a slightly longer setting time for Y-axis and angular tracking than the
BFANFTSMC.
Figure 10 shows the adaption control gain k ^ of the BFANFTSMC and the
MBFANFTSMC. In combination with the tracking error in Fig. 9, it can be seen that
the setting time's di®erence value is because the adaptive gain in the approach
process is limited to a ¯xed value. Since the initial error of the X-axis is not very big,
its adaptive gain is not limited. Moreover, Fig. 11 depicts the control inputs in a
circular movement with actuator saturation. It is evident that the MBFANFTSMC's
results in control input amplitudes are signi¯cantly smaller than those seen with the

2550119-21
 L. Chen et al.
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(a)
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

(b)

(c)

Fig. 9. (Color online) Tracking errors in circular movement with actuator saturation. (a) Tracking errors
of X-axis displacement. (b) Tracking errors of Y-axis displacement. (c) Tracking errors of orientation.

(a)

(b)

(c)

^ (a) The value of k

Fig. 10. (Color online) Adaption control gain k. ^ 1 . (b) The value of k
^ 2 . (c) The value of k
^3.

2550119-22
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

(a)
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(b)
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

(c)

(d)

Fig. 11. (Color online) Control inputs in lemniscate trajectory tracking with actuator saturation.
(a) Control law u1 . (b) Control law u2 . (c) Control law u3 . (d) Control law u4 .

BFANFTSMC for all four control inputs. This strongly demonstrates that the
MBFANFTSMC scheme can e®ectively prevent actuator saturation.

4.4. Data comparisons

According to Tables 2 and 3, it is evident that all tracking errors are less than its
requirements (2"1;2 ¼ 30 mm; 2"3 ¼ 0:14 radÞ, which con¯rms the theory that the
BFANFTSMC can restrict the tracking errors within a ¯xed region.
For further validation of the e®ectiveness and superiority of the proposed control
law, we utilize the evaluation indices of maximum errors (MAXE) and root mean
square errors (RMSE) for a comprehensive comparison. These indices are de¯ned as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N
uX e2 ðiÞ
RMSE ¼ t ; i ¼ 1; 2; 3; ð75Þ
i¼0
N

MAXE ¼ maxðjeðiÞjÞ; i ¼ 1; 2; 3; ð76Þ

where eðiÞ is the corresponding tracking error and N is the number of samples.
Table 2 shows the data comparisons of the control performance at lemniscate

2550119-23
 L. Chen et al.

Table 2. Control performance comparisons at lemniscate

movement.

Error Control method RMSE MAX

X-axis error CSMC 83.3443 19.5

NTSMC 80.3713 18.4
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

NFTSMC 75.0261 22.1

BFANFTSMC 64.3143 11.3
Y-axis error CSMC 29.8249 27.5
NTSMC 24.7306 27.1
NFTSM 22.6643 24.2
BFANFTSMC 20.8301 13.4
Orientation angle error CSMC 10.6019 4.22
NTSMC 9.9900 4.05
NFTSM 7.7377 4.95
BFANFTSMC 7.7185 3.88
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

Table 3. Control performance comparisons at circular movement.

Error Control method RMSE MAX

X-axis error BFANFTSMC 43.3134 17.5

MBFANFTSMC 43.6792 23.1
Y-axis error BFANFTSMC 116.699 18.9
MBFANFTSMC 121.771 19.1
Orientation angle error BFANFTSMC 9.6591 2.87
MBFANFTSMC 20.9871 3.71

movement with disturbance for the three controllers. We can see that the X-axis
MAXE and RMSE values of the proposed controller are the smallest.
Compared with the CSMC, NTSMC and NFTSMC, the proposed control
achieves improvements of 22.83%, 19.98% and 14.28% for RMSE values, respec-
tively. The MAXE and RMSE values for both the Y-axis and orientation angle under
the BFANFTSMC are noticeably smaller compared to the other two controllers.
Table 3 lists the data comparisons at the circular movement scenario. The data
indicate that the performance of BFANFTSMC is only marginally better than that
of MBFANFTSMC, with the di®erences not being particularly signi¯cant. This also
con¯rms the e®ectiveness of MBFANFTSMC in trajectory tracking performance,
indicating that its performance does not become signi¯cantly worse due to the lim-
itations imposed on the adaptation control gain.

5. Conclusion
In this paper, we have designed a BFANFTSMC scheme for the MWOMR's tra-
jectory tracking task in the presence of unknown disturbances. Initially, we intro-
duced a NFTSMC scheme to guarantee both ¯nite-time convergence and robustness
for the robot's trajectory tracking. Next, using the barrier function, the controller

2550119-24
 Adaptive fast terminal sliding mode control on MWOMR using barrier function

can adjust the output amplitude based on the tracking error and prevent excessive
generation of control inputs when the tracking errors are small, resulting in further
chattering reduction. Additionally, in response to the issue of actuator saturation in
practical applications, we have introduced the MBFANFTSMC by constraining the
maximum adaptive control gain. Both the BFANFTSMC and MBFANFTSMC
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

schemes have been theoretically proven to ensure that the tracking error converges to
a prede¯ned region near zero in ¯nite time, in the sense of Lyapunov. The real-time
experimental results clearly illustrate the superior control precision and robustness
of the BFANFTSMC and the strong capability of the MBFANFTSMC to handle
actuator saturation. The future work is to focus on fault tolerant control and multi-
agent control of MWOMR systems under both actuator and sensor faults.
Currently, due to limitations in our experimental platform, the position detection
system for our MWOMR relies on a combination of radar and odometry information.
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

However, this approach is subject to cumulative positional errors, resulting in a

gradually increasing discrepancy between the theoretical and actual positions. To
address this issue, we plan to incorporate a motion capture system in future studies to
achieve more precise localization. Our next research objective is to investigate
MWOMR control strategies based on prede¯ned-time SMC and disturbance observer.
Concurrently, we aim to leverage the motion capture system to explore formation
control and task allocation for MWOMRs. These advancements are expected to ef-
fectively enhance the precision and functionalities of our robotic systems.

Acknowledgments
This work was supported by the National Natural Science Foundation of China
under Grant No. 61771178.

ORCID
Long Chen https://orcid.org/0009-0006-2892-0716
Zeyuan Lu https://orcid.org/0009-0000-5984-1929
Bin Yan https://orcid.org/0009-0004-5433-3940
Peipei Jin https://orcid.org/0000-0003-3302-188X
Guangyi Wang https://orcid.org/0000-0002-6777-8234

References
1. J. H. Choi, K. Nam and S. Oh, IEEE Trans. Autom. Sci. Eng. 19 (2022) 3210.
2. N. Lashkari, M. Biglarbegian and S. X. Yang, IEEE Trans. Autom. Sci. Eng. 17 (2020)
1759.
3. M. Ge, H. Xu and Q. Song, Nonlinear Dyn. 17 (2023) 387.
4. W. H. Li, Z. C. Li, Y. Q. Liu, L. Ding, J. F. Wang, H. B. Gao and Z. Q. Deng, Mech. Syst.
Signal Process. 133 (2019) 234.
5. W. H. Chen, C. P. Chen, J. S. Tsai, J. K. Yang and P. C. Lin, Mech. Mach. Theory 68
(2013) 35.

2550119-25
 L. Chen et al.

6. S. Toyoshima, M. Tanaka and F. Matsuno, IEEE Trans. Autom. Sci. Eng. 11 (2014) 139.
7. T. Terakawa, M. Komori, K. Matsuda and S. Mikami, IEEE-ASME Trans. Mechatron.
23 (2018) 1716.
8. J. T. Huang, T. V. Hung and M. L. Tseng, IEEE Trans. Control Syst. Technol. 23 (2015)
1986.
9. A. Gfrerrer, Comput. Aided Geom. Des. 25 (2008) 784.
by ZHEJIANG UNIVERSITY on 03/19/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

10. G. Bayar and S. Ozturk, Mechatronics 72 (2020) 1452.

11. J. Kim, S. Woo, J. Kim, J. Do, S. Kim and S. Bae, Int. J. Precis. Eng. Manuf. 26 (2012) 379.
12. X. Lu, X. Zhang, G. Zhang, J. Fan and S. Jia, ISA Trans. 86 (2019) 201.
13. L. Ovalle, H. Ríos, M. Llama, V. Santibañez and A. Dzul, Control Eng. Pract. 85 (2019) 50.
14. Z. Yuan, Y. Tian, Y. Yin, S. Wang, J. Liu and L. Wu, IET Control Theory 14 (2020) 415.
15. H. Xiao, D. Yu and C. P. Chen, Inf. Sci. 590 (2022) 45.
16. Z. Sun, S. Hu, D. He, W. Zhu, H. Xie and J. Zheng, Comput. Electr. Eng. 96 (2021) 516.
17. H. Huang and S. Lin, IEEE Trans Ind. Inf. 15 (2019) 1062.
18. P. S. Yadav, V. Agrawal and J. C. Mohanta, Proceedings 56 (2022) 623.
Mod. Phys. Lett. B Downloaded from www.worldscientific.com

19. X. Wu, K. Xu, M. Lei and X. He, IEEE Trans. Autom. Sci. Eng. 17 (2020) 2182.
20. T. Huang and T. Li, Nonlinear Dyn. 111 (2023) 10183.
21. J. Zhang, H. Wang, M. Ma, M. Yu, A. Yazdani and L. Chen, IEEE Trans. Veh. Technol.
69 (2020) 14713.
22. A. Safa, R. Abdolmalaki, S. Sha¯ee and B. Sadeghi, ISA Trans. 77 (2018) 122.
23. J. Zhang, H. Wang, Z. Cao, J. Zheng, M. Yu, A. Yazdani and F. Shahnia, Neural Comput.
Appl. 32 (2019) 14447.
24. H. Xie, J. Zheng, Z. Sun, H. Wang and R. F. Chai, Nonlinear Dyn. 110 (2022) 1437.
25. Y. Wang, S. Li, D. Wang, F. Ju, B. Chen and H. Wu, IEEE Trans. Ind. Electron. 68
(2021) 2356.
26. J. Xu, IEEE Trans. Autom. Control 64 (2018) 3046.
27. S. Ding, Q. Hou and H. Wang, IEEE Trans. Energy Convers. 23 (2022) 1257.
28. L. Chen, B. Yan, H. Wang, K. Shao, E. Kurniawan and G. Wang, Control Eng. Pract.
121 (2022) 1050.
29. L. Chen, J. Liu, H. Wang, Y. Hu, X. Zheng, M. Ye and J. Zhang, Nonlinear Dyn. 104 (2021)
2291.
30. R. D. Xi, X. Xiao, T.-N. Ma and Z.-X. Yang, IEEE Robot. Autom. Lett. 7 (2022) 6139.
31. H. Han, X. Wu and J. Qiao, IEEE Trans. Syst. Man Cybern. Syst. 50 (2018) 4415.
32. H. Obeid, L. Fridman, S. Laghrouche, M. Harmouche and M. A. Golkani, Int. J. Control
91 (2018) 1345.
33. C. Dong, Y. Liu and Q. Wang, ISA Trans. 96 (2020) 163.
34. K. P. Tee, S. S. Ge and E. H. Tay, Automatica 45 (2009) 918.
35. X. Ji, S. Ding, B. Cui, C. Ding and X. Wei, IEEE Trans. Circuits Syst. II Express Briefs
70 (2023) 3024.
36. Y. Z. Sun, J. Y. Kuang, Y. B. Gao, W. L. Chen, J. H. Wang, J. X. Liu and L. G. Wu,
Nonlinear Dyn. 111 (2023) 14077.
37. T. D. Li, J. X. Zhang, S. B. Li, P. Zhou and D. C. Lv, Nonlinear Dyn. 111 (2023) 16187.
38. N. Ali, I. Tawiah and W. Zhang, Ocean Eng. 218 (2020) 1081.
39. K. Shao, J. Zheng, H. Wang, X. Wang, R. Lu and Z. Man, IEEE Trans. Ind. Inf. 179
(2021) 7479.
40. K. Shao, J. Zheng, K. Huang, H. Wang, Z. Man and M. Fu, IEEE Trans. Ind. Electron.
67 (2020) 6659.
41. H. Ghorbani, R. Vatankhah and M. Farid, Aerosp. Sci. Technol. 119 (2021) 1071.
42. Z. Sun, H. Xie, J. Zheng, Z. Man and D. He, Mech. Syst. Signal Process. 147 (2021) 1071.

2550119-26