IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO.

11, NOVEMBER 2023 6929

Adaptive Fixed-Time Stabilization for a Class of

Uncertain Nonlinear Systems
Qingtan Meng , Graduate Student Member, IEEE, Qian Ma , Senior Member, IEEE,
and Yang Shi , Fellow, IEEE

Abstract—This article studies the problem of fixed-time sta- methods can only ensure that the closed-loop systems are practically
bilization for a class of nonlinear systems with parameteric un- fixed-time stable. Few studies have been conducted to achieve the
certainties, where the nonlinear functions are constrained by an
unknown linear growth condition. By establishing a new adaptive fixed-time stability of systems. Cao et al. [28] made it possible to
fixed-time stability analysis criterion, it is proved that the closed- stabilize the system in a fixed time by designing exponential control
loop system is fixed-time stable and all the states can be regulated laws. A fixed-time stability condition was given in [29], under which
to zero. In addition, the singularity problem caused by the back- the system is fixed-time stable by a finite-time decentralized controller.
stepping method under fixed-time control can be avoided. Finally,
the effectiveness of the control strategy is verified by a simulation
In [30], a time-varying scale transformation was introduced to realize
example. that the asymptotic stability of the transformed system is equivalent to
the fixed-time stability of the original system. In the abovementioned
Index Terms—Adaptive control, backstepping method, fixed- few exceptions, the upper bounds of nonlinear functions should be
time control, uncertain nonlinear system.
known in advance. It is nontrivial to generalize such results to the case
where the bounds are unknown, mainly due to the lack of explicit adap-
I. INTRODUCTION tive controller design technique and stability analysis criterion. In this
article, we are committed to dealing with the fixed-time control problem
In the past few decades, the controller design of nonlinear systems for a class of uncertain nonlinear systems. The main contribution can
has been studied extensively [1], [2], [3], [4], [5], [6], [7], [8], in which be summarized as follows.
stability is the primary and basic objective. Note that in many practical 1) A novel adaptive fixed-time control framework for a class of
applications, systems need to be stabilized in finite time to meet some nonlinear systems with unknown nonlinear functions and parame-
specific requirements. Hence, finite-time control has received consider- teric uncertainties is designed for the first time. In order to prove
able attention and obtained fruitful results; see, for example, [9], [10], the fixed-time stability under these uncertainties, a new adaptive
[11], [12], and [13]. fixed-time stability analysis criterion is proposed.
Fixed-time control, as an extension of finite-time control, does 2) Different from [26], [29], [28], [30], and [27] where unknown non-
not require the convergence time to be dependent on the initial state linear functions were assumed to be bounded by known functions,
of the system. In the pioneering work [14], the fixed-time control the nonlinear functions with unknown bounds are considered in
technique was first proposed for uncertain linear plants. Since then, this article. Moreover, the singularity problem in the virtual control
by using the adaptive backstepping technique, fixed-time control of design can be avoided effectively.
uncertain nonlinear systems has attracted much attention, such as 3) Most of the adaptive fixed-time control techniques for uncertain
systems with parameteric uncertainty [15], systems with unknown nonlinear systems [20], [21], [22], [23], [24], [25], [26], [27] can
control direction [16] and [17], systems containing input saturation [18], only achieve practical fixed-time stability rather than fixed-time
error-constrained systems [19], just to name a few. stability. In this article, the designed adaptive controller guarantees
Although some developments of fixed-time control have been made that the closed-loop systems are fixed-time stable and all the states
in the literature, the singularity problem was introduced inevitably in can be regulated to zero.
the virtual control design, which may cause systems to escape before
convergence. From this point of view, an error conversion mechanism
II. PRELIMINARIES AND PROBLEM FORMULATION
was established in [20] to confine the position tracking error within a
positive interval to avoid singularities. The authors in [21], [22], [23], In this article, we study the following uncertain nonlinear system:
[24], [25], [26], [27] dealt with the singularity problem by designing
several switched virtual control laws. Notably, most of these control ẋi = xi+1 + fi (x̄i ) + ϑT φi (x̄i ), i = 1, . . . , n − 1

ẋn = u + fn (x) + ϑT φn (x) (1)

Manuscript received 14 July 2022; revised 19 November 2022; ac- y = x1
cepted 3 February 2023. Date of publication 10 February 2023; date
of current version 26 October 2023. This work was supported by the where x = [x1 , . . . , xn ]T ∈ Rn is the system state, x̄i =
NSFC under Grant 62173183. Recommended by Associate Editor M. [x1 , . . . , xi ]T , u ∈ R is the control input, y ∈ R is the system
E. Broucke. (Corresponding author: Qian Ma.)
Qingtan Meng and Qian Ma are with the School of Automation, output, ϑ = [ϑ1 , . . . , ϑm ]T ∈ Rm is an unknown constant parameter
Nanjing University of Science and Technology, Nanjing 210094, China vector, φi (x̄i ) : Ri → Rm is a known smooth vector-valued function
(e-mail: [email protected]; [email protected]). with φi (0) = 0, and fi (x̄i ) : Ri → R is an unknown nonlinear
Yang Shi is with the Department of Mechanical Engineering, Univer- function with fi (0) = 0.
sity of Victoria, Victoria, BC V8W 2Y2, Canada (e-mail: [email protected]).
Color versions of one or more figures in this article are available at
Definition 1 ([14], [31]): For the system in (1), assume that the
https://doi.org/10.1109/TAC.2023.3244151. origin is an equilibrium point. The equilibrium x = 0 of the system
Digital Object Identifier 10.1109/TAC.2023.3244151 is said to be finite-time stable if it is asymptotically stable and any

0018-9286 © 2023 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: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
6930 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023

solution x(t, x0 ) of (1) reaches the origin at some finite moment, i.e., Step 1: For the system in (1), define the Lyapunov function V1 =
1 2
x(t, x0 ) = 0, t ≥ T (x0 ), where T (x0 ) : Rn → R+ ∪ 0 is the conver- x
2 1
+ 2Γ1 1 θ̃12 , where Γ1 > 0 is an adaptive gain and θ̃1 = θ1 − θ̂1 is
gence time function. The equilibrium x = 0 of the system is said to be the estimate error. Then, the time derivative of V1 is calculated as
practically finite-time stable if there exists a constant  > 0 such that
1 ˙
any solution x(t, x0 ) satisfying x(t, x0 ) <  for all t ≥ T (x0 ). The V̇1 = z1 (x2 − α2 ) + z1 α2 + z1 f1 + z1 ϑT φ1 − θ̃1 θ̂1 . (6)
Γ1
equilibrium x = 0 of the system is said to be fixed-time stable if it is
finite-time stable and the convergence time function T (x0 ) is bounded, It follows from Assumption 1 and Young’s inequality that
i.e., there exists a finite constant Tmax ∈ R+ such that T (x0 ) ≤ Tmax 2+q
for all x0 ∈ Rn . The equilibrium x = 0 of the system is said to be z1 f1 ≤ δz12 ≤ k11 δ 2 z12+q + Δ11 ≤ k11 θ1 z12+q + Δ11 (7)
practically fixed-time stable if it is practically finite-time stable and
where k11 and Δ11 are positive design constants, respectively. Ac-
T (x0 ) ≤ Tmax for all x0 ∈ Rn .
cording to [32, Lemma 2.5], for i = 1, . . . , n, one has |φi (x̄i )| ≤
In order to achieve the control goal, the following assumption and i
lemmas are crucial. For brevity, the proof of Lemma 3 is given in the j=1 φ̄j (xj )|xj | with φ̄1 , . . . , φ̄i being known nonnegative smooth
functions. Hence, it can be verified that
Appendix.
Assumption 1: For i = 1, . . . , n, there exists an unknown constant z1 ϑT φ1 ≤ z12 |ϑT |φ̄1 ≤ k12 |ϑT |
2+q
2 z12+q + Δ12
δ > 0 such that
≤ k12 θ1 z12+q + Δ12 (8)
|fi (x̄i )| ≤ δ(|x1 | + · · · + |xi |). (2)
where k12 ≥ 0 is a smooth function with respect to x1 and Δ12 is a pos-
Lemma 1 ([9]): For the system in (1), if there is a positive def- itive constant. Let k1 = k11 + k12 and Δ1 = Δ11 + Δ12 . Substituting
inite and proper function V (x) such that V̇ (x) ≤ −cV α (x), where (7) and (8) into (6) yields
c > 0 and α ∈ (0, 1), then the system is finite-time stable and the finite
convergence time can be upper bounded by T (x0 ) ≤ V (x(0))
1−α
1 ˙
c(1−α)
. V̇1 ≤ z1 z2 + z1 α2 + k1 θ1 z12+q + Δ1 − θ̃1 θ̂1 . (9)
Lemma 2 ([29]): For any x1 , x2 , . . . , xn and p, one has Γ1
Let c1 > 0 be a constant. Design the virtual control law as
(|x1 | + · · · + |xn |)p ≤ max{np−1 , 1}(|x1 |p + · · · + |xn |p ).

Lemma 3: For the system in (1), if there exists a positive definite α2 = − z11+q l1 , l1 ≥ c1 + k1 θ̂1 (10)
and proper function V (x) such that V̇ (x) ≤ −μW q (x), where μ > 0, which leads to
q > 1, and W (x) ≥ cV (x) for some constant c > 0, then V (x) ≤ 1  
when t ≥ T1 = (q−1)c 1 1 ˙
qμ . V̇1 ≤ −c1 z12+q + z1 z2 + θ̃1 τ1 − θ̂1 + Δ1
Remark 1: Assumption 1 indicates that the unknown nonlinear Γ1
function fi (x̄i ) has an unknown bound. It is worth pointing out that
where τ1 = k1 z12+q .
when δ and ϑT are unknown, the fixed-time stabilization problem
Step k: For simplicity, we summarize the inductive step in the
for the system in (1) becomes much more involved due to the lack
following proposition.
of effective adaptive controller design technique and stability analysis
Proposition 1: Suppose at step k − 1, there exist a series of
criterion. This deficiency also results in that only practical fixed-time
virtual control laws α2 = −z11+q l1 (x1 , θ̂1 ), α3 = −z21+q l2 (x̄2 , θ̂1 ),
stability rather than fixed-time stability can be realized in most existing 1+q
. . . , αk = −zk−1 lk−1 (x̄k−1 , θ̂1 ) making the Lyapunov function
literature. Lemma 3 can help us overcome this difficulty.
Vk−1 = 12 i=1 zi2 + 2Γ1 1 θ̃12 satisfy
k−1

III. ADAPTIVE FIXED-TIME CONTROLLER DESIGN 

k−1

In this section, we propose a backstepping procedure to design the V̇k−1 ≤ − ck−1 zi2+q + zk−1 zk + Δk−1
i=1
adaptive fixed-time controller for the system in (1). To be clear, for a
certain function h(x), we occasionally use h to represent it. 
k−1  
∂αi 1 ˙
First, introduce the state coordinate transformation as + θ̃1 + Γ1 zi τk−1 − θ̂1 (11)
i=2 ∂ θ̂1 Γ1
z1 = x1 , zi = xi − αi (x̄i−1 , θ̂1 ), i = 2, . . . , n (3)
where ck−1 > 0 and Δk−1 > 0 are design constants, and l1 > 0,
where α2 , . . . , αn are virtual control laws to be designed and θ̂1 is an . . . , lk−1 > 0 and τk−1 ≥ 0 are C 1 design functions. Then, for the kth
estimate of Lyapunov function Vk = Vk−1 + 12 zk2 , there exists a virtual control law
 2+q 2+q

θ1  max δ, |ϑT |, δ 2+q , |ϑT |2+q , δ 2 , |ϑT | 2 (4) αk+1 = −zk1+q lk (x̄k , θ̂1 ) (12)

with q > 0 being a known ratio of an even and an odd integer, respec- such that
tively. Next, define the Lyapunov function W1 and the state set Ω as 
k
V̇k ≤ − ck zi2+q + zk zk+1 + Δk
   2+q
2  i=1
1 2
n
2Δn  
W1 = z , Ω= z : W1 ≤ max ,1 (5) 
k
2 i=1 i C ∂αi 1 ˙
+ θ̃1 + Γ1 zi τk − θ̂1 (13)
i=2 ∂ θ̂1 Γ1
where z = [z1 , . . . , zn ] , C and Δn are positive constants to be
T

designed. Then, we divide the fixed-time controller design into the where lk > 0 is a C 1 function, 0 < ck < ck−1 is a design constant, and
following two cases. τk = τk−1 + kk zk2+q with kk ≥ 0 being a C 1 function.
Case 1: For z ∈/ Ω. Proof: See the Appendix. 

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023 6931

Step n: By strictly following the design process of step k, one can x 1 1

where Ei = βii [s 1+(i−1)p − βi1+(i−1)p ]1−(i−1)p ds, dk−1 > 0 is a de-
get the adaptive controller as
sign constant, λ1 > 0, . . . , λk−1 > 0 are C 1 design functions, and
u1 = αn+1 = −zn1+q ln (x, θ̂1 ) (14) πk−1 ≥ 0 is a continuous function. Then, for the kth Lyapunov function
Uk = Uk−1 + Ek , there exists a virtual control law
˙ 
n
θ̂1 = Γ1 τn = Γ1 ki zi2+q , θ̂1 (0) ∈ (0, θ̂10 ] (15) βk+1 = −e1+kp λk (x̄k , θ̂2 ) (21)
k
i=1

which makes the Lyapunov function such that


k
1 2
n 1−(k−1)p
1 2 1 2 U̇k ≤ − dk e2+p + ek (xk+1 − βk+1 )
Vn = W 1 + θ̃1 = zi + θ̃ (16) i
2Γ1 2 i=1 2Γ1 1 i=1


k  
satisfy ∂Ei 1 ˙
+ θ̃2 − Γ2 πk − θ̂2 (22)

n
i=2 ∂ θ̂2 Γ2
V̇n ≤ − cn zi2+q + Δn (17)
i=1
where 0 < dk < dk−1 is a design constant, λk > 0 is a C 1 function, and
πk = πk−1 + ρk (x̄k , θ̂2 )e2+p
k with ρk ≥ 0 being a continuous function
where ln > 0 is a C 1 function, and c1 > c2 > · · · > cn > 0, Δn > 0,
and ρk (0, θ̂2 ) = 0.
and θ̂10 > 0 are design constants.
Proof: See the Appendix. 
Case 2: For z ∈ Ω
Step n: By strictly following the design process of step k, we can get
Before designing the adaptive controller for Case 2, we introduce
the adaptive controller as:
the following coordinate transformation:
1 1 u2 = βn+1 = −e1+np
n λn (x, θ̂2 ) (23)
1+(i−1)p 1+(i−1)p
e 1 = x1 , e i = xi − βi (x̄i−1 , θ̂2 ) (18)
˙ 
n

where i = 2, . . . , n, β2 , . . . , βn are virtual control laws to be designed, θ̂2 = Γ2 πn = Γ2 ρi e2+p

i , θ̂2 (0) ∈ (0, θ̂20 ] (24)
i=1
and θ̂2 is an estimate of
2+p 2+p
which makes the Lyapunov function
θ2  max δ, |ϑT |, |ϑT | 1−(i−2)p , δ 1−(i−2)p , i = 1, . . . , n
1 2 
n
1 2
Un = e1 + Ei + θ̃ (25)
2 2Γ2 2
with − n1 < p < 0 being a known ratio of an even integer and an odd i=2
integer. satisfy
Step 1: Consider the Lyapunov function U1 = 12 x21 + 2Γ1 2 θ̃22 , where

n
Γ2 > 0 is an adaptive gain and θ̃2 = θ2 − θ̂2 is the estimate error. Then, U̇n ≤ − dn e2+p
i (26)
the time derivative of U1 is calculated as i=1

1 ˙ where d1 > · · · > dn > 0 and θ̂20 > 0 are design constants, and λn >
U̇1 = e1 x2 + e1 f1 + e1 ϑT φ1 − θ̃2 θ̂2
Γ2 0 is a C 1 function. The whole design procedures are shown in Fig. 1.
1 ˙ Remark 2: Based on the definition of virtual control laws
≤ e1 (x2 − β2 ) + e1 β2 + (ρ11 + ρ12 )θ2 e2+p
1 − θ̃2 θ̂2 αi (x̄i−1 , θ̂1 ), i = 2, . . . , n, in (10) and (12), it is obvious that αi is
Γ2
a C 1 function with respect to x̄i−1 and θ̂1 since q > 0 and li−1 is
where ρ11 = e−p −p
1 ≥ 0 and ρ12 = φ̄1 e1 ≥ 0 are continuous functions a C 1 function. Hence, in the design process of virtual control laws
with respect to x1 . Let ρ1 = ρ11 + ρ12 with ρ1 (0) = 0 and d1 > 0 be of Case 1, it does not suffer from the singularity problem. For the
a design constant. Then, define the virtual control law virtual control laws βi (x̄i−1 , θ̂2 ), i = 2, . . . , n, in Case 2, we obtain
β2 = − e1+p λ1 , λ1 ≥ d1 + ρ1 θ̂2 (19) that βi cannot be differentiable with respect to x̄i−1 at ei−1 = 0 due
1
to p < 0 [for example, β2 in (19)]. This means that it will inevitably
such that introduce the singularities if we involve the derivative of βi in the virtual
  1
1 ˙ control design. However, it is not hard to conclude that βi1+(i−1)p is a
U̇1 ≤ −d1 e2+p
1 + e1 (x2 − β2 ) + θ̃2 π1 − θ̂2
Γ2 C 1 function with respect to x̄i−1 and θ̂2 . By designing the Lyapunov
where π1 = ρ1 e2+p and λ1 > 0 is chosen as a C 1 function. function (25) in integral form, we only use the partial derivative of
1 1
Step k: Similar to Case 1, we also summarize the inductive step in βi1+(i−1)p instead of βi , which also avoids introducing the singularity
the following proposition. problem. In this way, the whole controller design process is singularity
Proposition 2: Suppose at step k − 1, there exist a se- free.
ries of virtual control laws β2 = −e1+p 1 λ1 (x1 , θ̂2 ), . . . , βk = Remark 3: Recall that the control methods, such as in [21], [22],
1+(k−1)p
−ek−1 λk−1 (x̄k−1 , θ̂2 ) making the Lyapunov function Uk−1 = [23], [24], and [25], can avoid the singularity problem in the virtual
1 2
e + i=2
2 1
k−1
Ei + 2Γ1 2 θ̃22 satisfy control design, but it can only ensure the practical fixed-time stability
of the closed-loop system. There are two reasons for this conservative

k−1
1−(k−2)p
result: one is from the virtual control law and the other is from the
U̇k−1 ≤ − dk−1 e2+p
i + ek−1 (xk − βk ) adaptive update law. In the literature, the following form of virtual
i=1
control law is commonly used:

k−1  
∂Ei 1 ˙ zi α̌i2
+ θ̃2 − Γ2 πk−1 − θ̂2 (20) αi = −  (27)
i=2 ∂ θ̂2 Γ2 zi2 α̌i2 + ε2i

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
6932 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023

2+q q 2+q
where C = 2 2 n− 2 cn . Since z ∈ / Ω, it is easy to get CW1 2
>
2Δn . Then, (30) can be rewritten as
C 2+q
V̇n ≤ − W 2 . (31)
2 1
It follows from (5) that 0 ∈ Ω. Therefore, in (16), before z enters the set
Ω, we can find some constant 0 < σ1 < 1 such that 2Γ1 1 θ̃12 ≤ σ1 Vn <
Vn , i.e., W1 ≥ (1 − σ1 )Vn . Using Lemma 3, we have Vn ≤ 1 when t ≥
4
T1 = 2+q . Therefore, one has W1 ≤ Vn ≤ 1, i.e., the states
Cq(1−σ1 ) 2
that are not in the set Ω will converge to Ω in T1 .
In Case 2, according to (25), we define W2 = 12 e21 + n
i=2 Ei and
2
Q= n i=1 ei . Then, (26) can be rewritten as


n
2+p
U̇n ≤ −dn e2+p
i ≤ −dn Q 2 ≤0 (32)
i=1

which implies that x and θ̂2 are bounded. Furthermore, we have

1 2 
n
W2 ≤ e + |xi − βi ||ei |1−(i−1)p ≤ ω1 Q (33)
2 1 i=2

2+p 2+p 2+p

n−1
where ω1 = 2 n . Then, one has W2 2 ≤ ω1 2 Q 2 . It fol-
2+p p
lows from (24) that πn ≤ ρQ 2 , where ρ = n− 2 n i=1 ρi and
Fig. 1. Design procedures. ˙
ρ(0, θ̂2 ) = 0. By (24), θ̂2 is positive due to θ̂2 ≥ 0 and θ̂2 (0) > 0.
Without the loss of generality, we suppose θ̂2 ∈ (0, ], where  is a
constant with θ̂2 (0) and x(0). Combining (25), (32), and (33), one has
where α̌i is an intermediate variable and εi is a positive constant. Then, 2+p
by substituting this virtual control law (27) into the derivative of the Ẇ2 ≤ −dn Q 2 + θ̃2 πn
corresponding Lyapunov function candidate (i.e., V̇ ), the following  
2
dn 2+p 2+p dn 2+p 2( + |θ2 |)
inequality must be involved: ≤ − ω1 W 2 2 − Q 2 1− ρ . (34)
2 2 dn
zi2 α̌i2
zi αi ≤ −  2 2
≤ εi − zi α̌i (28) Define W3 (x, θ̂2 ) = 2(+|θ2 |)
ρ. It follows from ρ(0, θ̂2 ) = 0 that
zi α̌i + ε2i dn
W3 (0, θ̂2 ) = 0. Hence, we can find a constant η > 0 such that for any
which leads to V̇ ≤ −c1 V p − c2 V q + ρ (c1 > 0, c2 > 0, 0 < p < (x, θ̂2 ) ∈ Φ = {(x, θ̂2 ) : W2 (x, θ̂2 ) ≤ η}, W3 < 1. Correspondingly,
1, q > 1, ρ > 0). On the other hand, the commonly used adaptive 2+p
Q 2 [1 − 2(+|θ 2 |)
ρ] ≥ 0, i.e., it will be always in Φ if (x, θ̂2 ) ∈ Φ.
˙ dn
update law is θ̂1 = Γ1 τn − ς1 θ̂ − ς2 θ̂3 (ς1 > 0, ς2 > 0), which will If the initial condition (x(0), θ̂2 (0)) ∈ Φ, it is easy to deduce that
also contribute to ρ. Therefore, it is too difficult to realize the fixed- all (x, θ̂2 ) ∈ Φ. Then, (34) can be rewritten as
time stability of uncertain nonlinear systems due to the fact that both
2 2+p
the improvement of virtual control law and adaptive update law are dn 2+p
Ẇ2 ≤ − ω1 W 2 2 (35)
indispensable. In this article, based on the designed virtual control laws 2
(10), (12), (14), (19), (21), and (23), the adaptive update laws (15) and
such that W2 is finite-time stable and x converges to 0 in T2 =
(24), and the stability analysis criterion in Lemma 3, we successfully −
p
4W2 2 (x(0),θ̂2 (0))
solve this problem. − 2 . By Definition 1, since W2 (x, θ̂2 ) ≤ η for all
2+p
pdn ω1
p
−
4η 2
IV. STABILITY ANALYSIS (x, θ̂2 ) ∈ Φ, T2 ≤ T2 max = − 2 is a fixed time.
2+p
pdn ω1
In this section, we conclude the main results in Theorem 1.
Theorem 1: For the studied uncertain nonlinear system in (1) with If (x(0), θ̂2 (0)) ∈
/ Φ, then we first estimate its maximum arrival time
Assumption 1, if we design the adaptive controller as T3 to Φ. Before the state enters Φ, one has W2 > η. It is not hard to
 calculate that
˙ 2+q
−zn1+q ln (x, θ̂1 ), θ̂1 = Γ1 n i=1 ki zi , z∈/Ω
u= ˙ 2+p
(29) Un (x(0), θ̂2 (0)) ≥ Un (x(0), θ̂2 (0)) − Un (x(t), θ̂2 (t))
−e1+np
n λ n (x, θ̂2 ), θ̂ 2 = Γ 2
n
ρ
i=1 i ie , z ∈ Ω  t
2+p
then the system in (1) is fixed-time stable. ≥ dn Q 2 (x(s), θ̂2 (s))ds
0
Proof: In Case 1, according to (17) and Lemma 2, it is not difficult  t 2 2+p
to deduce
≥ dn ω12+p W2 2
(x(s), θ̂2 (s))ds

n 2+q 0
V̇n ≤ − cn zi2+q + Δn ≤ −CW1 2
+ Δn (30) 2 2+p
i=1 ≥ dn ω12+p η 2 t. (36)

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023 6933

Fig. 2. Responses of (39) with [x1 (0), x2 (0), θ̂1 (0), θ̂2 (0)] = Fig. 3. Responses of (39) with [x1 (0), x2 (0), θ̂1 (0), θ̂2 (0)] =
[−1.5, 9, 0.1, 0.3]. [0.8, −1, 0.5, 1].

Hence, (x, θ̂2 ) will enter Φ in ẋ2 = u − δ2 x2 + |x1 | + ϑT [x1 , x2 ]T (39)

2Γ2 W2 (x(0), θ̂2 (0)) + θ̃22 (0) where δ1 and δ2 are unknown constants and ϑ ∈ R2 is an unknown
T3 = 2 . (37)
2Γ2 dn ω1 2+p
η
2+p
2
constant parameter vector. Then, it is easy to calculate that φ̄1 = 1.2,
φ̄2 = 1, and δ ≥ max{|δ1 |, |δ2 |, 1} under Assumption 1. By Section
Then, based on the construction of W1 and Ω, z ∈ Ω indicates that III, we design
W1 has a constant upper bound, also means that there exists a con-
stant a > 0 such that |x| < a. In addition, it follows from (24) that z1 = x1 , z2 = x2 − α2
θ̂2 (0) ≤ θ̂20 . Then, there must exist a positive constant A such that  24 24 12 12

W2 (x(0), θ̂2 (0)) ≤ A. Therefore, we have θ1  max δ, |ϑT |, δ 11 , |ϑT | 11 , δ 11 , |ϑT | 11
 24 
2Γ2 W2 (x(0), θ̂2 (0)) + θ̃22 (0) ˙ 24

T3 = ≤ T3 max (38) θ̂1 = Γ1 0.939z111 + k2 z211

2
2+p
2Γ2 dn ω12+p η 2 13
α2 = − z111 l1 , l1 ≥ c1 + 0.939θ̂1
Γ2 A+θ2 2 +θ̂ 2
where T3 max = 2
20
2+p
is a fixed time. After the reaching 13 13
2+p
Γ 2 dn ω1 η 2 u1 = − z211 l2 , l2 ≥ c2 + 0.939Γ1 k2 z2 z111 + k2 θ̂1 + k27
time T3 , the state will be always in Φ. It will further take T2 max to 11 11
arrive at zero. Thus, in Case 2, x becomes zero in a fixed time T4 ≤ e1 = x1 , e2 = x29 − β29
T2 max + T3 max . In conclusion, under the adaptive controller (29), x  20 20 20 20

converges to zero and the adaptive parameters θ̂1 and θ̂2 tend to positive θ2  max δ, |ϑT |, δ 9 , |ϑT | 9 , |ϑT | 11 , δ 11
constants in t ≤ T1 + T4 , i.e., the system in (1) is fixed-time stable. The
˙ 20
proof of Theorem 1 is complete.  θ̂2 = Γ2 (2.2e21 + ρ2 e211 )
Remark 4: For the uncertain system in (1), our control objective is 9
to design a singularity-free control scheme to ensure that the system β2 = − e111 λ1 , λ1 ≥ d1 + (2.2 + 2.2e1 )θ̂2
in (1) is fixed-time stable rather than practically fixed-time stable. In 7
order to achieve this objective, there are two major difficulties. One u2 = − e211 λ2 , λ2 ≥ d2 + ρ21 + ρ25 + ρ27 + ρ28 + ρ2 θ̂2
difficulty is how to design a fixed-time control scheme that avoids
u1 , z ∈
/Ω
singularities. The other difficulty arises from how to ensure that the u=
u2 , z ∈ Ω
designed control laws and the adaptive laws make the system in (1)
fixed-time stable rather than practically fixed-time stable. It should be where c1 = 2.5, c2 = 0.5; Γ1 = 5, Γ2 = 3, d1 = 2.5, d2 = 0.5, k2 ,
noted that without the proper design of either the control law or the k27 , ρ21 , ρ25 , ρ27 , ρ28 , and ρ2 are known positive functions, Δ1 = 1.1,
adaptive law, the desired results cannot be obtained. In Remark 2 and Δ2 = 4.2, and C = 1. In simulation, choose ϑ1 = 1, ϑ2 = 1, δ1 = 1,
the proof of Theorem 1, we show that the abovementioned difficulties δ2 = 1, and δ = 1. From the selected parameters, we can obtain the state
can be overcome simultaneously based on the designed control scheme set Ω as Ω = { 12 z12 + 12 z22 ≤ 7.035}. Then, we select two different
in this article.
initial conditions as [x1 (0), x2 (0), θ̂1 (0), θ̂2 (0)] = [−1.5, 9, 0.1, 0.3]
and [x1 (0), x2 (0), θ̂1 (0), θ̂2 (0)] = [0.8, −1, 0.5, 1]. First, for the ini-
V. SIMULATION EXAMPLE tial value [x1 (0), x2 (0), θ̂1 (0), θ̂2 (0)] = [−1.5, 9, 0.1, 0.3], we can
Consider the following uncertain nonlinear system: deduce that [z1 (0), z2 (0)] ∈ / Ω. The responses of state variables x1
and x2 and the adaptive parameters θ̂1 and θ̂2 are shown in Fig. 2.
ẋ1 = x2 + sin(δ1 x1 ) + ϑT [x1 , 0.5x1 ]T Then, it follows from the initial value [x1 (0), x2 (0), θ̂1 (0), θ̂2 (0)] =

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
6934 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023

[0.8, −1, 0.5, 1] that [z1 (0), z2 (0)] ∈ Ω. Correspondingly, the simula- where kk3 ≥ 0 is a C 1 function and Δk3 is a positive design constant.
tion results are shown in Fig. 3. At this point, all cases of the initial Since αk is C 1 , it follows from the construction of αk and [32, Lemma
conditions have been taken into account, and our control strategy is 2.2] that there exists a nonnegative C 1 function k̄k4 such that | ∂αk
∂xi
|≤
valid from Figs. 2 and 3. k̄k4 . From the definition of ẋi and (3), one has

VI. CONCLUSION 
k−1
∂αk 2c̄k  2+q
k−1
− ẋi zk ≤ zi + (kk4 + kk5 θ1 )zk2+q + Δk4
In this article, we study the adaptive fixed-time control problem for i=1
∂x i 5 i=1
a class of uncertain nonlinear systems. The uncertainties are repre- (45)
sented by uncertain parameters and unknown nonlinear functions with
unknown bounds. In order to solve this problem, a novel framework is where kk4 ≥ 0 and kk5 ≥ 0 are C 1 functions, and Δk4 > 0 is a positive
proposed for the fixed-time controller design and stability analysis. It is constant. Similar to (45), from the definition of τk−1 , we have
proved that the singularity problem of virtual control laws is avoided as
well as the closed-loop system is fixed-time stable. Finally, a simulation
3c̄k  2+q
k−1
∂αk
example is given to demonstrate the effectiveness of this framework. − Γ1 zk τk−1 ≤ z + kk6 zk2+q (46)
∂ θ̂1 10 i=1 i
APPENDIX
where kk6 ≥ 0 is a C 1 function. Substituting (42)–(46) into (41) yields
Proof of Lemma 3: Since W (x) ≥ cV (x), we have
V̇ (x) ≤ −μcq V q (x) 
k−1
V̇k ≤ − ck zi2+q + zk zk+1 + zk αk+1 + kk θ1 zk2+q
which implies that i=1
 
t
V̇ (x) t ∂αk ˙ ∂αk
dt ≤ −cq μdt. (40) + kk7 zk2+q − zk θ̂1 + Γ1 zk τk−1
0 V q (x) 0 ∂ θ̂1 ∂ θ̂1
By computing the two sides of (40), we have 
k−1  
∂αi 1 ˙
1 1 + θ̃1 + Γ1 zi τk−1 − θ̂1 + Δk (47)
V 1−q (x(t)) − V 1−q (x(0)) ≤ −cq μt. i=2 ∂ θ̂1
Γ1
1−q 1−q
Therefore, it is easy to get the following inequality: where Δk = Δk−1 + Δk1 + Δk2 + Δk3 + Δk4 , kk = kk2 + kk3 +
1 1 kk5 , and kk7 = kk1 + kk4 + kk6 . Design the virtual control law as
V q−1 (x) ≤ ≤ .
(q − 1)cq μt + V 1−q (x(0)) (q − 1)cq μt
αk+1 = − zk1+q lk (x̄k , θ̂1 )
1
By defining T1 = (q−1)c q μ and for ∀t ≥ T1 , we have V
q−1
(x) ≤ 1,
i.e., V (x) ≤ 1. The proof is complete.  
k
∂αi
lk ≥ ck + kk θ̂1 + kk7 − Γ1 kk zi (48)
Proof of Proposition 1: Calculate the time derivative of Vk as ∂ θ̂1
 i=2
∂αk ∂αk
V̇k ≤ V̇k−1 + Γ1 zk τk−1 − Γ1 zk τk−1 + zk xk+1 which leads to (13). The proof of Proposition 1 is complete.
∂ θ̂1 ∂ θ̂1
Proof of Proposition 2: Calculate the time derivative of Uk as
 ∂αk
k−1 
∂αk ˙
+ f k + ϑT φk − ẋi − θ̂1 . (41) 1−(k−1)p ∂Ek ˙
i=1
∂xi ∂ θ̂1 U̇k ≤ U̇k−1 + ek (xk+1 + fk + ϑT φk ) + θ̂2
∂ θ̂2
We will deal with the right side of (41). Define c̄k = ck−1 − ck . First,
using the fact that |zk−1 | ≤ 1 + |zk−1 |1+q , one gets 
k−1
∂Ek
+ (xi+1 + fi + ϑT φi )
c̄k 2+q ∂x i
zk−1 zk ≤ z + kk1 zk2+q + Δk1 (42) i=1
10 k−1

k−1
∂Ek ˙
where kk1 and Δk1 are positive design constants. Second, it follows ≤ − dk−1 e2+p + ek−1
1−(k−2)p
(xk − βk ) + θ̂2
i
from Assumption 1 and (3) that i=1 ∂ θ̂2

k

k−1  
zk fk ≤ δ|zk | (1 + |zi |1+q + li−1 |zi−1 |1+q ) ∂Ei 1 ˙
+ θ̃2 − Γ2 πk−1 − θ̂2
i=1 i=2 ∂ θ̂2 Γ2

c̄k  2+q 1−(k−1)p

k−1
+ ek (xk+1 + fk + ϑT φk )
≤ z + kk2 θ1 zk2+q + Δk2 (43)
10 i=1 i

k−1
∂Ek
1
where kk2 ≥ 0 is a C function and Δk2 is a positive design constant. + (xi+1 + fi + ϑT φi ). (49)
i=1
∂x i
Third, we have

k
Define d¯k = dk−1 − dk . By Young’s inequality, we have
zk ϑT φk ≤ φ̄i (1 + |zi |1+q + li−1 |zi−1 |1+q )|zk ||ϑT |
i=1 1−(k−2)p
ek−1 (xk − βk ) ≤ 2−(k−1)p |ek−1 |1−(k−2)p |ek |1+(k−1)p
c̄k 
k−1
≤ zi2+q + kk3 θ1 zk2+q + Δk3 (44) d¯k 2+p
10 ≤ e + ρk1 e2+p (50)
i=1 12 k−1 k

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023 6935

where ρk1 > 0 is a design constant. Then, by Assumption 1, it is easy where ρk8 ≥ 0 is a continuous function with ρk8 (0, θ̂2 ) = 0. Substi-
to deduce tuting (50)–(56) into (49) yields
1−(k−1)p

k
(i−k−1)p
ek fk ≤ δ ek |ek |1−(i−2)p 
k−1
1−(k−1)p
i=1
U̇k ≤ − dk e2+p
i + ek (xk+1 − βk+1 )
i=1
× (|ei |1+(i−1)p + λi−1 |ei−1 |1+(i−1)p )  
1−(k−1)p

k−1
∂Ei 1 ˙
+ ek βk+1 + θ̃2 − Γ2 πk−1 − θ̂2
d¯k  2+p
k−1
∂ θ̂2 Γ2
≤ e + ρk2 (x̄k , θ̂2 )θ2 e2+p (51) i=2
12 i=1 i k

∂Ek 
k
∂Ei ∂Ek ˙
where ρk2 ≥ 0 is a continuous function and ρk2 (0, θ̂2 ) = − Γ2 πk−1 − Γ2 ρk e2+p
k + θ̂2
∂ θ̂2 ∂ θ̂2 ∂ θ̂2
0. According to the definition of ei , we have |φk | ≤ i=2
1+(i−1)p
k
i=1 |φ̄i xi | ≤
k
i=1 φ̄i (|ei | + λi−1 |ei−1 |1+(i−1)p ) ≤ + (ρk1 + ρk5 + ρk7 + ρk8 + ρk θ2 ) e2+p
k . (57)
k 1+(k−1)p
ρ̄k3 (x̄k , θ̂2 ) i=1 |ei | , where ρ̄k3 ≥ 0 is a continuous
function and ρ̄k3 (0, θ̂2 ) = 0. Then, by using Young’s inequality, one From [32, Lemma 2.2], we can design the virtual control law as
has
βk+1 = − e1+kp
k λk (x̄k , θ̂2 )
1−(k−1)p T

k
ek ϑ φk ≤ ρ̄k3 e−p
k |ϑ ||ek |
T 1−(k−2)p
|ei |1+(k−1)p λk ≥ dk + ρk1 + ρk5 + ρk7 + ρk8 + ρk θ̂2 (58)
i=1

d¯k 2+p which leads to (22), where λk is a positive definite C 1 function. The
≤ e + ρk3 (x̄k , θ̂2 )θ2 e2+p (52) proof of Proposition 2 is complete.
12 1 k

where ρk3 ≥ 0 is a continuous function and ρk3 (0, θ̂2 ) = 0. From the REFERENCES
definition of βk , we have
  [1] M. Krstić, P. V. Kokotovic, and I. Kanellakopoulos, Nonlinear and Adap-
 1
 
 ∂[−βk1+(k−1)p ]  k−1 tive Control Design. Hoboken, NJ, USA: Wiley, 1995.
  ≤ ρk4 (x̄k−1 , θ̂2 ) |ej |−(i−1)p (53) [2] H. Deng and M. Krstić, “Stochastic nonlinear stabilization–I: A backstep-
 ∂xi 
  j=i−1 ping design,” Syst. Control Lett., vol. 32, no. 3, pp. 143–150, 1997.
[3] Z. Pan and T. Basar, “Backstepping controller design for nonlinear stochas-
where ρk4 ≥ 0 is a C 1 function. Define tic systems under a risk-sensitive cost criterion,” SIAM J. Control Optim.,
 xk 1
vol. 37, no. 3, pp. 957–995, 1999.
1
Ēk = [1 − (k − 1)p] [s 1+(k−1)p − βk1+(k−1)p ]−(k−1)p ds [4] W. Lin and C. Qian, “Adaptive control of nonlinearly parameterized
βk systems: A nonsmooth feedback framework,” IEEE Trans. Autom. Control,
vol. 47, no. 5, pp. 757–774, May 2002.
where we calculate Ēk ≤ [1 − (k − 1)p]2−(k−1)p |ek |. Therefore, it is [5] Z. P. Jiang and L. Praly, “Design of robust adaptive controllers for non-
not difficult to calculate that linear systems with dynamic uncertainties,” Automatica, vol. 34, no. 7,
pp. 825–840, 1998.

k−1
∂Ek [6] P. Zhang, T. Liu, and Z.-P. Jiang, “Event-triggered stabilization of a class
(xi+1 + fi + ϑT φi ) of nonlinear time-delay systems,” IEEE Trans. Autom. Control, vol. 66,
i=1
∂x i
no. 1, pp. 421–428, Jan. 2021.
1 [7] C. L. P. Chen, Y.-J. Liu, and G.-X. Wen, “Fuzzy neural network-based

k−1
∂[−βk1+(k−1)p ] adaptive control for a class of uncertain nonlinear stochastic systems,”
≤ Ēk (xi+1 + fi + ϑT φi ) IEEE Trans. Cybern., vol. 44, no. 5, pp. 583–593, May 2014.
i=1
∂xi [8] Y. Huang and Y. Liu, “Practical tracking via adaptive event-triggered
feedback for uncertain nonlinear systems,” IEEE Trans. Autom. Control,
d¯k  2+p
k−1
vol. 64, no. 9, pp. 3920–3927, Sep. 2019.
≤ e + (ρk5 (x̄k , θ̂2 ) + ρk6 (x̄k , θ̂2 )θ2 )e2+p (54) [9] S. P. Bhat and D. S. Bernstein, “Finite-time stability of continuous au-
2 i=1 i k
tonomous systems,” SIAM J. Control Optim., vol. 38, no. 3, pp. 751–766,
2000.
where ρk5 ≥ 0 and ρk6 ≥ 0 are continuous functions with ρk5 (0, θ̂2 ) = [10] Y. Hong, J. Wang, and D. Cheng, “Adaptive finite-time control of nonlinear
ρk6 (0, θ̂2 ) = 0. Then, it can be verified that systems with parametric uncertainty,” IEEE Trans. Autom. Control, vol. 51,
1
no. 5, pp. 858–862, May 2006.
∂Ek ∂[−βk1+(k−1)p ] [11] H. Li, S. Zhao, W. He, and R. Lu, “Adaptive finite-time tracking control
Γ2 πk−1 ≤ Ēk Γ2 πk−1 of full state constrained nonlinear systems with dead-zone,” Automatica,
∂ θ̂2 ∂ θ̂2 vol. 100, pp. 99–107, 2019.
[12] J. X. Zhang and G. H. Yang, “Global finite-time output stabilization of
d¯k  2+p
k−1
nonlinear systems with unknown measurement sensitivity,” Int. J. Robust
≤ e + ρk7 (x̄k , θ̂2 )e2+p (55) Nonlinear Control, vol. 28, no. 16, pp. 5158–5172, 2018.
12 i=1 i k
[13] Y. X. Li, “Finite time command filtered adaptive fault tolerant control for a
class of uncertain nonlinear systems,” Automatica, vol. 106, pp. 117–123,
where ρk7 ≥ 0 is a continuous function with ρk7 (0, θ̂2 ) = 0. Let ρk = 2019.
ρk2 + ρk3 + ρk6 and πk = πk−1 + ρk e2+pk , then, we have [14] A. Polyakov, “Nonlinear feedback design for fixed-time stabilization of
1 linear control systems,” IEEE Trans. Autom. Control, vol. 57, no. 8,

k
∂Ei 
k
∂[−βi1+(i−1)p ] pp. 2106–2110, Aug. 2012.
Γ2 ρk e2+p
k ≤ Ēi Γ2 ρk e2+p
k [15] F. Wang and G. Lai, “Fixed-time control design for nonlinear uncer-
i=2 ∂ θ̂2 i=2 ∂ θ̂2 tain systems via adaptive method,” Syst. Control Lett., vol. 140, 2020,
Art. no. 104704.
d¯k  2+p
k−1
[16] M. Chen, H. Wang, and X. Liu, “Adaptive fuzzy practical fixed-time
≤ e + ρk8 (x̄k , θ̂2 )e2+p (56) tracking control of nonlinear systems,” IEEE Trans. Fuzzy Syst., vol. 29,
6 i=1 i k
no. 3, pp. 664–673, Mar. 2021.

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.
6936 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 68, NO. 11, NOVEMBER 2023

[17] X. Hu, Y.-X. Li, and Z. Hou, “Event-triggered fuzzy adaptive fixed-time [25] J. Ni, Z. Wu, L. Liu, and C. Liu, “Fixed-time adaptive neural network
tracking control for nonlinear systems,” IEEE Trans. Cybern., vol. 52, control for nonstrict-feedback nonlinear systems with deadzone and output
no. 7, pp. 7206–7217, Jul. 2022, doi: 10.1109/TCYB.2020.3035779. constraint,” ISA Trans., vol. 97, pp. 458–473, 2020.
[18] X.-N. Shi, Z.-G. Zhou, D. Zhou, and R. Li, “Event-triggered fixed-time [26] Q. Meng, Q. Ma, and Y. Shi, “Fixed-time stabilization for nonlinear
adaptive trajectory tracking for a class of uncertain nonlinear systems with systems with low-order and high-order nonlinearities via event-triggered
input saturation,” IEEE Trans. Circuits Syst. II, Exp. Briefs, vol. 68, no. 3, control,” IEEE Trans. Circuits Syst. I. Reg. Papers, vol. 69, no. 7,
pp. 983–987, Mar. 2021. pp. 3006–3015, Jul. 2022, doi: 10.1109/TCSI.2022.3164552.
[19] Q. Zhou, P. Du, H. Li, R. Lu, and J. Yang, “Adaptive fixed-time control [27] H. Du, G. Wen, D. Wu, Y. Cheng, and J. Lü, “Distributed fixed-time
of error-constrained pure-feedback interconnected nonlinear systems,” consensus for nonlinear heterogeneous multi-agent systems,” Automatica,
IEEE Trans. Syst., Man, Cybern. Syst., vol. 51, no. 10, pp. 6369–6380, vol. 113, 2020, Art. no. 108797.
Oct. 2021. [28] Y. Cao, C. Wen, S. Tan, and Y. Song, “Prespecifiable fixed-time control
[20] Y. Pan, P. Du, H. Xue, and H.-K. Lam, “Singularity-free fixed-time fuzzy for a class of uncertain nonlinear systems in strict-feedback form,” Int. J.
control for robotic systems with user-defined performance,” IEEE Trans. Robust Nonlinear Control, vol. 30, no. 3, pp. 1203–1222, 2020.
Fuzzy Syst., vol. 29, no. 8, pp. 2388–2398, Aug. 2021. [29] C. Hua, Y. Li, and X. Guan, “Finite/fixed-time stabilization for nonlin-
[21] X. Jin, “Adaptive fixed-time control for MIMO nonlinear systems ear interconnected systems with dead-zone input,” IEEE Trans. Autom.
with asymmetric output constraints using universal barrier func- Control, vol. 62, no. 5, pp. 2554–2560, May 2017.
tions,” IEEE Trans. Autom. Control, vol. 64, no. 7, pp. 3046–3053, [30] F. Gao, Y. Wu, and Z. Zhang, “Global fixed-time stabilization of switched
Jul. 2019. nonlinear systems: A time-varying scaling transformation approach,”
[22] S. Ling, H. Wang, and P. X. Liu, “Fixed-time adaptive event-triggered IEEE Trans. Circuits Syst. II, Exp. Briefs, vol. 66, no. 11, pp. 1890–1894,
tracking control of uncertain nonlinear systems,” Nonlinear Dyn., vol. 100, Nov. 2019.
no. 4, pp. 3381–3397, 2020. [31] Y. Sun, F. Wang, Z. Liu, Y. Zhang, and C. L. P. Chen, “Fixed-time fuzzy
[23] H. Wang, K. Xu, and J. Qiu, “Event-triggered adaptive fuzzy fixed-time control for a class of nonlinear systems,” IEEE Trans. Cybern., vol. 52,
tracking control for a class of nonstrict-feedback nonlinear systems,” no. 5, pp. 3880–3887, May 2022, doi: 10.1109/TCYB.2020.3018695.
IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 68, no. 7, pp. 3058–3068, [32] X. Zhang, W. Lin, and Y. Lin, “Nonsmooth feedback control of time-delay
Jul. 2021. nonlinear systems: A dynamic gain based approach,” IEEE Trans. Autom.
[24] Y. Xie, Q. Ma, J. Gu, and G. Zhou, “Event-triggered fixed-time practical Control, vol. 62, no. 1, pp. 438–444, Jan. 2017.
tracking control for flexible-joint robot,” IEEE Trans. Fuzzy Syst., vol. 31,
no. 1, pp. 67–76, Jan. 2023, doi: 10.1109/TFUZZ.2022.3181463.

Authorized licensed use limited to: YANGZHOU UNIVERSITY. Downloaded on November 01,2024 at 02:47:42 UTC from IEEE Xplore. Restrictions apply.