IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 53, NO.

4, APRIL 2023 2363

Fixed-Time and Fault-Tolerant Path-Following

Control for Autonomous Vehicles With Unknown
Parameters Subject to Prescribed Performance
Zhongchao Liang , Zhongnan Wang , Jing Zhao , Pak Kin Wong , Zhixin Yang , Member, IEEE,
and Zhengtao Ding , Senior Member, IEEE

Abstract—With the consideration of actuator faults, includ- pressures [1]. As a key technology of AGVs, path-following
ing the unknown steering mechanism misalignments and motor control is the primary target to reasonably operate the vehicle
traction losses, this article presents a fixed-time control pro- steering system, and to eliminate the deviation between the
tocol to follow reference paths and velocities for autonomous
ground vehicles (AGVs) with preset performance constraints. actual path and the reference path [2]. In this sense, the path-
To provide sufficient large boundaries for the initial states, the following control strategy with strong robustness and rapid
hyperbolic tangent function is employed to predefine the con- convergence is of great significance to guarantee the track-
straints with respect to the path-following and velocity control ing accuracy and dynamic stability of the vehicles during
performance. Based on the homeomorphic mapping and bar- autonomous driving [3].
rier Lyapunov theorem, the fixed-time prescribed performance
control (PPC) objective-integrated fault-tolerant scheme can be Up to now, a series of path-following control protocols have
achieved for the controlled AGV. In comparison to three different been developed for AGVs to track reference trajectories, such
fixed-time controllers without the fault-tolerant or PPC scheme, as model-predictive control (MPC) [4], sliding mode control
the hardware-in-the-loop (HIL) test results demonstrate that the (SMC) [5], and H-infinity control [6]. In fact, these con-
proposed control protocol can always provide superior control trol schemes mostly focus on the steady-state errors, whereas
performance for the AGV under various maneuvering conditions.
the control performance during the whole tracking process
Index Terms—Fault-tolerant control, fixed-time convergence, is less concerned. To this end, the prescribed performance
path-following control, prescribed performance control (PPC). control (PPC) approaches have been proposed to satisfy the
preset performance requirements and the transient error con-
straints [7]. In particular, the transient performance of the
I. I NTRODUCTION controlled system can be limited within the predefined resid-
ITH the development of sensing, data processing and ual set through using homeomorphic mapping and Lyapunov
W telecommunication technologies, intensive researches
have focused on the autonomous ground vehicles (AGVs) due
theorem [8]. Resorting to the PPC techniques with bound-
ary functions, the control performance can be effectively
to great potentials of accident reductions, as the autonomous improved [9], [10], [11].
driving system can effectively alleviate driver works and traffic As a widely applicable control technique, the SMC has been
utilized in path-following systems to cope with perturbations
Manuscript received 1 August 2022; accepted 28 September 2022. Date of and uncertainties [12], [13]. However, we notice that the set-
publication 14 October 2022; date of current version 17 March 2023. This tling time of the conventional SMC-based schemes are related
work was supported in part by the National Natural Science Foundation of
China under Grant 51975109; in part by the Fundamental Research Funds to the initial states of the controlled system. Consequently,
for the Central Universities under Grant N2103018; in part by the Joint the fixed-time control method has been proposed to drive the
Project of Natural Science Foundation of Liaoning Province of China under tracking errors into a small neighborhood of zero in a fixed
Grant 2021-KF-11-02; in part by the University of Macau under Grant
MYRG2020-00045-FST; in part by the Science and Technology Development settling time independent of initial states [14], [15], [16], and
Fund, Macau, under Grant 0018/2019/AKP and Grant SKL-IOTSC(UM)- can provide a systematic framework in the controller design.
2021-2023; and in part by the Guangdong Science and Technology As a matter of fact, the fixed-time control schemes integrated
Department under Grant 2018B030324002 and Grant 2020B1515130001.
This article was recommended by Associate Editor S. C. Subramanian. with PPC have been usually used in quadrotor unmanned aerial
(Corresponding author: Jing Zhao). vehicles (QUAVs) [17], marine surface vessels (MSVs) [18]
Zhongchao Liang and Zhongnan Wang are with the School of Mechanical and autonomous underwater vehicles (AUVs) [19]. In terms
Engineering and Automation, Northeastern University, Shenyang 110819,
China (e-mail: [email protected]; [email protected]). of the AGVs, a different dynamic system with the steering
Jing Zhao, Pak Kin Wong, and Zhixin Yang are with the Department of mechanism will introduce more theoretical and practical issues
Electromechanical Engineering, University of Macau, Macau, China (e-mail: to achieve fixed-time and PPC objectives. While the system
[email protected]; [email protected]; [email protected]).
Zhengtao Ding is with the Department of Electrical and Electronic parameters are not available, the adaptive algorithm can be
Engineering, University of Manchester, M13 9PL Manchester, U.K. employed to estimate the unknown parameters [20], [21], [22].
(e-mail: [email protected]). Moreover, the actuator faults are also of significance,
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TSMC.2022.3211624. whereas less considered in fixed-time control and PPC
Digital Object Identifier 10.1109/TSMC.2022.3211624 methods.
2168-2216 
c 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
2364 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 53, NO. 4, APRIL 2023

In practical applications, it is difficult to establish an accu-

rate vehicle dynamic model including all nonlinearities and
uncertainties [23]. Meanwhile, the malfunctioned actuators
may lead to serious accidents of the vehicle. Therefore,
fault-tolerant control is of necessity to be considered in
the controller design for the system reliability [24]. In the
autonomous driving system, the steering angle is calculated
by the path-following controller, and is performed by the
mechanical structure of the steering system to track the desired
paths [25]. Faults caused by actuator failures or mechanism
errors are inevitable during the vehicle operation [26] and,
thereby, the fault-tolerant control has been enticed consid-
erable attentions to maintain the system performance while
the faulty status occurs [27]. In terms of these system
uncertainties in the actuators, the adaptive control tech-
Fig. 1. Dynamic model for the vehicle.
nology can also be used to compensate for the actuator
faults with completely unknown efficiencies, and realizes
self-regulation [28].
Since the AGVs always run on the highway or civil is integrated to estimate all the vehicle parameters
road with high safety requirements, the preset constraints for and to achieve the control objectives within the
the path-following errors can predict the limitations of the fixed time.
future states for the vehicle system. Meanwhile, the fixed- The structure of this article is organized as the following.
time control approach can be used to provide fixed settling Section II presents the vehicle dynamic model, path-following
time independent of initial states. Therefore, with integration model, and problem formulation. The path-following and
of prescribed performance and fixed-time control, the prior velocity controllers are designed in Section III. In Section IV,
knowledge boundaries for the future path-following error and the hardware-in-the-loop (HIL) tests are conducted to ver-
settling time can effectively constrain the system status and in ify the effectiveness of the proposed controllers. Finally, the
turn enhance the safety performance for the AGVs in practi- conclusions are drawn in Section V.
cal applications. Besides, the actuator faults of the steering and
traction systems should be considered in the controller design. II. M ODELING AND P ROBLEM F ORMULATION
To this end, by combining the fixed-time control, PPC and
In this section, the vehicle dynamics and the path-following
fault-tolerant control, the AGV system can acquire the supe-
model will be established with the consideration of actuator
rior path-following performance and improve the reliability of
faults and system uncertainties, and the problem formulation
the control system.
will be discussed.
Motivated by the above observations, the PPC can be uti-
lized to predefine the transient and steady-state performance
of the path-following and traction systems during the whole A. Vehicle Dynamic Model
control process. The main contributions and features of this As illustrated in Fig. 1, the vehicle dynamics can be
article are summarized as follows. described as follows [29]:
1) A novel control protocol is designed for the AGV ⎧  
⎪
⎪ m v̇x − vy γ  = Ffax cos δf + Frax − Ffay sin δf
to follow the desired paths and velocities subject to ⎨ 
m v̇y − vx γ = Frax sin δf + Fray + Fray cos δf
the preset performance constraints. In contrast to the (1)
⎪
⎪ Iz γ̇ = lf Ffax sin δf + lf Ffay cos δf − lr Ffay + dFrbx
existing path-following control schemes [1], [2], [3], ⎩
−dFfby sin δf
the steering faults caused by the unknown steering
mechanism misalignments and the tire force nonlin- where Ffax , Frax , Ffay , Fray , Ffby , and Frby are the tire forces
earities, as well as the traction faults caused by the with Ffax = Fflx + Ffrx , Frax = Frlx + Frrx , Ffay = Ffly + Ffry ,
unknown motor efficiency and the tire slip, are con- Fray = Frly + Frry , Ffby = Ffry − Ffly , and Frby = Frry − Frly ,
sidered in the vehicle dynamics. In the meantime, the in which the subscripts represent the wheel identities, i.e., fl
fixed-time PPC objectives can always be guaranteed, is front left, fr is front right, rl is rear left, and rr is rear right;
even if the unknown actuator faults exist in the AGV vx and vy denote the longitudinal and lateral velocities of the
system. vehicle, respectively; β stands for the sideslip angle of the
2) Unlike most of the existing PPC approaches with expo- vehicle, and β = arctan(vy /vx ); ψ and γ stand for the yaw
nential decay boundaries [8], [9], [10], [19], [20], [31], angle and yaw rate of the vehicle, respectively; δf stands for
the application of the hyperbolic tangent function for the the steering angle of the front wheels with δf = cs u, in which
preset constraint can effectively enlarge the boundaries cs is the steering ratio, and u is the steering column input of
for the initial states, and the convergence time of the the vehicle; lf and lr represent the distances from the front
enlarged boundaries is only delayed by an additional and rear axles to the center of the gravity (CG), respectively;
time constant. Meanwhile, the adaptive algorithm d represents the distance from the CG to the left/right wheel;

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
LIANG et al.: FIXED-TIME AND FAULT-TOLERANT PATH-FOLLOWING CONTROL FOR AUTONOMOUS VEHICLES 2365

and m and Iz represent the vehicle mass and the yaw moment into (1), the lateral dynamics of the vehicle can be given as
of inertia, respectively. ⎧
⎪
⎪ β̇ = (a11 + a11 )β + (a12 + a12 )γ
Furthermore, in order to ensure the vehicle operation at the ⎨
+(b1 + b1 )(cs + cs )u + dβ
target speed, the traction motion of the wheel is modeled with (6)
⎪
⎪ γ̇ = (a21 + a21 )β + (a22 + a22 )γ
the consideration of uncertain slips and traction losses as [30] ⎩
+(b2 + b2 )(cs + cs )u + dγ
Fix = R−1 (1 − ηi )Ti − Fiz (as si + bs ) + R−1 dF where dβ and dγ include the perturbations, and the system
⎧ ⎫
⎪
⎨ ⎪
⎬ parameters are calculated by
−1
  
=R Ti − ηi Ti + RFiz (as si + bs ) − dF (2) Cf + Cr Cf + Cr v̇x
⎪
⎩ ⎪
⎭ a11 = − , a11 = − −
Ti mvx mvx vx
Cr lr − Cf lf lr Cr − lf Cf
where Ti represents the torque of motor i; R represents the a12 = 2
− 1, a12 =
mvx mv2x
effective radius of wheel; si is the slip ratio of wheel i, and lr Cr − lf Cf lr Cr − lf Cf
i ∈ G = {fl, rl, fr, rr}; as and bs are the fitting coefficients; a21 = , a21 =
Iz I
ηi is the traction loss coefficient of motor i; dF represents  2z 
the unmodeled traction terms in wheel dynamics; and Ti lf2 Cf + lr2 Cr lf Cf + lr2 Cr
a22 = − , a22 = −
represents the total traction loss of wheel i. Iz vx Iz vx
According to the vehicle dynamics (1) and the wheel trac- Cf Cf
tion (2), the longitudinal dynamics of the vehicle can be b1 = , b1 =
mvx mvx
expressed as
lf Cf lf Cf
b2 = , b2 = .
v̇x = (mR)−1 (1 − ηT )Ta + dv (3) Iz Iz

where Ta is the total torque of wheels; dv includes the uncer- B. Path-Following Model
−1 −1
tainty terms, and dv = (cos δf − 1)m Ffax − m Ffay sin δf + The path-following errors mainly include the distance off-
vy γ ; ηT = (1/Ta ) i∈G Ti is the total torque loss coefficient set ed from the CG to the orthogonal projection point on the
of the traction system. desired path, and the heading direction offset eψ between the
Based on the vehicle kinematics, the sideslip angles of tires actual velocity direction ψv and the tangent direction of the
can be expressed as orthogonal projection point ψs . If ed and eψ can be driven to
 zero, then the path-following control target can be achieved.
βf = δf − β − lf v−1
x γ In order to consider above two offsets, the preview error is
(4)
βr = −β − lr v−1
x γ employed as follows:
where βf and βr represent the equivalent sideslip angles of the ep = ed + Leψ (7)
front and rear tires, respectively.
where ep is the preview error that incorporates ed and eψ ; and
From (4), the lateral tire forces can be calculated by
the weight of eψ in ep is determined by the preview length L.
⎛ ⎞
Invoking (6), the dynamics of ed and eψ can be given
⎜ ⎟ by [30]
Ffay = Ffly + Ffry = ⎝Cfl + Cfr + Cfl + Cfr ⎠βf
Cf Cf ėd = vx sin eψ + vy cos eψ ≈ vx eψ + vx β
 
ėψ = ψ̇v − ψ̇s ≈ γ − κ(s)vs (8)
= (Cr + Cr ) δf − β − lf v−1
x γ
⎛ ⎞ where κ(s) is the curvature of the desired path, and s is related
⎜ ⎟ to the curvilinear coordinate of the projection point; vs denotes
Fray = Frly + Frry = ⎝Crl + Crr + Crl + Crr ⎠βr
the speed of the corresponding projection point of the AGV
Cr Cr on the ideal path, and can be calculated by
  
= Cf + Cf −β − lr v−1
x γ vs = ṡ ≈
vx
. (9)
  1 − ed κ(s)
Ffby = (Cfr − Cfl )βf = (Cfr − Cfl ) δf − β − lr v−1
x γ (5)
In order to elucidate the dynamics of the preview error, the
first-order and second-order derivatives of ep can be given by
where Cf and Cr stand for the equivalent cornering stiffnesses
of the front and rear tires running in the linear regions, respec- ėp = vx eψ + vx β − Lκvx + Lγ (10)
tively; and the uncertain and nonlinear parts in the cornering   T
ëp = ge + a11 a21 L a12 a22 L vx β β vx γ γ
stiffnesses of front and rear tires are considered by Cf and
Cr , respectively. AT X
With consideration of the steering mechanism misalignment,   T
+ a11 a21 L a12 a22 L vx β β vx γ γ
an additional fault ratio cs in the steering mechanism is intro-
duced, and the actual steering angle of the vehicle wheels can AT X
be calculated by δf = (cs + cs )u. Then, by substituting (5) = ge + AT X + bu + bu + de (11)
Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
2366 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 53, NO. 4, APRIL 2023

where the infinity ρ(0) may cause singular issue in the controller
design. Nevertheless, we can still select a small ri2 to
b = (b1 vx + b2 L)cs
achieve sufficient large boundaries for the initial conditions.
b = (b1 vx + b2 L)(cs + cs ) + (b1 vx + b2 L)cs While using exponential decay function in conventional PPC
dκ schemes [8], [9], [10], [19], [20], [31], overlarge initial con-
ge = v̇x eψ − Lvs vx − κv2x , de = AT X + dβ + dγ .
ds straint in exponential decay boundaries will deteriorate the
convergence rate. In comparison, the proposed hyperbolic
C. Problem Formulation
tangent function can provide the constant deviation of the
The performance of the path-following and velocity con- boundary convergence time between ri2 = 0 and ri2 = 0
trollers for the autonomous vehicle can be indicated by the with respect to the time constant ri2 /ri1 independent of ini-
preview error ep and the velocity error ev , respectively. tial states. This implies that the convergence rate of the preset
To follow the ideal speed, ev can be calculated by boundaries with ri2 = rA will be consistent with the one with
ev = vx − vd (1 − ed κ) (12) ri2 = rB > rA since t = (rB − rA )/ri1 .
Remark 3: Through homeomorphic mapping in (15), the
where vd is the desired speed on the reference path. issue of the convergence for ei in the preset region can be
In order to preset the desired performance, the constraints converted to the stability of the radially unbounded zi . In par-
for ep and ev are predefined as follows: ticular, zi grows to infinity when ei approaches its bounds, and
− α i ρi (t) < ei (t) < α i ρi (t) (13) zi = 0 if and only if ei = 0. In addition, ei can be calculated
by the following functions from zi :
where α i and α i are the lower and upper constant gains of
the boundary function, and i = p, v; ρi (t) can be given by ei = ρi (t)fi (zi ) (17)
a hyperbolic tangent function with the positive and strictly α i exp(zi + i ) − α i exp[−(zi + i )]
fi (zi ) = . (18)
decreasing properties as follows: exp(zi + i ) + exp[−(zi + i )]

ρi (t) = coth(ri1 t + ri2 ) − 1 + ρi ∞

D. Preliminaries
exp(ri1 t + ri2 ) + exp[−(ri1 t + ri2 )]
= − 1 + ρi ∞ (14) Lemma 1 [32]: For ∀x, y ∈ R, the following inequality
exp(ri1 t + ri2 ) − exp[−(ri1 t + ri2 )]
holds:
where ri1 and ri2 are the positive constants related to the con-
vergence rate and the envelope range of the initial constraints; xy ≤ λm m−1 |x|m + λn n−1 |y|n (19)
ρi∞ is the steady state of the boundary function ρi (t).
To guarantee the constraint (13) during the converging pro- where λ > 0, m > 1, and (n − 1) (m − 1) = 1.
cess, the following homeomorphic mapping function from ei Corollary 1: For ∀x, y ∈ R, it can be obtained that
to zi is implemented [31]: − x(x + y)3 ≤ kα x4 + kβ y4 (20)
 
1 yi α i + α i α i ei
zi = ln , yi = (15) where kα = 1 − (9/4)p4/3 and kβ = (3/4p4 ) + (1/12), p > 0.
2 α i α i − yi α i ρi
Proof: From (20), we have
which implies that ei can be maintained within the prescribed
performance constraint (13) for t ≥ 0 if zi can be stabilized. − x(x + y)3 = −xy3 − 3x2 y2 − 3x3 y − x4 . (21)
Remark 1: The proposed path-following controller is per-
Invoking Lemma 1, the following inequality holds:
formed by a steering mechanism in the autonomous vehicle,
! !m  
and transfers the steering inputs from the steering column to ! ! −1
− x3 y ≤ m−1 λm !−x3 ! + nλn |y|n . (22)
the vehicle front wheels. In the steering mechanism, it is dif-
ficult to measure the accurate transmission ratio cs . In the Letting m ≥ 4/3 and n ≥ 4, then (22) can be rewritten as
meantime, the unknown nonlinear terms of the tire forces, as
given in b1 and b2 of (6), will lead to additional input faults 3λ4/3 !! 3 !!4/3 1 3λ4/3 4 1
− x3 y ≤ !−x ! + 4 |y|4 = x + 4 y4 . (23)
in the path-following system (11). In terms of the velocity 4 4λ 4 4λ
control, the traction losses from the motors to the tires in (2) Resorting to the following inequality:
and (3), such as the unknown motor efficiency and tire slip,  2
are considered as the actuator faults in the longitudinal vehicle 6xy + y2 > 0 (24)
dynamics. In (11), b is always constant, and A is time-invariant
with the constant speed. In the case of speed variation, A with we have
the initial speed is constructed as the time-invariant one for
− xy3 < 3x2 y2 + 12−1 y4 . (25)
the adaptive law, and the inaccurate part is involved in A.
Remark 2: From (14), we can further obtain that Substituting (22) and (25) into (20), one can obtain
2    
ρ(t) = + ρi∞ (16) 9 3 1 4
exp(2ri2 ) exp(2ri1 t) − 1 − x(x + y) ≤ − 4/3 + 1 x +
3 4
+ y . (26)
4p 4p4 12
which implies that ρ is infinity while t = 0 with ri2 = 0,
kα kβ
and the radially unbounded ρ(0) can include all the ini-
tial conditions with bounded states. As a matter of fact, This completes the proof.
Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
LIANG et al.: FIXED-TIME AND FAULT-TOLERANT PATH-FOLLOWING CONTROL FOR AUTONOMOUS VEHICLES 2367

Lemma 2 [33]: For ∀xi ∈ R, i = 1, . . . , k, one has updated by

 k n ⎧
" "k
⎪ ˙
|xi | ≤ |xi |n , 0 < n < 1 ⎪
⎪ b̂ = pb b̂3 (u1 + u2 ) + pbb b̂−1
(27) ⎪
⎨˙  T
i=1 i=1 Â = pa Xh − paa â31 â32 â33 â34
 k n ⎪ ˙ (36)
" "k ⎪
⎪ d̂m = |h|pm − pmm d̂m 3
k 1−n
|xi | ≤ xin , 1 < n. (28) ⎪
⎩˙
λ̂3 = λ1 |h|pλ − pλ λ λ̂33
i=1 i=1
Corollary 2: For ∀x ∈ R, m > 1, and 0 < n < 1, it follows: where k1 , k2 , k3 , pa , paa , pb , pbb , pm , pmm , pλ , and pλλ are pos-
itive control gains; ÂT = [â1 â2 â3 â4 ]; h is defined in (39).
− x2m ≤ −21−m x2n + 1. (29) Then, the preview error can converge to a small neighborhood
Proof: From (27) and (28), we have of zero within a fixed time, and the preset constraint (13) can
 n  m   be guaranteed.
x2n ≤ x2 + 1 ≤ x2 + 1 ≤ 2m−1 x2m + 1 . (30) Proof: Differentiating zp and invoking (11) produce
# −1 $
This completes the proof.  −1  
Lemma 3 [34]: For ẋ = f (x) with positive definite function żp = 2−1 ρp−1 yp + α p − yp − α p ėp − ep ρ̇p ρ −1
p .
V(x), if the following inequality holds:
ξ1
V̇ ≤ −kα V m − kβ V n + C (31) (37)
with 0 < kα , 0 < kβ , 1 < m, 0 < n < 1, and 0 < C < ∞,
A Lyapunov function candidate is selected as
then the system can be stabilized to the following set within
a fixed time:
⎧ V1 = z2p /2. (38)
$1/m % &1/n ⎫
⎨# C C ⎬
V(x) ≤ VM = min , (32) The time derivative of V1 is given as
⎩ kα (1 − ω) kβ (1 − ω) ⎭ ⎛ ⎞
⎜ ⎟
where 0 < ω < 1, and the settling time T is estimated by V̇1 = zp żp = zp ξ1 ⎝−k1 z3p ξ1 −1 + ėp − ep ρ̇p ρp−1 + k1 z3p ξ1 −1 ⎠
 −1
T ≤ Tmax = [kα ω(m − 1)]−1 + kβ ω(1 − n) . (33) h
(39)
III. C ONTROLLER D ESIGN
which implies zp can be stabilized to zero in a fixed time if
In this section, the path-following and velocity controllers h = ėp − ep ρ̇p ρp−1 + k1 z3p ξ1 −1 = 0 with a positive control
will be designed to stabilize the preview and velocity gain k1 .
errors within the fixed time, respectively, and the prescribed Considering the virtual law h, the Lyapunov function can-
performance constraints will be guaranteed. didate is rewritten as
1 1 2 (1 − εb ) 2 1 T 1 2
A. Path-Following Controller Design V2 = V1 + h2 + d̃ + λ̃3 + Ã Ã + b̃
2 2pm m 2pλ 2pa 2pb
The states of the controlled vehicle, including the longitu-
(40)
dinal speed vx , sideslip angle β, and yaw rate γ , are assumed
to be measured by sensors or estimated by observers. In where d̃m = dm − d̂m , b̃ = b − b̂−1 , λ̃3 = λ3 − λ̂3 , and
addition, the unavailable and unmodeled terms are defined à = A − Â.
as perturbations in the vehicle system with the following Differentiating (40), and applying controller (34) and adap-
assumption. tive law (36) produce
Assumption 1: With consideration of actuator faults in
T˙ −1 −2 ˙ ˙
system (11), we assume that the steering error coefficient V̇2 = V̇1 + hḣ − p−1 −1
a à  + pb b̃b̂ b̂ + pm d̃m d̃ m
! !
ct = (b/b) is bounded by an unknown positive constant εb ! ! ˙
+ p−1λ (1 − εb )!bb̂!λ̃3 λ̃3
such that |ct | ≤ εb < 1. In terms of the perturbation in (11),  
we assume that there exist an unknown positive constant dm = zp ξ1 −k1 z3 ξ1 −1 + h + hξ2 + hge + hAT X
such that |de | ≤ dm . T˙ −1 −2 ˙ 
− p−1a à  + pb b̃b̂ b̂ − h(b + b)b̂ zp ξ1
Then, we are ready to give the first result of this article.
Theorem 1: Consider the following fixed-time fault-tolerant + ge + ξ2 + ÂT X + k2 h3 + k3 h1/3 + d̂m sgn(h)
˙ 1 !! !! ˙
steering controller with the prescribed performance: + ηsgn(h)) + hde − p−1 m d̃m d̂ m − (1 − εb ) !bb̂!λ̃3 λ̂3
pλ
u = −b̂(u1 + u2 ) (34) ≤ −k1 z4p − k2 h4 − k3 h4/3 + ÃT Xh
  3 
with − p−1
T
a à pa Xh − paa â1
T
⎧ â32 â33 â34
⎪
⎪ u = z ξ + ge + ξ2 + ÂT X + k2 h3 + k3 h1/3  
⎨ 1 p 1 + p−1 −2 ˙
b b̃b̂ b̂ − b̃b̂ zp ξ1 + ge + ξ2 + Â X + k2 h + k3 h
T 3 1/3
u2 = d̂m + η sgn(h) (35) # $
⎪
⎪   b
⎩ η = λ̂3 − 1 λ1 , λ1 = |u1 + u2 | + λ2 , λ2 > 0 − η|h| + bb̂(u1 + u2 )h
b

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
2368 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 53, NO. 4, APRIL 2023

 
˙ 1 !! !!
d̃m d̂m − (1 − εb ) !bb̂!λ̃3 λ̂˙ 3
1
+ de |h| − d̂m |h| − According to Lemma 3, zp is bounded by zpM , which can
pm pλ
be calculated by
≤ −k1 z4p − k2 h4 − k3 h4/3 + ÃT Xh 2#
  $1 # $13
 3 ! ! 1 4C 4 2nC 2n
− p−1a à pa Xh − paa â1
T
â32 â33 â34
T
!zp ! ≤ 2V2 ≤ zpM = min ,
  ζ (1 − ω2 ) ϑ(1 − ω2 )
−2 ˙
+ p−1
1
b b̃b̂ b̂ − b̃b̂ zp ξ1 + ge + ξ2 + Â X + k2 h + k3 h
T 3 3
(43)
  ! !
1 ˙ ! !
+ de |h| − d̂m |h| − d̃m d̂m − εb !bb̂z!λ2 with 0 < ω2 < 1.
pm
⎡ ⎤ By substituting (17) and (18) into (13), the preview error
⎢  ! ! −1 ⎥ ! !  will converge into the following residual set:
! ! ⎥ εb !!bb̂!! − 1 λ1 |z|     
+ ⎢⎣λ̂3 + εb !bb̂! − 1 ⎦ ! ! α exp zpM + p − α p exp − zpM + p
!ep ! < ρ p      (44)
−λ3
exp zpM + p + exp − zpM + p
≤ −k1 z4p − k2 h4 − k3 h4/3 − paa p−1
a
 3 and the fixed settling time T is bounded by
× Ã â1 â2 â3 â4 + pbb p−1
T 3 3 3 T
b̃b̂
−1 −1
!b!
! ! T ≤ Tmax = (ζ ω2 )−1 + [ϑω2 (1 − n)]−1 . (45)
− pmm pm d̃m d̂m − pλλ pλ (1 − εb )!bb̂!λ̃3 λ̂33
3
(41)
Then, we will show that the preview error ep will never
where violate the preset constraint for any t ≥ 0.
 
ξ2 = −ρp−2 ėp ρ̇p + ep ρ̈p ρp − ep ρ̇p2 From (42), one can obtain that
  
V̇2 ≤ −ζ V2 2 + C ≤ −ζ V2 + C, 1 ≤ V2
− k1 3z2p żp ξ1 −1 − z3 ξ1 −2 ξ̇1 . V̇2 ≤ −ϑV2 n + C ≤ −ϑV2 + C, 0 ≤ V2 < 1
(46)

Invoking Lemmas 1 and 2, Corollary 1, and Assumption 1, which implies

and substituting (34)–(36) into (41) yield
V̇2 ≤ − min(ζ, ϑ)V2 + C. (47)
1 1
V̇2 ≤ − k1 z4p − k2 h4 − k2 h4 − k3 h4/3 − 2ka1 paa
2 2 Then, we have
 2   # $
1 " 4
1 2 2 C C
× pa2 2
ãi − 2kb1 pbb pb2
b̃ V2 ≤ V2 (0) − √ exp[− min(ζ, ϑ)t] + .
2pa 2pb min(ζ, ϑ) min(ζ, ϑ)
i=1 (48)
 
1 2 2
− 2kλ1 (1 − εb )pλλ p2λ λ̃3 − 2km1 pmm It follows that, zp is always limited
√ by monotonically
2pλ
 n decreasing boundaries since |zp | ≤ 2V2 holds. Resorting to
 2
1 " 2
4
1 2 homeomorphic mapping and barrier Lyapunov theorem, the
× pm2
d̃ − 2 kb1 pbb pb
n−2 n
ãi
2pm m 2pa bounded zp can always guarantee yp subject to the limited
i=1
  region [−α p , α p ], and then the preset boundaries for ep can
1 2 n
− 2 kb1 pbb pb
n−2 n
b̃ + pλλ p−1
λ λ3 km1 pmm
4 be, in turn, guaranteed.
2pb Invoking (48), the ultimate bounds of the preview error are
 
1 2 n 4 - 4 .
× pmn
d̃ − 2n−2 kλ1 (1 − εb )pλλ pnλ 2C
min(ζ,ϑ) +
2C
− min(ζ,ϑ) +
2pm m ! ! α e − αpep p

# $ !ep (∞)! < ρ p 4 - 4 . . (49)

(1 − εb ) 2 n 2C
+ p 2C
− min(ζ,ϑ) + p
× λ̃3 + ka2 paa p−1 a a −2
4 n−2
e min(ζ,ϑ)
+e
2pλ
This completes the proof.
+ kb2 pbb p−1
b b + kλ2 (1 − εb ) + km2 pmm pm dm
4 −1 4
Remark 4: In the PPC protocol, the preview error
≤ −ζ V2 2 − ϑV2 n + C (42) is proposed to be limited by the performance functions.
where 0 < n < 1 and the positive constants ka1 , ka2 , kb1 , kb2 , According to (14), the decay rate of the preset boundary can
km1 , km2 , kλ1 , and kλ2 satisfy the conditions in Corollary 1. be predefined by ri1 . In the meantime, a small value of ri2
Furthermore, the parameters in (42) are given by can avoid a singular issue in the proposed controller, and pro-
⎧ - duces sufficient large initial boundaries to include initial state
−1
⎪
⎪ ζ = 3−1 min k1 , 2k2 , ka1 paa p−1 a , kb1 pbb pb errors. Large values of the control gains k1 , k2 , and k3 can pro-
⎪
⎪ .
⎪
⎪ vide fast convergence rate, whereas produce serious chattering
⎪
⎪ pλ λ (1 − εb )−1 kλ1 p−1 , k p p −1
⎪ λ m1 mm m
⎪ ϑ = 2n−2 min/k , k 28/3−n , k p pn , k p pn
⎪ in the controller. Furthermore, ksat in the saturation function
⎪
⎪ aa a b1 bb0 b is introduced to replace the sign function. [pa , paa ], [pb , pbb ],
⎪
⎨
1 3 a1
q
 (1 − ε b ) 1−n n
pλ pλ λ k λ1 , km1 pmm pm [pm , pmm ], and [pλ , pλλ ] in adaptive laws are employed to sta-
⎪ −1
⎪
⎪ C = 2−1 k1 + ka1 p−1 a paa + (1 − εb )kb1 pb pbb
bilize the estimate errors of the unknown parameters A, b, dm ,
⎪
⎪ and λ3 in the controlled system, respectively. In particular,
⎪
⎪ +km2 p−1 −1
⎪
⎪ m pmm + (1 − εb )kλ1 pλ pλ λ
⎪
⎪
paa , pbb , pmm , and pλλ are introduced to achieve the fixed-
⎪ +pka2 −1 paa a4 + kb2 p−1 −1
⎪ b pbb b + km2 pm
4
⎪
⎩
time convergence, and can also avoid the divergence of the
−1
×pmm (dm ) + kλ2 (1 − εb )pλ pλλ λ3 .
4 4 estimate errors.

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
LIANG et al.: FIXED-TIME AND FAULT-TOLERANT PATH-FOLLOWING CONTROL FOR AUTONOMOUS VEHICLES 2369

Remark 5: To demonstrate the superiority of the proposed

control scheme, three different comparative path-following
controllers are constructed in the Appendix. A fixed-time
prescribed performance controller without considering actu-
ator faults is constructed in Proposition 1. In Proposition 2,
a fixed-time fault-tolerant controller without the preset con-
straint is given to analyze the PPC scheme. In addition, a
conventional fixed-time controller without fault-tolerant and
PPC schemes is constructed in Proposition 3. The proofs of
Propositions 1–3 are omitted for brevity as they are not much
Fig. 2. HIL test implementation. (a) HIL tests incorporating cRIO-9038 and
different from the Theorem 1. CarSim. (b) Schematic diagram.

B. Velocity Controller Design TABLE I

In order to follow ideal speeds for the vehicle, a fixed-time M AIN V EHICLE PARAMETERS
prescribed performance velocity controller with fault-tolerant
control scheme will be designed. In the meantime, the follow-
ing assumptions are required.
Assumption 2: dv in the longitudinal vehicle dynamics (3)
is bounded by an unknown positive constant dvm , such that
|dv | ≤ dvm .
Assumption 3: ηT in (3) is bounded by an unknown positive
constant ηTm , such that |ηT | ≤ ηTm < 1.
Theorem 2: Consider the following velocity controller
Ta = −b̂v (T1 + T2 ) (50)
with
⎧
T = k z3 ξ −1 + k z ξ −1 − ev ρ̇v ρv−1
1/3
⎪
⎪
⎨ 1 v1 v v  v2 v v
T2 = d̂vm + ηv sgn zv vξ (51) TABLE II
⎪
⎪   C ONTROLLER PARAMETERS
⎩ ηv = λ̂v3 − 1 λv1 , λv1 = |u1 + u2 | + λv2 , λv2 > 0

updated by
⎧ 
⎪ ˙ 3 k z3 ξ −1 + k z1/3 ξ −1 − e ρ̇ ρ −1
⎪
⎪ b̂ v = pvb b̂v v1 v v v1 v v v v v
⎪
⎨
+ηv sgn(zv ) + pvb b̂v −1
˙ ! ! (52)
⎪
⎪ d̂vm = pvm !zv ξv ! − pvm d̂vm
3
⎪
⎪ ! !
⎩˙
λ̂v3 = pvλ !zv ξv !λ3 − pvλλ λ̂3v3
where ξv = (2ρv )−1 [(yv + α v )−1 − (yv − α v )−1 ]. Then, the
velocity error in (12) will converge to a small neighborhood
of zero within the fixed time, and is subject to the preset
constraint (14).
Proof: The proof is omitted for brevity since it is similar to
the proof of Theorem 1.

IV. H ARDWARE - IN -L OOP T ESTS vehicle parameters are listed in Table I. In particular, the
In this section, the proposed controllers are con- above-mentioned path-following controllers are all combined
ducted in HIL tests. As illustrated in Fig. 2, the com- with the same velocity controller in Theorem 2 to perform
parative path-following control protocols, including the fair comparisons, and the controller parameters are selected
prescribed performance fault-tolerant controller (PPC-FTC) in Table II. Furthermore, the data interface is implemented by
in Theorem 1, the prescribed performance controller without the LabVIEW platform, which guarantees the online iteration.
fault-tolerant control (PPC) in Proposition 1, the conven- The S-curve and double-lane-change maneuvers with var-
tional fixed-time-based controller with fault-tolerant control ious driving conditions are implemented in the HIL tests to
(CBC-FTC) in Proposition 2, and the conventional fixed- demonstrate the effectiveness and robustness of the proposed
time-based controller (CBC) in Proposition 3, are respectively controllers. To conduct the actuator faults in the HIL tests, the
loaded in a hardware controller cRIO-9038. The nonlinear nonlinear tire models and tire slips have been incorporated by
vehicle model is provided by the CarSim platform, and the the CarSim platform. In particular, additional fault coefficient,

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
2370 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 53, NO. 4, APRIL 2023

ct = 0.8 sin(t), is added in the steering mechanism to demon-

strate the effectiveness of the fault-tolerant control scheme in
the path-following controller. Furthermore, the actuator faults
are all unknown to the designed controllers.

A. S-Curve in HIL Tests

The vehicle is expected to follow an S-curve path with the
continuous clothoid curvature as given in Fig. 3(a), and runs
on a dry road with the high friction coefficient (μ = 0.85). The
desired longitudinal acceleration is planned as follows [30]:
⎧
⎪
⎪ −αs s, 0 ≤ s ≤ 0.25l
⎪
⎪
⎨ −αs (0.5l − s), 0.25l ≤ s ≤ 0.75l
v̇d (s) = αs (l − s), 0.75l ≤ s ≤ 1.25l (53)
⎪
⎪
⎪ −αs (1.5l − s), 1.25l ≤ s ≤ 1.75l
⎪
⎩
αs (2l − s), 1.75l ≤ s ≤ 2l
where αs = 0.02; 2l is the length of the desired S-curve
path; and the ideal initial speed is set as vd (0) = 60 km/h.
From (53), we can observe that the varying speed strategy
performs the decelerating maneuver to avoid overlarge lateral
acceleration while the vehicle enters the S-curve path. The
accelerating maneuver is operated to restore the initial entry
speed while the vehicle exits in the S-curve path. Thus, the
varying speed strategy can provide the continuous accelera-
tion and guarantee the smooth running for the vehicle. The
parameters in the preset constraint for the velocity controller
are chosen as α v = 10, α v = 20, rv1 = 0.1, rv2 = 0.4, and
ρv∞ = 0.05. With the actual initial speed vx (0) = 35 km/h,
the proposed velocity controller can obtain the prescribed
performance subject to the preset velocity constraint, as shown
in Fig. 3(b).
The parameters in the prescribed performance constraint for
the path-following controller are selected as α p = 3, α p = 4,
rp1 = 0.05, rp2 = 0.6, and ρp∞ = 0.01. Meanwhile, various
initial preview errors are implemented in the HIL tests, i.e.,
e+p (0) = 2.952, in which ed (0) = 3 m and eψ (0) = −0.03
rad, and e− p (0) = −1.968, in which ed (0) = −2 m and
eψ (0) = 0.02 rad.
The results for the S-curve path in the HIL tests are plotted
in Fig. 3. It can be seen that all the preview errors can converge Fig. 3. Results in HIL tests for S-curve path. (a) Curvatures of S-curve path.
(b) Velocity in S-curve. (c) and (e) Preview errors of proposed controllers
to the neighborhood of zero by using the fault-tolerant control with e+ − +
p and ep , respectively. (d) Steering inputs with ep . (f) Steering inputs
schemes, including the PPC-FTC and CBC-FTC. In particu- −
with ep . (g) Vehicle trajectories. (h) Enlarged vehicle trajectories.
lar, by using the proposed PPC-FTC, the preview error can be
strictly limited within the predefined boundaries, as shown in
Fig. 3(c) and (e). In contrast, the preview error of the vehi-
cle with the CBC-FTC transgresses the preset performance B. Double-Lane-Change in HIL Tests
constraint. In addition, we notice that the vehicle controlled To demonstrate the robustness of the proposed PPC-FTC,
by the PPC or CBC without the fault-tolerant control scheme the driving maneuver of double-lane-change is implemented
looses the stability while large steering actuator faults occur. in the HIL tests. In this case, the same controller parame-
From Fig. 3(d) and (f), one can observe that large fluctuations ters are utilized, and the controlled vehicle runs on a slippery
exhibit in the steering inputs of the PPC and CBC, as the large wet road with a lower tire-road friction coefficient (μ = 0.5).
steering faults are not considered in the controller design. The The ideal velocity is set as a constant value of 60 km/h, and
steering inputs of the PPC-FTC and CBC-FTC can stabilize the actual initial velocity is given as vx (0) = 50 km/h. The
the preview errors to a small region. To further analyze the parameters of the velocity constraint are chosen as α v = 10,
path-following results, the trajectories of the controlled vehi- α v = 20, rv1 = 0.7, rv2 = 0.4, and ρv∞ = 0.01. By using the
cle are plotted in Fig. 3(g) and (h), and it can be found that proposed velocity controller in Theorem 2, the vehicle can
the path-following performance of the PPC-FTC can always follow the desired constant velocity, and always evolve within
outperform the other comparative controllers. the prescribed performance constraint, as shown Fig. 4(b).

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
LIANG et al.: FIXED-TIME AND FAULT-TOLERANT PATH-FOLLOWING CONTROL FOR AUTONOMOUS VEHICLES 2371

and no instability issue occurs in the HIL tests due to the

short length of the double-lane-change path. As a result, the
PPC-FTC can still obtain superior path-following performance
over the other comparative controllers, as shown in Fig. 4(c)
and (e). We can observe that the preview errors of the vehi-
cle controlled by the PPC-FTC and PPC never violate the
predefined performance constraint, whereas the CBC and FTC
without the prescribed performance algorithm cannot guaran-
tee the preview errors evolving within the preset constraint.
Thus, this demonstrates the effectiveness of the proposed PPC
scheme. From the trajectories of the controlled vehicle, as
given in Fig. 4(g) and (h), we can analyze that the proposed
PPC-FTC, which incorporates the PPC and fault-tolerant con-
trol schemes, can achieve the control objectives and always
outperforms the other comparative controllers.
In particular, as shown in Fig. 3(c) and (e), one can observe
that the serious input faults lead to the vehicle instability while
using the PPC and CBC without the fault-tolerant control
scheme. This is because that the input faults result in addi-
tional path-following errors, which are, in turn, obtained by
the PPC or the CBC, and extra steering inputs are required
to compensate for the additional path-following errors. Then,
the overlarge control inputs are generated and lead to the
vehicle instability. In contrast, the boundaries of the actuator
faults are estimated by the adaptive law (37) in the proposed
PPC-FTC, and the accurate control inputs for the vehicle
system can be implemented through compensating for the
actuator faults before they produce the path-following errors.
In Fig. 4(c) and (e), the vehicle can be operated in stable
situations due to the small range of the double-lane-change
maneuver. However, we can find that the vehicle controlled
by the PPC or CBC still exhibits large path-following errors.
Resorting to the estimation of the actuator faults, the PPC-FTC
can always guarantee the preview error evolving within the
preset performance constraints.

V. C ONCLUSION
In this article, a fixed-time PPC scheme has been proposed
to follow the desired paths and velocities for the AGVs with
Fig. 4. Results in HIL tests for double-lane-change path. (a) Curvatures of the consideration of actuator faults. Based on the homeo-
double-lane-change path. (b) Velocity in double-lane path. (c) and (e) Preview
errors of proposed controllers with e+ −
p and ep , respectively. (d) Steering inputs
morphic mapping, the hyperbolic tangent function has been
with e+p . (f) Steering inputs with e− . (g) Vehicle trajectories. (h) Enlarged
p utilized to constrain the system errors subject to the preset
vehicle trajectories. constraints. The adaptive laws have been designed to esti-
mate the unknown vehicle parameters and the actuator faulty
coefficients. Finally, the effectiveness of the proposed control
On this basis, the proposed PPC-FTC and other comparative
protocol has been verified in the HIL tests. Compared with the
path-following controllers are implemented in the HIL tests.
controllers without the fault-tolerant control or PPC scheme,
The parameters of the prescribed performance function for
the proposed controller, which integrates the fault-tolerant
double-lane-change maneuver are preset as α p = 2, α p = 1,
control, PPC, and adaptive control schemes, can always per-
rp1 = 0.07, rp2 = 0.6, and ρp∞ = 0.01. The positive and
form a superior path-following performance over the other
negative types of the initial preview errors are set as e+
p (0) = comparative controllers.
0.672, in which ed (0) = 0.8 m and eψ (0) = −0.08 rad, and
e−
p (0) = −1.088, in which ed (0) = −1.2 m and eψ (0) =
0.07 rad. A PPENDIX
The results in HIL tests are plotted in Fig. 4. Unlike the Proposition 1: Without considering the actuator faults, a
cases in the S-curve path, the preview errors of the controlled prescribed performance controller is given as
vehicle can be driven to the neighborhood of zero by using all 
the controllers of the PPC-FTC, PPC, CBC-FTC, and CBC, u = −b̂ u1 + d̂m sgn(h) (54)

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
2372 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 53, NO. 4, APRIL 2023

updated by [7] C. Hu et al., “RISE-based integrated motion control of autonomous

 ground vehicles with asymptotic prescribed performance,” IEEE Trans.
˙
b̂ = pb b̂3 u1 + d̂m sgn(h) + pbb b̂−1 (55) Syst., Man, Cybern., Syst., vol. 51, no. 9, pp. 5336–5348, Sep. 2021.
[8] C. Hu, Z. Wang, Y. Qin, Y. Huang, J. Wang, and R. Wang, “Lane keeping
control of autonomous vehicles with prescribed performance considering
in which u1 , d̂m , and h have the same definitions with the the rollover prevention and input saturation,” IEEE Trans. Intell. Transp.
related parameters and variables in Theorem 1. Then, the path- Syst., vol. 21, no. 7, pp. 3091–3103, Jul. 2020.
following system (7) will be driven to a stable region in a fixed [9] G. Guo and D. Li, “Adaptive sliding mode control of vehicular pla-
toons with prescribed tracking performance,” IEEE Trans. Veh. Technol.,
time, and is subject to the preset constraint in (13). vol. 68, no. 8, pp. 7511–7520, Aug. 2019.
Proposition 2: A fixed-time controller considering the actu- [10] H. Dong and X. Yang, “Finite-time prescribed performance control for
ator faults in the vehicle system is given as follows: space circumnavigation mission with input constraints and measure-
ment uncertainties,” IEEE Trans. Aerosp. Electron. Syst., vol. 58, no. 4,
u = −b̂(u1c + u2c ) (56) pp. 3209–3222, Aug. 2022, doi: 10.1109/TAES.2022.3148356.
⎧ 1/3 [11] C. P. Bechlioulis and G. A. Rovithakis, “Robust adaptive con-
⎪
⎪ u = k e + k
3
3 hep + ge + Â X
T
⎨ 1c 2 p trol of feedback linearizable MIMO nonlinear systems with
prescribed performance,” IEEE Trans. Autom. Control, vol. 53,
u2c = d̂m + η sgn(hc ) (57)
⎪
⎪   no. 9, pp. 2090–2099, Oct. 2008.
⎩ η = λ̂ − 1 λ , λ = |u + u | + λ , λ > 0 [12] C. Hu, R. Wang, and F. Yan, “Integral sliding mode-based com-
c 3c 1c 1c 1c 2c 2c 2c
posite nonlinear feedback control for path following of four-wheel
independently actuated autonomous vehicles,” IEEE Trans. Transport.
updated by Electrific., vol. 2, no. 2, pp. 221–230, Jun. 2016.
⎧
⎪ ˙ [13] Z. Liang, Z. Wang, J. Zhao, and X. Ma, “Fast finite-time path
⎪
⎪ b̂ = pb b̂3 (u1c + u2c )
⎪
⎨˙  following control for autonomous vehicle via complete model-free
 = pa Xh − paa â31 â32 â33 â34 approach,” IEEE Trans. Ind. Informat., early access, Apr. 7, 2022,
˙ (58) doi: 10.1109/TII.2022.3165630.
⎪
⎪ d̂m = |hc |pm − pmm d̂m 3
⎪
⎪
[14] Z. Zuo, “Nonsingular fixed-time consensus tracking for second-order
⎩˙ multi-agent networks,” Automatica, vol. 54, pp. 305–309, Apr. 2015.
λ̂ = λ |h |p − p λ̂3
3c 1 c λ λ λ 3c [15] J. Sun, J. Yi, Z. Pu, and X. Tan, “Fixed-time sliding mode distur-
bance observer-based nonsmooth backstepping control for hypersonic
where hc = ėp + k1 e3p .
Then, the path-following system (7) vehicles,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 50, no. 11,
will be driven to a stable region in a fixed time. pp. 4377–4386, Nov. 2020.
Proposition 3: Without consideration of the prescribed [16] J. Sun, Z. Pu, J. Yi, and Z. Liu, “Fixed-time control with uncertainty
and measurement noise suppression for hypersonic vehicles via aug-
performance and actuator faults, a fixed-time controller is mented sliding mode observers,” IEEE Trans. Ind. Informat., vol. 16,
designed as no. 2, pp. 1192–1203, Feb. 2020.
 [17] G. Cui, W. Yang, J. Yu, Z. Li, and C. Tao, “Fixed-time prescribed
u = −b̂ u1c + d̂m sgn(hc ) (59) performance adaptive trajectory tracking control for a QUAV,” IEEE
Trans. Circuits Syst. II, Exp. Briefs, vol. 69, no. 2, pp. 494–498,
updated by Feb. 2022, doi: 10.1109/TCSII.2021.3084240.
 [18] J. Zhang, S. Yu, D. Wu, and Y. Yan, “Nonsingular fixed-time terminal
˙
b̂ = pb b̂3 u1c + d̂m sgn(hc ) + pbb b̂−1 (60) sliding mode trajectory tracking control for marine surface vessels with
anti-disturbances,” Ocean Eng., vol. 217, no. 1, pp. 108–158, Dec. 2020.
in which u1c and hc have the same definitions with the related [19] C. P. Bechlioulis, G. C. Karras, S. Heshmati-Alamdari, and
K. J. Kyriakopoulos, “Trajectory tracking with prescribed performance
parameters and variables in Proposition 2. Then, the path- for underactuated underwater vehicles under model uncertainties and
following system (7) will be driven to a stable region in a external disturbances,” IEEE Trans. Control Syst. Technol., vol. 25,
fixed time. no. 2, pp. 429–440, Mar. 2017.
[20] Z.-G. Zhou, D. Zhou, X.-N. Shi, R.-F. Li, and B.-Q. Kan, “Prescribed
performance fixed-time tracking control for a class of second-order non-
R EFERENCES linear systems with disturbances and actuator saturation,” Int. J. Control,
vol. 94, no. 1, pp. 223–234, Mar. 2019.
[1] Y. Wu, L. Wang, J. Zhang, and F. Li, “Path following control of
[21] Q. Hu, W. Chen, L. Guo, and J. D. Biggs, “Adaptive fixed-time atti-
autonomous ground vehicle based on nonsingular terminal sliding mode
tude tracking control of spacecraft with uncertainty-rejection capability,”
and active disturbance rejection control,” IEEE. Trans. Veh. Technol.,
IEEE Trans. Syst., Man, Cybern., Syst., vol. 52, no. 7, pp. 4634–4647,
vol. 68, no. 7, pp. 6379–6390, Jul. 2019.
Jul. 2022, doi: 10.1109/TSMC.2021.3100903.
[2] N. Guo, X. Zhang, Y. Zou, B. Lenzo, and T. Zhang, “A computationally
efficient path-following control strategy of autonomous electric vehi- [22] Q. Chen, M. Tao, X. He, and L. Tao, “Fuzzy adaptive non-
cles with yaw motion stabilization,” IEEE Trans. Transport. Electrific., singular fixed-time attitude tracking control of quadrotor UAVs,”
vol. 6, no. 2, pp. 728–739, Jun. 2020. IEEE Trans. Aerosp. Electron. Syst, vol. 57, no. 5, pp. 2864–2877,
[3] R. Wang, C. Hu, F. Yan, and M. Chadli, “Composite nonlinear feed- Oct. 2021.
back control for path following of four-wheel independently actuated [23] X. Cao, Y. Tian, X. Ji, and B. Qiu, “Fault-tolerant controller design
autonomous ground vehicles,” IEEE Trans. Intell. Transp. Syst., vol. 17, for path following of the autonomous vehicle under the faults in
no. 7, pp. 2063–2074, Jul. 2016. braking actuators,” IEEE Trans. Transport. Electrific., vol. 7, no. 4,
[4] W. Wang, Y. Zhang, C. Yang, T. Qie, and M. Ma, “Adaptive model pp. 2530–2540, Dec. 2021.
predictive control-based path following control for four-wheel indepen- [24] G. Zhang, H. Zhang, X. Huang, J. Wang, H. Yu, and R. Graaf, “Active
dent drive automated vehicles,” IEEE Trans. Intell. Transp. Syst., vol. 23, fault-tolerant control for electric vehicles with independently driven rear
no. 9, pp. 14399–14412, Sep. 2022, doi: 10.1109/TITS.2021.3128268. in-wheel motors against certain actuator faults,” IEEE Trans. Control
[5] J. Chen, Z. Shuai, H. Zhang, and W. Zhao, “Path following control of Syst. Technol., vol. 24, no. 5, pp. 1557–1572, Sep. 2016.
autonomous four-wheel-independent-drive electric vehicles via second- [25] C. Huang, F. Naghdy, and H. Du, “Fault tolerant sliding mode predictive
order sliding mode and nonlinear disturbance observer techniques,” control for uncertain steer-by-wire system,” IEEE Trans. Cybern.,
IEEE Trans. Ind. Electron., vol. 68, no. 3, pp. 2460–2469, Mar. 2021. vol. 49, no. 1, pp. 261–272, Jan. 2019.
[6] R. Wang, H. Jing, C. Hu, F. Yan, and N. Chen, “Robust H∞ [26] D. Zhang, G. Liu, H. Zhou, and W. Zhao, “Adaptive sliding
path following control for autonomous ground vehicles with delay mode fault-tolerant coordination control for four-wheel independently
and data dropout,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 7, driven electric vehicles,” IEEE Trans. Ind. Electron., vol. 65, no. 11,
pp. 2042–2050, Jul. 2016. pp. 9090–9100, Nov. 2018.

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.
LIANG et al.: FIXED-TIME AND FAULT-TOLERANT PATH-FOLLOWING CONTROL FOR AUTONOMOUS VEHICLES 2373

[27] B. Guo and Y. Chen, “Robust adaptive fault-tolerant control of four- Jing Zhao received the Ph.D. degree in electrome-
wheel independently actuated electric vehicles,” IEEE Trans. Ind. chanical engineering from the University of Macau,
Informat., vol. 16, no. 5, pp. 2882–2894, May 2020. Macau, China, in 2016.
[28] Y.-H. Jing and G.-H. Yang, “Fuzzy adaptive fault-tolerant control for He is currently working with the Department
uncertain nonlinear systems with unknown dead-zone and unmodeled of Electromechanical Engineering, University of
dynamics,” IEEE Trans. Fuzzy Syst., vol. 27, no. 12, pp. 2265–2278, Macau. His research interests include dynamics
Dec. 2019. and control, mechanism and machine theory, fluid
[29] Z. Liang, J. Zhao, Z. Dong, Y. Wang, and Z. Ding, “Torque vectoring mechanics, and finite element analysis.
and rear-wheel-steering control for vehicle’s uncertain slips on soft and
slope terrain using sliding mode algorithm,” IEEE Trans. Veh. Technol.,
vol. 69, no. 4, pp. 3805–3815, Apr. 2020.
[30] Z. Liang, J. Zhao, B. Liu, Y. Wang, and Z. Ding, “Velocity-based path
following control for autonomous vehicles to avoid exceeding road fric-
tion limits using sliding mode method,” IEEE Trans. Intell. Transp. Syst.,
vol. 23, no. 3, pp. 1947–1958, Mar. 2022. Pak Kin Wong received the Ph.D. degree in
[31] Z. Zheng and M. Feroskhan, “Path following of a surface vessel mechanical engineering from The Hong Kong
with prescribed performance in the presence of input saturation and Polytechnic University, Hong Kong, in 1997.
external disturbances,” IEEE/ASME Trans. Mechatronics, vol. 22, no. 6, He is currently a Professor with the Department
pp. 2564–2575, Dec. 2017. of Electromechanical Engineering, University of
[32] H. Wang, B. Chen, and C. Lin, “Adaptive neural tracking control for a Macau, Macau, China. He has published over
class of stochastic nonlinear systems,” Int. J. Robust Nonlinear Control, 279 scientific papers in refereed journals, book
vol. 24, no. 7, pp. 1262–1280, May 2014. chapters, and conference proceedings. His research
[33] L. Yu, G. He, X. Wang, and S. Zhao, “Robust fixed-time sliding mode interests include automotive engineering, fluid trans-
attitude control of tilt trirotor UAV in helicopter mode,” IEEE Trans. mission and control, mechanical vibration, and med-
Ind. Electron., vol. 69, no. 10, pp. 10322–10332, Oct. 2022. ical engineering.
[34] D. Ba, Y.-X. Li, and S. Tong, “Fixed-time adaptive neural track-
ing control for a class of uncertain nonstrict nonlinear systems,”
Neurocomputing, vol. 363, pp. 273–280, Oct. 2019.

Zhixin Yang (Member, IEEE) received the Ph.D.

degree in industrial engineering and engineering
management from the Hong Kong University of
Science and Technology, Hong Kong, China.
He is currently an Associate Professor with
the Department of Electromechanical Engineering,
Zhongchao Liang received the M.S. and Ph.D. University of Macau, Macau, China. His current
degrees in mechanical engineering from the Harbin research interests include prognostic health moni-
Institute of Technology, Harbin, China, in 2011 and toring of engineering system, computer vision, and
2015, respectively. artificial-intelligence-based robotic control.
He was an Academic Visitor with the Department
of Electrical and Electronic Engineering, The
University of Manchester, Manchester, U.K., from
2019 to 2020. He is currently an Associate Professor
with the School of Mechanical Engineering and
Automation, Northeastern University, Shenyang, Zhengtao Ding (Senior Member, IEEE) received
China. His research interests include nonlinear and B.Eng. degree in thermal energy from Tsinghua
adaptive control for intelligent vehicles and robots. University, Beijing, China, in 1984, and the M.Sc.
degree in systems and control and the Ph.D.
degree in control systems from the University of
Manchester Institute of Science and Technology,
Manchester, U.K., in 1986 and 1989, respectively.
After working as a Lecturer with Ngee Ann
Polytechnic, Singapore, for ten years, he joined
the University of Manchester, Manchester, in 2003,
Zhongnan Wang received the B.S. and M.S. where he is currently a Professor of Control Systems
degrees in mechanical engineering from the with the Department of Electrical and Electronic Engineering. He has authored
Changchun University of Technology, Changchun, the book: Nonlinear and Adaptive Control Systems (IET 2013) and a num-
China, in 2016 and 2020, respectively. He is ber of journal papers. His research interests include nonlinear and adaptive
currently pursuing the Ph.D. degree with the control theory and their applications, more recently network-based control,
School of Mechanical Engineering and Automation, distributed optimization, and distributed machine learning, with applications
Northeastern University, Shenyang, China. to power systems and robotics.
His research interests include vehicle dynamics Prof. Ding has served as an Associate Editor for the IEEE T RANSACTIONS
and intelligent vehicle. ON AUTOMATIC C ONTROL , IEEE C ONTROL S YSTEMS L ETTERS , Journal of
Franklin Institute, and several other journals.

Authorized licensed use limited to: University of Ulsan. Downloaded on June 01,2023 at 03:44:51 UTC from IEEE Xplore. Restrictions apply.