Received: 11 May 2017 Revised: 31 October 2018 Accepted: 3 November 2018

DOI: 10.1002/rnc.4423

RESEARCH ARTICLE

Adaptive finite-time control for bilateral teleoperation

systems with jittering time delays

Ziwei Wang1 Zhang Chen2 Yiman Zhang1 Xingyao Yu2 Xiang Wang2 Bin Liang1

1
Department of Automation, Tsinghua
University, Beijing, China Summary
2
Graduate School at Shenzhen, Tsinghua In this paper, we present a robust adaptive control algorithm for a class of bilat-
University, Shenzhen, China
eral teleoperation systems with system uncertainties and jittering time delays.
Correspondence The remarkable feature of jittering delays is that time delays change sharply
Zhang Chen, Graduate School at and randomly. Conventional controllers would fail because jittering time delays
Shenzhen, Tsinghua University, Shenzhen
introduce serious chattering. To address the jittering issue, a novel jittering-free
518055, China.
Email: [email protected] scheme is developed by relaxing and extending the frequently used constant
upper bound. Moreover, an adaptive law was incorporated with the Cheby-
Funding information
National Natural Science Foundation of shev neural network to deal with the system uncertainties. To obtain finite-time
China, Grant/Award Number: 61703228 synchronization performance, a fast terminal sliding mode controller is pro-
and 61673239; Science and Technology
posed through the technique of “adding a power integrator.” With the proposed
Project of Shenzhen, Grant/Award
Number: JCYJ20160428182227081; control scheme, the robust finite-time convergence performance is guaranteed.
The Science and Technology Planning The settling time can be further calculated with the controller parameters. The
Project of Guangdong Province,
simulation and experiment results have demonstrated the effectiveness of the
Grant/Award Number: 2017B010116001
proposed method.

K E Y WO R D S
adaptive control, bilateral control, finite-time control, jittering time delay, teleoperation

1 I N T RO DU CT ION

Teleoperation is an effective approach to expand human's sensibility and operational capability for a remote environ-
ment. Bilateral teleoperation means that a human operator manipulates the master robot while generating command
signals to be sent to the slave robot and interacting with the environment indirectly. As a method of human replacement,
bilateral teleoperation has a wide range of applications, such as space exploration,1-3 underwater robotics,4 telediagnosis,5
telesurgery,6 and nuclear material handling.7
In practice, teleoperation systems inevitably suffer from time delays, which would result in degraded performance or
system instability. Therefore, time delay is a main issue that much research focuses on. Classic passivity-based approaches
were proposed firstly, such as scattering-based method8 and wave-variable scheme.9,10 Many subsequent passivity-based
studies, such as symmetric impedance matching, position drift eliminating, and damping injection schemes, were con-
ducted based on the two methods.11 For more advanced surveys, readers may refer to the work of Hokayem and Spong,12 in
which time-domain passivity and four-channel structure were introduced. High transparency and stability were achieved
by a wave-based time-domain passivity approach13 for a four-channel nonlinear teleoperation system. Position drift and
force chattering were compensated by a wave-variable–based control scheme14 for the bilateral telemanipulation. In fact,
although the time-delay issue in teleoperation systems has been extensively studied, an assumption that time-delay deriva-
tive is bounded was always used in the works of Hua and Liu,15 Hashemzadeh et al,16 and Chen et al.17 Due to complicated

Int J Robust Nonlinear Control. 2018;1–24. wileyonlinelibrary.com/journal/rnc © 2018 John Wiley & Sons, Ltd. 1
2 WANG ET AL.

processing links in the information transmission channel, time delay is very complex, including constant and random
components. Thus, time delay in teleoperation systems is characterized by rapid changes, namely, jittering. Jittering time
delay will introduce mismatched terms into the control loop, resulting in serious undesirable chattering. However, tradi-
tional methods are difficult to solve this problem. To the best knowledge of the authors, rare attempt has been made to
design jittering-free schemes in the framework of continuous nonlinear systems.
As another crucial issue that deserves further research, the tracking performance shows the position and velocity syn-
chronization between the master and slave robots. Steady-state error and settling time are two main aspects of expressing
tracking performance. Many previous studies focused on the convergence problems. In the works of Chopra et al18 and
Nuño et al,19 position tracking controllers were designed for bilateral teleoperation to achieve asymptotical convergence
of the synchronization errors. In the work of Islam et al,20 the synchronization errors of the master and slave manipu-
lators were bounded by the adaptive control scheme in the presence of time-varying delays and constant input forces.
Recently, researchers have taken the settling time problem into account. Zhai and Xia21 proposed a finite-time bilateral
control scheme, achieving the finite-time position tracking when the robots continuously interact with the human oper-
ator or environment in the presence of time-varying delays. Yang et al22 proposed a terminal sliding mode (TSM)–based
controller to obtain the finite-time synchronization performance. It is noticed that the settling time depends on initial
states in the aforementioned approaches. In fact, external disturbances and measurement errors will lead to inexact ini-
tial states, which should be restricted in practical teleoperation tasks. Therefore, a finite-time control scheme excluding
the knowledge of initial values for teleoperation systems is more effective and challenging.
In this paper, a robust adaptive control scheme for a type of bilateral teleoperation systems with jittering delays is
proposed through the technique of “adding a power integrator.” With the proposed control law, the coordination errors
between master and slave manipulators will converge within a finite time. The controller guarantees fast tracking, high
accuracy, and robustness against model uncertainties and jittering delays. The main contributions of this paper are stated
as follows.
1. A novel TSM algorithm is proposed to address the jittering problem. Different from the common assumptions in other
works,23-26 the disturbances caused by jittering time delays are relaxed and extended to be bounded by positive functions
rather than frequently used constant bounds. By flexible parameter settings, the proposed sliding mode method can
reduce the control effort, and the synchronization performance can be guaranteed simultaneously.
2. The settling time is finite and independent of initial values. In addition, an adaptive law is incorporated with Chebyshev
Neural Network (ChNN) to eliminate the unknown system uncertainties and external disturbances. The proposed
approach relaxes the requirements of the exact system model and the bound of the system uncertainties.
The remainder of this paper is organized as follows. In Section 2, the preliminary knowledge of mechanical teleopera-
tion dynamics and the ChNN basis are presented. A nonsingular finite-time controller (FTC) is proposed in Section 3. In
Section 4, simulation and experiment results are shown and discussed to demonstrate the effectiveness of the proposed
control method, followed by the conclusions presented in Section 5.
Notation. Throughout this paper, we use the subscript i = m, s to denote the master and slave manipula-
tors, respectively. For x = [x1 , x2 , … , xn ] ∈ Rn and 𝛾 > 0 , ||x|| represents the Euclidean norm of x, and sig(x)𝛾 =
[sgn(x1 )|x1 |𝛾 , sgn(x2 )|x2 |𝛾 , … , sgn(xn )|xn |𝛾 ]T with sgn(·) being the sign function. I stands for the identity matrix, and 0
is the null matrix. For any matrix A ∈ Rn×n , tr(A)denotes the trace of A.

2 PROBLEM FO RMULATION AND PRELIMINARIES

Consider a class of nonlinear teleoperation systems composed of n-DOF revolute-joint manipulators. The dynamics can
be expressed as
{ ( ) ( ) ( )
M m qm q̈ m + Cm qm , q̇ m q̇ m + gm qm = F m (t) + F h (t)
( ) ( ) ( ) (1)
M s qs q̈ s + Cs qs , q̇ s q̇ s + gs qs = F s (t) + F e (t) ,
where qi ∈ Rn is the joint position of the master and slave robots; Mi (qi ) ∈ Rn×n is the inertia matrix; Ci (qi , q̇ i ) ∈ Rn×n is
the Coriolis and centrifugal effect matrix; gi (qi ) ∈ Rn is the gravitational force; Fi ∈ Rn is the control torque; Fh ∈ Rn is
the torque of the human operator imposed on the master device; and Fe ∈ Rn is the contact torque when the slave robot
interacts with the environment. Two properties of the dynamic model are given as follows.
WANG ET AL. 3

Property 1. Mi (qi ) is positive definite such that

{ ( )} ( ) { ( )}
0 < 𝜆min M i qi I ≤ M i qi ≤ 𝜆max M i qi I < +∞, (2)
where 𝜆min {Mi (qi )} and 𝜆max {Mi (qi )} are the minimum and maximum eigenvalues of Mi (qi ), respectively.

Property 2. M ̇ i (qi ) − 2Ci (qi , q̇ i )is skew symmetric.

The states of system (1) are augmented to be q = [qm , qs ]T and q̇ = [q̇ m , q̇ s ]T . The desired signals of the dual robots
are denoted as qd = [qm (t − ΔTs ), qs (t − ΔTm )]T and q̇ d = [q̇ m (t − ΔTs ) , q̇ s (t − ΔTm )]T , where ΔTm and ΔTs are the
forward and backward jittering delays, respectively. Thus, the tracking errors can be rewritten as follows:
em = qm − qs (t − ΔTs ) , es = qs − qm (t − ΔTm ) . (3)
Since the gravity terms can be canceled out by the feed-forward control input, the gravitational force is not
considered in the later analysis. Thus, the system is simplified to be
( ) ( )
M m qm q̈ m (t) + Cm qm , q̇ m q̇ m (t) = F m (t) + F h (t) (4)
( ) ( )
M s qs q̈ s (t) + Cs qs , q̇ s q̇ s (t) = F s (t) + F e (t) . (5)
In practical bilateral teleoperation, it is difficult to obtain the extra dynamic model because of unknown external
disturbances and parameter uncertainties. Thus, systems (1), (2) can be expressed as
( ) ( )
M m0 qm q̈ m (t) + Cm0 qm , q̇ m q̇ m (t) = F m (t) + Pm + F h (t) (6)
( ) ( )
M s0 qs q̈ s (t) + Cs0 qs , q̇ s q̇ s (t) = F s (t) + Ps + F e (t) , (7)
where Mi = Mi0 + ΔMi , Ci = Ci0 + ΔCi with Mi0 and Ci0 denoting the nominal part of Mi and Ci , respectively. ΔMi
and ΔCi represent the uncertain part of corresponding matrices. In addition, we can obtain the lumped uncertainty as
Pm = −ΔM m q̈ m (t) − ΔCm q̇ m (t) + dm (t) (8)

Ps = −ΔM s q̈ s (t) − ΔCs q̇ s (t) + ds (t) . (9)

To proceed with the design of the control scheme, the following assumption and lemmas are required.

Assumption 1. The external disturbance di (t) and model uncertainties in (8), (9) are unknown and bounded. Thus,
there exist positive but unknown constants di and 𝜎 i such that
‖ ( ̇ ̈ )‖
‖Pi qi , qi , qi ‖ ≤ di + 𝜎i . (10)
‖ ‖

Lemma 1 (See the work Polyakov27 ).

Consider the following differential system:
ẋ (t) = 𝑓 (x, t), x (0) = x0 . (11)
Suppose that there exists a continuous positive-definite differentiable function V : U → R+ , and the derivative of V with
respect to time satisfies
( )k
V̇ (x) ≤ − 𝛼V p (x) + 𝛽V g (x) , x ∈ U0 ⧵ {0} , (12)
where 𝛼, 𝛽, p, g, k are positive constants such that pk < 1, gk > 1, and U0 ⊂ U is an open neighborhood of the origin. Then,
system (11) is fixed-time stable. Moreover, if U = U0 = Rn , the system is globally fixed-time stable, with the setting time
T ≤ 1/𝛼 k (1 − pk) − 1/𝛽 k ( gk − 1). In particular, T can be rewritten as follows if the initial value of V(x) is known exactly:
( )
1 1 − 𝑝𝑘 1 − 𝑝𝑘 𝛽
T≤ k |V (x0 )|1−𝑝𝑘 · G k, ,1 + , − |V (x0 )|g−p , (13)
𝛼 (1 − 𝑝𝑘) g−p g−p 𝛼
where V(x0 ) denotes the initial value of V, and G stands for the Gauss hypergeometric function.

Lemma 2 (See the work of Zhu et al28 ).

Suppose there is a continuous positive-definite differentiable function V : U → R+ , which satisfies
( )k
V̇ (x) ≤ − 𝛼V p (x) + 𝛽V g (x) + 𝜃, x ∈ U0 ⧵ {0} (14)
4 WANG ET AL.

where 𝛼, 𝛽, p, g, k are positive constants with pk < 1, 0 < 𝜃 < + ∞, and gk > 1, and U0 ⊂ U is an open neighborhood of
origin, then system (11) is practically fixed-time stable. Moreover, if U = U0 = Rn , the trajectory of the closed-loop system
will converge to a small neighborhood Dwithin the fixed time T

⎧T ≤ 1 1
+ 𝛽 k 𝜁 k (𝑔𝑘−1)
⎪ {
𝛼 k 𝜁 k (1−𝑝𝑘) { ( )1 ( ) 1 }}
⎨ − p1 𝜃 𝑘𝑝 − p1 𝜃 𝑘𝑔
(15)
⎪D = lim x |V (x) ≤ min 𝛼 1−𝜁 k
,𝛽 1−𝜁 k
⎩ t→T

where 𝜁 is a scalar with 0 < 𝜁 ≤ 1. Lemma 2 relaxes and extends the proof conditions of Lemma 1, and explicitly gives
the relationship between the settling time and the invariant set. In particular, Lemma 2 guarantees the state variables to
become arbitrarily small based on proper parameters.

Lemma 3 (See the work of Ding et al29 ).

If p1 > 0 and 0 < p2 ≤ 1, for ∀x, y ∈ R+ , the following inequality holds:

|sig(x)p1 p2 − sig(𝑦)p1 p2 | ≤ 21−p2 |sig(x)p1 − sig(𝑦)p1 |p2 .

| | | |

Lemma 4 (See the work of Qian and Lin30 ).

If c, d are positive constants and 𝛾 is a positive function, the following inequality holds:

c d − dc c+d
𝛾|x|c+d + 𝛾 |𝑦| ≥ |x|c |𝑦|d .
c+d c+d

Chebyshev Neural Network has been widely applied because of its good approximation abilities and low computational
burden. Different from the normal adaptive techniques for estimation of the uncertainties,31 ChNN can be adopted to
approximate the constant or time-varying terms that are unknown. The prominent feature of ChNN is that it is inde-
pendent of system state measurements but only requires local information. Therefore, ChNN can be used to approximate
external disturbances and parameter uncertainties. In addition, Chebyshev polynomials are superior to other polynomials
of the same order in terms of both approximation accuracy and convergence rate.32 Actually, ChNN is a single-layer neural
network with a set of orthogonal Chebyshev polynomials, which can be derived with the following recurrence formula:

Ti+1 (x) = 2𝑥𝑇 i (x) − Ti−1 (x) , T0 (x) = 1, (16)

where T1 (x) = x, and the Chebyshev polynomial is expressed as

𝝑 (X) = [1, T1 (x1 ) , … , Tn (x1 ) , … , T1 (xm ) , … , Tn (xm )]T , (17)

where X ∈ Rm ,𝝑(X) ∈ R(nm + 1) , 𝝑(X) is the basis function of Chebyshev polynomials, and n denotes the order of Chebyshev
polynomials. An arbitrary continuous nonlinear function can be approximated by ChNN, namely,

fN (X) = WT 𝝑 (X) + 𝜀i , (18)

where W ∈ R(nm + 1) × n is the optimal weight matrix, and 𝜀i ∈ Rn is the approximation error. The following two
assumptions are required for further analysis.

Assumption 2. W is bounded by tr(WT W) ≤ WM , where WM is a positive constant.

Assumption 3. 𝜀i is bounded by ||𝜀i || ≤ 𝜀M , where 𝜀M is a positive constant.

With the preliminaries above, the control objective can be stated as follows. Consider a class of bilateral teleoper-
ation systems with jittering time delays, external disturbances and system uncertainties that satisfy Assumptions 1
to 3, and design a control law to ensure that the tracking errors converge within finite time. In addition to this, the
settling time is required to be independent of initial states.
WANG ET AL. 5

3 CO N T RO L LAW DE SIG N

Considering the good approximation capability for unknown nonlinear function of neural network, ChNN is adopted to
approximate the lumped uncertain term Pm (qm , q̇ m , q̈ m ) and Ps (qs , q̇ s , q̈ s ), which can be given by
( ) ( )
Pm qm , q̇ m q̈ m =W Tm 𝝑m qm , q̇ m , q̈ m + 𝜺m (19)

( ) ( )
Ps qs , q̇ s , q̈ s = W Ts 𝝑s qs , q̇ s , q̈ s + 𝜺s , (20)

where the estimation of Pi (qi , q̇ i , q̈ i )is given as follows:

( ) ( )
̂ Ti 𝝑i qi , q̇ i , q̈ i .
P̂ i qi , q̇ i , q̈ i = W (21)

Note that W ̃ T = WT − W
̂ T is the estimate of W T and the estimate error is defined as W ̂ T.
For the bilateral control of the teleoperation system, inspired by the works of Jiang et al33 and Wang et al,34 a nonsingular
fast TSM is designed as

Si = ė i + a1i ei + a2i Swzwi , (22)

where a1i > 0, a2i > 0. Swzwi is the column vector composed of Swzwij , which is given by

⎧ [ ( )p1i ( )p ]k1i
⎪sig b1i sig e𝑖𝑗 + b2i sig e𝑖𝑗 2i , if S𝑖𝑗 = 0
⎪ [ ]k
Swzwij = ⎨ sig b sig(e )p1i + b sig(e )p2i 1i , if S ≠ 0, |e | ≥ 𝛿 (23)
⎪
1i 𝑖𝑗 2i 𝑖𝑗 𝑖𝑗 | 𝑖𝑗 | i
( )2
⎪r1i e𝑖𝑗 + r2i sig e𝑖𝑗 , | |
if S𝑖𝑗 ≠ 0, |e𝑖𝑗 | < 𝛿i ,
⎩

where 𝛿 i , b1i , b2i , p1i , p2i , and k1i are positive constants; S𝑖𝑗 and eij denote the jth ( j = 1, 2, … , n) component of Si and ei ,
p −1∕k1i p −1∕k1i k p −2∕k1i p −2∕k1i k
respectively; r1i = (2 − k1i )(b1i 𝛿i 1i + b2i 𝛿i 2i ) 1i and r2i = (k1i − 1)(b1i 𝛿i 1i + b2i 𝛿i 2i ) 1i .
The auxiliary sliding mode Si is designed as
[ ]k
Si = ė i + a1i ei + a2i sig b1i sig(ei )p1i + b2i sig(ei )p2i 1i . (24)

It can be observed that the conventional sliding mode controllers22,35 are designed utilizing Si and Ṡ i . Therefore, jittering
time delays will result in seriously undesirable chattering, which indicates that a larger gain is required to suppress the
⌢
disturbances. To eliminate such undesirable chattering, the mismatched term is considered in Ṡ i = S i + 𝜹 (ΔTi , Si ), where
⌢ 1
̇ 𝜹(ΔTi , Si ) is an unknown continuous differential function with |𝛿𝑖𝑗 (ΔTi , S𝑖𝑗 )| ≤ 𝜌(S𝑖𝑗 )|S𝑖𝑗 | 2 , and 𝜌(Sij ) ≥ 0
S i = (𝜕Si ∕𝜕e) e.
is a known continuous function (see Figure 1).
Remark 1. Deriving directly of e may cause jittering problems in the conventional sliding mode controllers22,35 because
the additional coefficient 1 − ΔṪ i will result in unknown disturbances, especially in the case of random time delays.
Therefore, to address the jittering delay issue, the disturbance bounded by a positive function is considered rather
than frequently used constant upper bounds.

Remark 2. Recall the existing result in the work of Al-Wais et al,36 which is used to investigate the synchronization
problem of telerobotic systems with known lower and upper bounds. A simple control architecture was provided
to achieve asymptotic stability.36 Motivated by the aforementioned work,36 this paper considers the jittering delay
problem. In addition, convergence performance, such as settling time, is improved to achieve finite-time control. It is
worth pointing out that combining the delay upper bounds with the control laws (25)-(29) is conducive to optimizing
the present control architecture, which is one subject of future studies.

Remark 3. The jittering problem is difficult to address because time delays and time-delay derivative are both
⌢
unknown and varying drastically. In brief, if the time delay is determined, S i can be estimated by an observer, which
can compensate the adverse effects of time delays because 𝜹(ΔTi , Si ) is only dependent on Si in many typical cases.37
6 WANG ET AL.

FIGURE 1 The teleoperation system with the proposed controller. AFTC, adaptive finite-time controller; ChNN, Chebyshev Neural
Network; NFTSM, nonsingular fast terminal sliding mode

Theorem 1. For the teleoperation system (6), (7) with jittering time delays, disturbances, and uncertainties, if Assump-
tions 1 to 3 hold and the following bilateral controllers (25)-(29) are adopted, then all states of the closed-loop system will
be bounded, and the sliding mode between the master and slave robots will converge within finite time:
( )
Fm = Mm0 qm q̈ s (t − ΔTs ) + Fm (25)

( )
Fs = Ms0 qs q̈ m (t − ΔTm ) + Fs (26)

( )
Fi = Ci0 qi , q̇ i q̇ i (t) − a2i Mi0 𝜳 wzwi ė i − c1i sig(Si )p3i
(27)
p ̂ i 𝝑i (·) − Ci0 Si − a1i Mi0 ė i
−c2i Si 4i − W
T

⎧ | [ ( )p1i ( )p ]|k1i −1
⎪diag||k1i b1i sig e𝑖𝑗 + b2i sig e𝑖𝑗 2i ||
⎪ | |
𝜳 wzwi =⎨ [ ( )p1i −1 ( )p2i −1 ] (28)
⎪· b1i p1i diag e𝑖𝑗 + b2i p2i diag e𝑖𝑗 , if Si = 0 or Si ≠ 0, ||e𝑖𝑗 || ≥ 𝛿i
⎪r I + 2r diag (e ) , if Si ≠ 0, ||e𝑖𝑗 || < 𝛿i
⎩ 1i 2i 𝑖𝑗

[ ]
̂̇ i = 𝛾i 𝝑i (·) ST − 𝜇i W
W ̂i , (29)
i

where 0 < p3i < 1, p4i > 1, c1i and c2i are positive constants; 𝛾 i and 𝜇i are positive adaptive parameters. The choice of p4i
is given later.
WANG ET AL. 7

Proof. Choose the candidate Lyapunov function as follows:

V = V1 + V2 + V3
𝜒1 ∑ 2𝜌+𝜒2
V1 = |S𝑖𝑗 | 𝜒1
2𝜌 + 𝜒2 i={m,s} | |
𝑗={1,2 … n}
[ ( )]
∑ 1 T 1
V2 = Si M i0 Si + tr W ̃i
̃ Ti W (30)
i={m,s}
2 2𝛾i
⌢ [ ( )𝛼]
2𝜌−𝜒2
∑ S 𝑖𝑗 ( )𝛼 ⌢ 𝜒
𝜒1
V3 = sig sig 𝜅𝑖𝑗 𝜒2
− sig S 𝑖𝑗 2 d𝜅𝑖𝑗 ,
i={m,s}
∫⌢
S
∗
𝑖𝑗
𝑗={1,2 … n}

where 𝜌, 𝛼, 𝜒 1 , 𝜒 2 are positive constants such that 𝜌 ≥ 𝛼 ≥ 𝜒 1 ≥ 𝜒 2 > 0. The proof is divided into four parts, namely,
calculating V̇ 1 , V̇ 2 , V̇ 3 , and V.
̇
Step 1: Calculate V̇ 1 .
Taking the derivative of V1 with respect to time yields
∑ ( ) 2𝜌−𝜒2 ⌢ ( ) 2𝜌−𝜒2 ( )
V̇ 1 = sig S𝑖𝑗 𝜒1 S 𝑖𝑗 + sig S𝑖𝑗 𝜒1 𝛿𝑖𝑗 t, S𝑖𝑗
i={m,s}
𝑗={1,2, … ,n}

∑ ( ) 2𝜌−𝜒2 (⌢ ⌢∗
) ( ) 2𝜌−𝜒2 ⌢∗ ( ) 2𝜌−𝜒2 (31)
≤ sig S𝑖𝑗 𝜒1 S 𝑖𝑗 − S 𝑖𝑗 + sig S𝑖𝑗 𝜒1 S 𝑖𝑗 + 𝜌 S𝑖𝑗 S𝑖𝑗 𝜒1 ,
i={m,s}
𝑗={1,2, … ,n}

⌢∗ ⌢∗
where S 𝑖𝑗 is a virtual control law with S 𝑖𝑗 = −𝛽1 (S𝑖𝑗 )sig(𝜉1𝑖𝑗 )𝜒2 ∕𝛼 , and 𝛽 1 (Sij ) is a smooth function, which satisfies
𝛽 1 (Sij ) ≥ 𝜌(Sij ) + 𝛽 0 , 𝛽 0 > 0 with 𝜉1𝑖𝑗 = sig(S𝑖𝑗 )𝛼∕𝜒1 .
Hence, we further have
[ )]
∑ 2𝜌 ( ) 2𝜌−𝜒2 (⌢ ⌢∗
̇V1 ≤ −𝛽0 𝜉1𝑖𝑗 𝛼 + sig 𝜉1𝑖𝑗 1 S 𝑖𝑗 − S 𝑖𝑗 .
𝜒
(32)
i={m,s}
𝑗={1,2 … n}

Step 2: Calculate V̇ 2 .
The derivative of V2 is given by
[ ( )]
∑ 1 T ̇ 1 T ̇
V̇ 2 = ̇ ̃ ̃
S M i0 Si + Si M i0 Si + tr W i W i .
T
(33)
i={m,s}
2 i 𝛾i

With the sliding mode (22) and controllers (25)-(27), we have

∑ 1 ( )
V̇ 2 = STi Ṁ i0 Si + ST M i0 Ṡ i + 1 tr W ̃̇ i
̃ Ti W
i={m,s}
2 i
𝛾i
∑ 1
≤ STi (M ̇ i0 Si − 2Ci 0 Si + 2Ci0 Si + 2M i0 ë i + 2M i0 a1i ė i
i={m,s}
2
( )
+ 2M i0 a2i Ṡ wzwi + 2𝜹 (ΔTi , Si ) 𝜆max (M i0 )) − tr W
1 ̂̇ i
̃ Ti W
𝛾i
∑ ( )
≤ S Ci0 Si + M i0 ë i + M i0 a1i ė i + M i0 a2i Ṡ wzwi
T
i (34)
i={m,s}
( )
1
+ ‖Si ‖ 𝛽1 𝜆max (M i0 ) − tr W ̂̇ i
̃ Ti W
𝛾i
∑ ( ( ))
= T p3i p
Si −c1i sig(Si ) − c2i Si 4i − W ̂ Ti 𝝑i (·) + Pi qi , q̇ i , q̈ i
i={m,s}
( )
+ ‖Si ‖ 𝛽1 𝜆max (M i0 ) −
1
tr W ̂̇ i ,
̃ Ti W
𝛾i
where p4i = 2𝜌/𝛼.
8 WANG ET AL.

Substituting the adaptive law (29) into (34) gives

∑ [ p ]
V̇ 2 ≤ ̃ Ti 𝝑i (·)
STi −c1i sig(Si )p3i − c2i Si 4i + STi W
i={m,s}
[ ] ( ) (35)
̃ Ti 𝝑i (·) ST + 𝜇i tr W
−tr W ̂ i + ‖Si ‖ (𝜀i + 𝛽1 𝜆max (M i0 )) .
̃ Ti W
i

Note that the following equality exists:

( ) ( )
̃ Ti 𝝑i (·) = tr ST W
STi W ̃ Ti 𝝑i (·) = tr W̃ Ti 𝝑i (·) ST . (36)
i i

By (36), (35) can be presented as

∑ [ ( )
p ] ̂ i + ‖Si ‖ (𝜀i + 𝛽1 𝜆max (M i0 )) .
̃ Ti W
V̇ 2 ≤ STi −c1i sig(Si )p3i − c2i Si 4i + 𝜇i tr W (37)
i={m,s}

Step 3: Calculate V̇ 3 .
Taking the derivative of V3 with respect to time yields
[ ]
∑ 𝜕V3 ̇ ( ) 2𝜌 ⌢̇
V̇ 3 = S𝑖𝑗 + sig 𝜉2𝑖𝑗 S 𝑖𝑗 ,
𝛼
(38)
i={m,s}
𝜕S𝑖𝑗
𝑗={1,2, … ,n}

𝛼 𝛼
⌢ ⌢∗
where 𝜉2𝑖𝑗 = sig( S 𝑖𝑗 ) 𝜒2 − sig( S 𝑖𝑗 ) 𝜒2 . By Lemmas 3 and 4, we can obtain

| ( ∗) | 𝛼
⎡ | 𝜕sig ⌢ 𝜒2
| ⎤
∑ ⎢ 2𝜌 1− 𝜒2 |
𝜒2 +2𝜌−𝛼 S 𝑖𝑗 | ⎥
̇V2 ≤ ⎢ 𝛼 2 𝛼 ||𝜉2𝑖𝑗 || 𝛼 |
| 𝜕S𝑖𝑗
| Ṡ 𝑖𝑗 ⎥
|
i={m,s} ⎢ | | ⎥
𝑗={1,2, … ,n} ⎣
| | ⎦
| |
( ∗)𝛼 | (39)
⎡ |
∑ ⎢ 2𝜌 1− 𝜒2 | ⌢
𝜒2 +2𝜌−𝛼 | 𝜕sig S 𝑖𝑗
𝜒2
| ∗⎤
| ⌢̇ ⎥
| | | | S 𝑖𝑗 ⎥ .
+ ⎢𝛼2 |𝜉2𝑖𝑗 |
𝛼
𝛼
| ⌢∗ |
i={m,s} ⎢ | 𝜕 S | ⎥
𝑗={1,2, … ,n} ⎣
| 𝑖𝑗 | ⎦
| |

With Lemmas 3-4 and (39), the following inequalities exist:

( ∗)𝛼
⎡| ⌢ | ⎤
∑ ⎢|| 𝜕sig S 𝑖𝑗 |
𝜒2
| ⎥ ∑ [( 𝜒 𝜒 )]

⎢|| | Ṡ 𝑖𝑗 ⎥ ≤ 𝜓2 ||𝜉1𝑖𝑗 || 𝛼 + 𝜓3 ||𝜉2𝑖𝑗 || 𝛼

1− 2 1− 2
| (40)
i={m,s} ⎢||
𝜕S𝑖𝑗 | ⎥ i={m,s}
𝑗={1,2, … ,n} ⎣
| ⎦ 𝑗={1,2, … ,n}
| |

( ∗)𝛼
⎡|| ⌢ 𝜒2 | ⎤
| ∗
∑ ⎢| 𝜕sig S 𝑖𝑗 | ⌢̇ ⎥ ∑ [( 𝜒2 𝜒2 )]

⎢|| | S 𝑖𝑗 ⎥ ≤ 𝜓 |𝜉1𝑖𝑗 |1− 𝛼 + 𝜓5 |𝜉2𝑖𝑗 |1− 𝛼 , (41)

⌢∗ | 4 | | | |
i={m,s} ⎢|| 𝜕 S 𝑖𝑗 | ⎥ i={m,s}
| ⎦ 𝑗={1,2, … ,n}
𝑗={1,2, … ,n} ⎣
| |
{ } { }
| 𝜕𝛽1 𝜒𝛼2 𝜒𝛼1 | 𝛼𝛽1 𝜒𝛼2 | 𝜕𝛽1 𝜒𝛼2 𝜒𝛼1 | 𝛼𝛽1 𝜒𝛼2
| | |
where 𝜓2 = max | 𝜕S 𝜉1𝑖𝑗 | + 𝜒 , 𝛽1 𝜌 ,𝜓3 = max | 𝜕S 𝜉2𝑖𝑗 | + 𝜒 , 𝛽1 𝜌 |
{ | 𝑖𝑗 | 1
} { | 𝑖𝑗 | 1
}
| 𝜕𝛽1 𝜒𝛼2 𝜒𝛼1 | 𝛼𝛽1 𝜒𝛼2 | 𝜕𝛽1 𝜒𝛼2 𝜒𝛼1 | 𝛼𝛽1 𝜒𝛼2
𝜓4 = max || ⌢ 𝜉1𝑖𝑗 || + 𝜒 , 𝛽1 𝜌 𝜓5 = max || ⌢ 𝜉2𝑖𝑗 || + 𝜒 , 𝛽1 𝜌 .
| 𝜕 S 𝑖𝑗 | 1 | 𝜕 S 𝑖𝑗 | 1

Step 4: Calculate V. ̇
Taking the derivative of V yields

V̇ = V̇ 1 + V̇ 2 + V̇ 3 . (42)
WANG ET AL. 9

Substitute (32), (37), (40), and (41) into (42), and there is
∑ ( ) 2𝜌−𝜒2 (⌢ ⌢∗
) ∑ { [ ( )]
̂i
̃ Ti W
2𝜌
V̇ ≤
p
−𝛽0 𝜉1𝑖𝑗 𝛼 + sig 𝜉1𝑖𝑗 𝜒1 S 𝑖𝑗 − S 𝑖𝑗 + STi −c1i sig(Si )p3i − c2i Si 4i + 𝜇i tr W
i={m,s} i={m,s}
𝑗={1,2, … ,n}
} ∑ 2𝜌 1− 𝜒2 | | 𝜒2 +2𝜌−𝛼 ( 𝜒 𝜒 )
𝜓2 ||𝜉1𝑖𝑗 || 𝛼 + 𝜓3 ||𝜉2𝑖𝑗 || 𝛼
1− 2 1− 2
+ ‖Si ‖ (𝜀i + 𝛽1 𝜆max (M i0 )) + 2 𝛼 |𝜉2𝑖𝑗 | 𝛼
i={m,s}
𝛼
𝑗={1,2, … ,n}

∑ 2𝜌 1− 𝜒2 | | 𝜒2 +2𝜌−𝛼 𝜒 ( 𝜒 )
𝜓4 ||𝜉1𝑖𝑗 || 𝛼 + 𝜓5 ||𝜉2𝑖𝑗 || 𝛼
1− 2 1− 2
+ 2 𝛼 |𝜉2𝑖𝑗 | 𝛼
i={m,s}
𝛼
𝑗={1,2, … ,n} (43)
∑ 2𝜌 1 2𝜌 2𝜌
≤ −𝛽0 𝜉1𝑖𝑗 𝛼 + 𝛽0 𝜉1𝑖𝑗 𝛼 + 𝜓1 𝜉2𝑖𝑗 𝛼
i={m,s}
4
𝑗={1,2, … ,n}
∑ ( ( )
p ) ̃ Ti W ̂ i + 1 S T S i + 𝜓7 𝜀
+ STi −c1i sig(Si )p3i − c2i Si 4i + 𝜇i tr W
i={m,s}
2𝜓7 i 2
∑ 1 2𝜌 2𝜌 ∑ 1 2𝜌 2𝜌
+ 𝛽0 𝜉1𝑖𝑗 𝛼 + 𝜓4 𝜉2𝑖𝑗 𝛼 + 𝛽0 𝜉1𝑖𝑗 𝛼 + 𝜓6 𝜉2𝑖𝑗 𝛼 ,
i={m,s}
4 i={m,s}
4
𝑗={1,2, … ,n} 𝑗={1,2, … ,n}

𝛼−𝜒2
𝜒 2𝜌+𝜒2 −𝛼
where 𝜓1 = 2𝛼−𝜒2 ∕𝛼 4𝜌1 [(4𝜌 − 𝜒1 ) 2𝛼−𝜒2 ∕𝛼 ∕𝜌𝛽0 ]4𝜌−𝜒1 ∕𝜒1 , 𝜓6 = 2𝜌
𝛼
2 𝛼 |𝜉2𝑖𝑗 | 𝛼
max {𝜓2 , 𝜓4 },
𝜀 = (𝜀M + 𝛽1 𝜆max (M i0 )) , and 𝜓 7 is a positive constant. Note that the following inequality is utilized in (43):
2

√ 1 T 𝜓7
‖Si ‖ 𝜀 ≤ S Si + 𝜀. (44)
2𝜓7 i 2
Then, (43) can be rewritten as follows:
∑ ( ( )
p ) ̂ i + 1 S T S i + 𝜓7 𝜀
̃ Ti W
V̇ ≤ STi −c1i sig(Si )p3i − c2i Si 4i + 𝜇i tr W
i={m,s}
2𝜓7 i 2
∑ 1 2𝜌 2𝜌
+ − 𝛽0 𝜉1𝑖𝑗 𝛼 + (𝜓1 + 𝜓4 + 𝜓6 ) 𝜉2𝑖𝑗 𝛼
i={m,s}
4
𝑗={1,2 … n}
∑ ( p ) 𝜇i ( ) 𝜇 i ( T ) 𝜓7 (45)
≤ STi −c1i sig(Si )p3i − 𝜓8 Si 4i + tr W Ti W i − tr W ̃i +
̃ iW 𝜀
i={m,s}
2 2 2
∑ 1 2𝜌 1 2𝜌
+ − 𝛽0 𝜉1𝑖𝑗 𝛼 − 𝛽0 𝜉2𝑖𝑗 𝛼 ,
i={m,s}
4 4
𝑗={1,2, … ,n}

where 𝜓 8 is a positive constant.

From the integral form of V3 , the following inequality exits:
⌢ [ ( ) 𝛼 ] 𝜒1 2
2𝜌−𝜒
∑ S 𝑖𝑗 ( )𝛼 ⌢ 𝜒2
V3 = sig sig 𝜅𝑖𝑗 𝜒2
− sig S 𝑖𝑗 d𝜅𝑖𝑗
i={m,s}
∫⌢
S
∗
𝑖𝑗
𝑗={1,2, … ,n}
∑ |⌢ ⌢∗ | | 2𝜌
| (46)
≤ | S 𝑖𝑗 − S 𝑖𝑗 | |𝜉2𝑖𝑗 𝛼 |
| || |
i={m,s}
𝑗={1,2, … ,n}
𝛼−𝜒2 2𝜌+𝜒2
=2 𝜒2 |𝜉2𝑖𝑗 | 𝛼 .
| |
Thus, by Assumption 2, (45), and (46), we have
∑ ( p ) 𝜇i ( ) 𝜓7 ∑ 1 2𝜌 1 2𝜌
V̇ ≤ STi −c1i sig(Si )p3i − 𝜓8 Si 4i + tr W Ti W i + 𝜀+ − 𝛽0 𝜉1𝑖𝑗 𝛼 − 𝛽0 𝜉2𝑖𝑗 𝛼
i={m,s}
2 2 i={m,s}
4 4
𝑗={1,2, … ,n}
[ ]
( ) p3i +1 ( ) p4i +1 𝛽0 2𝜌 (47)
≤ −c1i ‖Si ‖2 2 − 𝜓8 ‖Si ‖2 2 − 6𝜌+2𝜒2
(V1 + V3 ) 2𝜌+𝜒2 +c
2 2𝜌+𝜒2
2𝜌 p4i +1
≤ −𝜒3 V 2𝜌+𝜒2
− 𝜒4 V 2 + c,
10 WANG ET AL.

{ } { }
where 𝜒3 = min c1i ,∕(𝜆max (M i0 ) ∕2)p3i +1∕2 , 𝛽0 ∕28𝜌+3𝜒2 ∕2𝜌+𝜒2 , 𝜒4 = min 𝜓8 ∕(𝜆max (M i0 ,) ∕2)p4i +1∕2 , 𝛽0 ∕28𝜌+3𝜒2 ∕2𝜌+𝜒2
𝜇i 𝜓7
and c = 2 WM + 2 𝜀. According to Lemma 2, the trajectory of the closed-loop system will converge to a small
neighborhood of the sliding surface Si in T1i , which is given by
2𝜌 + 𝜒2 1
T1i ≤ + , (48)
2𝜒2 𝜒3 𝜍i 𝜒4 𝜍i (p4i − 1)
where 𝜍 i is a scalar such that 0 < 𝜍 i ≤ 1, and the convergence neighborhood D is presented as
{ | [( ) 2𝜌−𝜒
2 ( ) 2 ]}
| c 2
c 1+p4i
lim Si ||V ≤ min
1+
D= 2𝜌+𝜒2
. (49)
t→T1i , | 𝜒3 (1 − 𝜍i ) 𝜒4 (1 − 𝜍i )
|
This completes the proof of Theorem 1.

Remark 4. The sliding mode control scheme and the technique of “adding a power integrator”30 are incorporated to
achieve finite-time convergence of system (6), (7). Besides this, the settling time is independent of initial states. Thus,
superior convergence performance is available by flexible parameter selection. Furthermore, this paper improves the
result in the work of Ding and Li37 from the following two aspects.

1. The controller37 is a finite-time second-order sliding control scheme, which implies that the settling time depends
on initial values. However, initial values may be difficult to exactly obtain due to various measurement errors and
external disturbances, which will reduce the estimation effect of the settling time. In this works, the settling time
only depends on the pre-set parameters of the controllers (25)-(29), indicating a wider range of applications.
2. External disturbances and system uncertainties are not considered in the aforementioned work.37 An adaptive
law with ChNN is employed to compensate the adverse effects of external disturbances, system uncertainties and
jittering time delays.
Remark 5. Note that, in the case of S𝑖𝑗 ≠ 0 , the terminal sliding manifold is switched into the general nonsingular
sliding manifold when |eij | < 𝛿 i , and while in the case of S𝑖𝑗 = 0 , the sliding manifold is obviously nonsingular. Thus,
singularity phenomenon is eliminated by the switching scheme.
The convergence of the sliding mode implies the trajectory of state variables will move along the sliding manifold.
Based on the sliding mode (22), the following theorem is obtained.

Theorem 2. For the bilateral control systems (7), (8) with jittering time delays, external disturbances, and system uncer-
tainties, the tracking errors will converge to the origin within finite time after the state trajectory reaches the sliding
manifold.

Proof. Consider the following Lyapunov function:

∑ 1
V4 = eTi ei . (50)
i={m,s}
2

Taking the derivative of (50) yields

∑
V̇ 4 = eTi ė i
i={m,s}
∑
=− eTi (a1i ei + a2i Swzwi )
i={m,s}
∑ [ ]k
≤− eTi a2i sig b1i sig(ei )p1i + b2i sig(ei )p2i 1i
i={m,s}
(51)
∑ || |k1i
1 1
( T ) k1i +p1i ( T ) k1i +p2i |
| |
=− |a2i b1i ei ei + a2i b2i ei ei |
2 2

|
i={m,s}| |
|
| 1 +p 1 +p
1 +p |
k1i
| ( ) k1i 1i 1 +p
( ) k1i 2i 2i |
| 1 2
k1i 1i
1 2
k1i
|
= −|a2i b1i (V1 ) 2 + a2 b2 (V1 ) 2 | .
| 2 2 |
| |
| |
WANG ET AL. 11

According to Lemma 1, the state trajectories will converge to the origin within the finite time T2i after reaching the
sliding surface. T2i is bounded by
3+k1i p1i 3+k1i p2i
2 2 2 2
T2i ≤ k k
+ k k
. (52)
a2i1i b1i1i (1 − k1i p1i ) a2i1i b2i1i (1 − k1i p2i )

If the initial valueV4 (0) is known, a more accurate estimation of settling time can be obtained as
{ 3+k1i p1i
}
2 1−k1i p1i
T̂ 2i ≤ max
2

k k
|V4 (0)| 2 𝛯i , (53)
a2i1i b1i1i (1 − k1i p1i )
( p2i −p1i
)
1−k1i p1i 1−k1i p1i b
where 𝛯i = G k1i , p −p ,1 + p2i −p1i
, − b2i |V4 (0)| 2 , and G stands for the Gauss hypergeometric function.
2i 1i 1i

Remark 6. The total settling time T is fixed such that one hasT = max i=m,s {T1i + T2i }, which means a higher conver-
gence efficiency than that of conventional FTCs. In addition, T can be estimated and controlled by proper choices of
controller parameters c1i ,c2i ,𝜇i ,a2i , b2i , k1i , p2i , p3i , p4i according to the requirement of tasks.

Remark 7. Similarly to the works of Yang et al,22,35 the acceleration measurements of the master and slave manipula-
tors are required. In fact, accurate acceleration information is not easy to obtain because accelerometers are sensitive
to external noise. Moreover, accelerometers are impractical due to high cost. To solve this problem, the design of
acceleration observer has been investigated based on pseudo-differentiator38 and low-pass filter technique.39 Combin-
ing acceleration observer and teleoperation control scheme is a challenging issue because time-varying delays would
degrade the observer performance. It becomes one subject of our future studies to address this issue.

4 SIMULATION AND EXPERIMENT

4.1 Simulation on teleoperation of two-DOF manipulators

In this section, simulation results for the bilateral teleoperation system are presented to illustrate the effectiveness of the
proposed adaptive FTC (AFTC). Two serial linked manipulators with the identical two DOF are chosen as the master
and slave manipulators, respectively. In particular, for comparison, the FTC proposed in the work of Yang et al35 is also
simulated as
( ) −1 −1 ( ) ( )
𝝉 m = −M m qm 𝛽m 𝛾m sig(ė m )2−𝛾m + M m qm q̈ s (t − ΔTs ) + Cm qm , q̇ m q̇ m
( )( ) (54)
− M m qm K m Sm + Bm sig(Sm )𝜌m
( ) ( ) ( )
𝝉 s = −M s qs 𝛽s−1 𝛾s−1 sig(ė s )2−𝛾s + M s qs q̈ m (t − ΔTm ) + Cs qs , q̇ s q̇ s
( )( ) (55)
− M s qs K s Ss + Bs sig(Ss )𝜌s ,

where Sm = em + 𝛽m sig(ė m )𝛾m and Ss = es + 𝛽s sig(ė s )𝛾s . The corresponding parameters are given in Table 1. The nominal
mass and length of the bilateral teleoperation system are set as m ̂ 𝑖𝑗 = 1.0 kg, ̂lm1 = 1.0 m, ̂lm2 = 2.0 m, ̂ls1 = 1.0 m, and
̂ls2 = 2.0 m, respectively. The controller and system parameters are shown in Tables 1 and 2.
The sign function in the proposed controller (25)-(27) will result in chattering phenomena. To solve this problem,
the hyperbolic tangent function is adopted to represent the sign function. The jittering time delays are set as
ΔTm = (0.4 + 0.3 sin 2t + 𝜛) seconds and ΔTs = (0.3 + 0.2 sin 3t + 𝜛) seconds, where 𝜛 follows Gaussian distribution
(see Figures 2 and 3). The statistical parameters 0.3 and 0.02 seconds are for the mean and variance, respectively.
Case 1 (Free motion).
To verify the effectiveness of the proposed controller, the unmanned motion is simulated firstly. The initial states of
the master and slave manipulators are set as qm (0) = [𝜋/6, 𝜋/6 ]T and qs (0) = [0, 0 ]T , respectively.
Figures 4 to 7 show the responses of position errors driven by two controllers (AFTC and FTC), respectively. The sys-
tem controlled with AFTC scheme behaves with a better convergence property. The position errors converge within
12 WANG ET AL.

TABLE 1 Controller
parameters in the simulation

a1i 0.3 p3i 1.0

a2i 0.3 p4i 2.0
b1i 0.75 k1i 1.2
b2i 1.2 𝛿i 0.01
c1i 0.5 𝛾i 1.0
c2i 0.6 𝜇i 20
p1i 0.6 𝜷i diag(0.8, 0.8)
p2i 1.5 Ki diag(6, 6)
𝜌i 0.33 Bi diag(0.8, 0.8)
me 0 de 0.01
ke 0.5 xe [2.5, −1.5]T

TABLE 2 System parameters in

the simulation

mm1 0.3 kg lm1 1.0 m

mm2 0.3 kg lm2 1.0 m
ms1 0.3 kg ls1 1.0 m
ms2 0.3 kg ls2 1.0 m

FIGURE 2 Forward jittering time delay [Colour figure can be viewed at wileyonlinelibrary.com]

a shorter settling time, with higher accuracy and smaller overshoot, whereas jittering time delays and system uncer-
tainties result in the chattering problem in the system controlled by FTC. Figures 8 to 11 depict the velocity errors
and their responses of the master and slave manipulators, respectively. It can be seen that the chattering phenomena
reflected in the velocity error responses is more obvious in Figures 10 and 11. Comparing with FTC, the AFTC scheme
overcomes the problem, and fast convergence performance can be further obtained, as shown in Figures 8 and 9.

Case 2 (Human manipulation).

The human operator input is further considered. The initial values of the master and slave manipulators are set as
qm (0) = [𝜋/4, 𝜋/4 ]T rad and qs (0) = [0, 0 ]T rad, respectively. Fh is modeled as a spring-damper mechanical system
{ ( ) ( )
kh qd − q + dh q̇ d − q̇ , 0 ≤ t ≤ 10
F h (t) = (56)
0, t > 10,
where kh = 0.001 and dh = 0.001.
WANG ET AL. 13

FIGURE 3 Backward jittering time delay [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 4 Position errors of the master manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 5 Position errors of the master manipulator in finite-time controller (FTC) [Colour figure can be viewed at wileyonlinelibrary.com]
14 WANG ET AL.

FIGURE 6 Position errors of the slave manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 7 Position errors of the slave manipulator in finite-time controller (FTC) [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 8 Velocity errors of the master manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]
WANG ET AL. 15

FIGURE 9 Velocity errors of the master manipulator in finite-time controller (FTC) [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 10 Velocity errors of the slave manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 11 Velocity errors of the slave manipulator in finite-time controller (FTC) [Colour figure can be viewed at wileyonlinelibrary.com]
16 WANG ET AL.

FIGURE 12 Position errors of the master manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 13 Position errors of the master manipulator in finite-time controller (FTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 14 Position errors of the slave manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]
WANG ET AL. 17

Figures 12 to 15 show the responses of position errors in the case of human operator's manipulation. Although
the initial values are different from those in Case 1, the position errors driven by the AFTC scheme converge within
the same settling time. While a human torque is applied to the master robot, the position errors remain bounded,
and when the human operator releases the robot, the position tracking is established rapidly. The bilateral teleopera-
tion system controller by the AFTC scheme performs better in terms of convergence rate and accuracy; however, the
applied force degrades the control performance of FTC. Figures 16 to 19 demonstrate the responses of velocity errors.
The chattering problem at velocity level is further solved effectively, and the settling time and accuracy are guaranteed
in the AFTC scheme.

Case 3 (Passive and nonpassive environments).

To verify the effectiveness of the AFTC scheme, we further consider the force interaction between the slave manipula-
tor and the environment, where passive and nonpassive environments are simulated simultaneously. In addition, the
human operator input in Case 23 is considered. For the passive case, the dynamics40 of the environment interacting
with the end-effector of the slave manipulator is presented as

̈ + de (ẋ e − x)
F e (t) = me (ẍ e − x) ̇ + ke (x e − x) 2 ≤ t ≤ 8, (57)

where me , de , and ke represent the mass, damping, and stiffness coefficients of the environment, respectively. xe and
x denote the displacement of the environment and the end-effector in Cartesian space, respectively. For the nonpas-
sive case, the force41 between the environment and the slave manipulator is set as Fe (t) = [0.5 sin (2t), 0.5 sin (2t)]T ,

FIGURE 15 Position errors of the slave manipulator in finite-time controller (FTC) [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 16 Velocity errors of the master manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]
18 WANG ET AL.

FIGURE 17 Velocity errors of the master manipulator in finite-time controller (FTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 18 Velocity errors of the slave manipulator in adaptive finite-time controller (AFTC) [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 19 Velocity errors of the slave manipulator in finite-time controller (FTC) [Colour figure can be viewed at wileyonlinelibrary.com]
WANG ET AL. 19

FIGURE 20 Position errors of the master manipulator. AFTC, adaptive finite-time controller [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 21 Position errors of the slave manipulator. AFTC, adaptive finite-time controller [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 22 Velocity errors of the master manipulator. AFTC, adaptive finite-time controller [Colour figure can be viewed at
wileyonlinelibrary.com]
20 WANG ET AL.

FIGURE 23 Velocity errors of the slave manipulator. AFTC, adaptive finite-time controller [Colour figure can be viewed at
wileyonlinelibrary.com]

FIGURE 24 Master-slave position tracking response of the first joint. AFTC, adaptive finite-time controller [Colour figure can be viewed
at wileyonlinelibrary.com]

FIGURE 25 Master-slave position tracking response of the second joint. AFTC, adaptive finite-time controller [Colour figure can be
viewed at wileyonlinelibrary.com]
WANG ET AL. 21

40 ≤ t ≤ 48. Thus, the passive and nonpassive environments are both contacted during the motion of the slave
manipulator.
Figures 20 and 21 show the position errors of the master and slave manipulators interacting with passive and non-
passive environments. It can be seen that synchronous performance is guaranteed even under jittering delays, passive
or nonpassive environment, and system uncertainties. Moreover, Figures 22 and 23 depict that the velocity errors con-
verge and contact stability can be maintained. Figures 24 and 25 gives the master-slave position tracking responses,
where the tracking performance is achieved.

4.2 Experiment on teleoperation of three-DOF PHANTOM manipulator

The teleoperation system for the experiments consists of two three-DOF PHANTOM Premium 1.5A manipulators (see
Figure 26). Incremental rotary encoders are equipped to calculate the endpoint position in the Cartesian space. The two
robotic manipulators are connected to two computers that communicate via the local area network. The controllers and
network environment are set in MATLAB/Simulink blocks. The time delay is set as (0.4 + 0.2 sin 2t + 𝜛) seconds, where
𝜛 is a random number in the range of [0, 0.3].

FIGURE 26 Experimental teleoperation system [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 27 Tracking response of q1 (blue solid line: qm1 ; red dotted line: qs1 ) [Colour figure can be viewed at wileyonlinelibrary.com]
22 WANG ET AL.

FIGURE 28 Tracking response of q2 (blue solid line: qm2 ; red dotted line: qs2 ) [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 29 Tracking response of q3 (blue solid line: qm3 ; red dotted line: qs3 ) [Colour figure can be viewed at wileyonlinelibrary.com]

Figures 27 to 29 depict the position tracking responses of the master and slave manipulators when the human operator
manipulates the master robot, and the slave robot moves in the Cartesian space. Although the operator exerts a force
on the master robot, the position errors tend to be bounded. Synchronization performance can be obtained within short
settling time, and thus, the ability is available to overcome irregular time varying delays.
In summary, the proposed AFTC with ChNN can achieve bilateral teleoperation with high tracking accuracy, fast con-
vergence rate, and finite-time stability in the presence of system uncertainties and jittering delays. Moreover, the flexible
choice of control parameters is conductive to achieving desirable performance in terms of convergence accuracy and
settling time.

5 CO N C LU S I O N

This paper investigates and discusses the control scheme for a class of bilateral teleoperation systems with jittering delays.
Jittering time delays result in chattering problems, which will degrade the control performance seriously. To eliminate
the adverse effect of jittering delays, an anti-jittering scheme is developed to relax and extend the common assumptions of
time delays. The disturbances bounded by a positive function are considered rather than frequently used constant upper
bounds. For the bilateral teleoperation systems with jittering delays, external disturbances, and system uncertainties, an
adaptive sliding mode control scheme is proposed to achieve finite-time convergence through the technique of “adding a
power integrator,” in this case, the settling time is independent of initial values. An adaptive law and ChNN are incorpo-
rated to compensate the system uncertainties. The numerical simulation and experiment results show the effectiveness
of the proposed approach.
WANG ET AL. 23

ACKNOWLEDGEMENTS
This research is supported by the National Natural Science Foundation of China (Grant no. 61703228, 61673239), Science
and Technology Project of Shenzhen (Grant no. JCYJ20160428182227081), and The Science and Technology Planning
Project of Guangdong Province (Grant no. 2017B010116001).

ORCID

Ziwei Wang https://orcid.org/0000-0003-4588-8501

REFERENCES
1. Sheridan TB. Space teleoperation through time delay: review and prognosis. IEEE Trans Robotics Autom. 1993;9(5):592-606.
2. Imaida T, Yokokohji Y, Doi T, Oda M, Yoshikawa T. Ground-space bilateral teleoperation of ETS-VII robot arm by direct bilateral coupling
under 7-s time delay condition. IEEE Trans Robotics Autom. 2004;20(3):499-511.
3. Landzettel K, Preusche C, Albu-Schaffer A, Reintsema D, Rebele B, Hirzinger G. Robotic on-orbit servicing-DLR's experience and
perspective. Paper presented at: 2006 IEEE/RSJ International Conference on Intelligent Robots and Systems; 2006; Beijing, China.
4. Khatib O, Yeh X, Brantner G, et al. Ocean one: a robotic avatar for oceanic discovery. IEEE Robotics Autom Mag. 2016;23(4):20-29.
5. Sanchez LA, Le MQ, Liu C, Zemiti N, Poignet P. The impact of interaction model on stability and transparency in bilateral teleop-
eration for medical applications. Paper presented at: 2012 IEEE International Conference on Robotics and Automation (ICRA); 2012;
Saint Paul, MN.
6. Preusche C, Ortmaier T, Hirzinger G. Teleoperation concepts in minimal invasive surgery. Control Eng Pract. 2002;10(11):1245-1250.
7. Desbats P, Geffard F, Piolain G, Coudray A. Force-feedback teleoperation of an industrial robot in a nuclear spent fuel reprocessing plant.
Ind Robot Int J. 2006;33(3):178-186.
8. Anderson RJ, Spong MW. Bilateral control of teleoperators with time delay. IEEE Trans Autom Control. 1989;34(5):494-501.
9. Niemeyer G, Slotine JE. Stable adaptive teleoperation. IEEE J Ocean Eng. 1991;16(1):152-162.
10. Niemeyer G, Slotine JE. Telemanipulation with time delays. Int J Robotics Res. 2004;23(9):873-890.
11. Nuño E, Basañez L, Ortega R. Passivity-based control for bilateral teleoperation: a tutorial. Automatica. 2011;47(3):485-495.
12. Hokayem PF, Spong MW. Bilateral teleoperation: an historical survey. Automatica. 2006;42:2035-2057.
13. Sun D, Naghdy F, Du HP. Wave-variable-based passivity control of four-channel nonlinear bilateral teleoperation system under time
delays. IEEE/ASME Trans Mechatron. 2016;21(1):238-253.
14. Li H, Kawashima K. Achieving stable tracking in wave-variable-based teleoperation. IEEE/ASME Trans Mechatron. 2014;19(5):1574-1582.
15. Hua CC, Liu XP. Delay-dependent stability criteria of teleoperation systems with asymmetric time-varying delays. IEEE Trans Robotics.
2010;26(5):925-932.
16. Hashemzadeh F, Hassanzadeh I, Tavakoli M. Teleoperation in the presence of varying time delays and sandwich linearity in actuators.
Automatica. 2013;49(9):2813-2821.
17. Chen Z, Liang B, Zhang T, Wang X. Bilateral teleoperation in cartesian space with time-varying delay. Int J Adv Robotic Syst. 2012;9(4):110.
18. Chopra N, Spong MW, Lozano R. Synchronization of bilateral teleoperators with time delay. Automatica. 2008;44(8):2142-2148.
19. Nuño E, Basañez L, Ortega R, Spong MW. Position tracking for non-linear teleoperators with variable time delay. Int J Robotics Res.
2009;28(7):895-910.
20. Islam S, Liu PX, Saddik AE, Yang Y. Bilateral control of teleoperation systems with time delay. IEEE/ASME Trans Mechatron.
2014;20(1):1-12.
21. Zhai D, Xia Y. Adaptive finite-time control for nonlinear teleoperation systems with asymmetric time-varying delays. Int J Robust Nonlinear
Control. 2016;26(12):2586-2607.
22. Yang Y, Hua C, Guan X. Finite time control design for bilateral teleoperation system with position synchronization error constrained.
IEEE Trans Cybern. 2016;46(3):609-619.
23. Chen Z, Liang B, Zhang T, Wang X, Zhang B. Adaptive bilateral control for nonlinear uncertain teleoperation with guaranteed transient
performance. Robotica. 2016;34(10):2205-2222.
24. Ahmadi B, Nourisola H, Tavakoli S. Robust adaptive sliding mode control design for uncertain stochastic systems with time-varying delay.
Int J Dyn Control. 2018;6:413-424.
25. Sun D, Naghdy F, Du H. Neural network-based passivity control of teleoperation system under time-varying delays. IEEE Trans Cybern.
2017;47(7):1666-1680.
26. Wang Z, Chen Z, Liang B, Zhang B. A novel adaptive finite time controller for bilateral teleoperation system. Acta Astronaut.
2018;144:263-270.
27. Polyakov A. Nonlinear feedback design for fixed-time stabilization of linear control systems. IEEE Trans Autom Control.
2012;57(8):2106-2110.
28. Zhu Z, Xia Y, Fu M. Attitude stabilization of rigid spacecraft with finite-time convergence. Int J Robust Nonlinear Control.
2011;21(6):686-702.
24 WANG ET AL.

29. Ding S, Li S, Zheng WX. Non-smooth stabilization of a class of nonlinear cascaded systems. Automatica. 2012;48(10):2597-2606.
30. Qian CJ, Lin W. A continuous feedback approach to global strong stabilization of nonlinear systems. IEEE Trans Autom Control.
2001;34(5):1061-1079.
31. Mohajerpoor R, Sharifi I, Talebi HA, Rezaei SM. Adaptive bilateral teleoperation of an unknown object handled by multiple robots under
unknown communication delay. Paper presented at: 2013 IEEE/ASME International Conference on Advanced Intelligent Mechatronics;
2013; Wollongong, Australia.
32. Zou AM, Kumar KD, Hou ZG. Distributed consensus control for multi-agent systems using terminal sliding mode and Chebyshev neural
networks. Int J Robust Nonlinear Control. 2013;23(3):334-357.
33. Jiang B, Hu Q, Friswell MI. Fixed-time attitude control for rigid spacecraft with actuator saturation and faults. IEEE Trans Control Syst
Technol. 2016;24(5):1892-1898.
34. Wang L, Chai T, Zhai L. Neural-network-based terminal sliding-mode control of robotic manipulators including actuator dynamics. IEEE
Trans Ind Electron. 2009;56(9):3296-3304.
35. Yang Y, Hua C, Ding H, Guan X. Finite-time coordination control for networked bilateral teleoperation. Robotica. 2015;33(2):451-462.
36. Al-Wais S, Mohajerpoor R, Shanmugam L, Abdi H, Nahavandi S. Improved delay-dependent stability criteria for telerobotic systems with
time-varying delays. IEEE Trans Syst Man Cybern Syst. 2018;48:2470-2484.
37. Ding S, Li S. Second-order sliding mode controller design subject to mismatched term. Automatica. 2017;77:388-392.
38. Sakaino S, Furuya T, Tsuji T. Bilateral control between electric and hydraulic actuators using linearization of hydraulic actuators. IEEE
Trans Ind Electron. 2017;64(6):4631-4641.
39. Yokokura Y, Ohishi K, Katsura S. Fine force reproduction based on motion-copying system using acceleration observer. IEEE Trans Ind
Electron. 2014;61(11):6213-6221.
40. Li HJ, Song AG. Virtual-environment modeling and correction for force-reflecting teleoperation with time delay. IEEE Trans Ind Electron.
2007;54(2):1227-1233.
41. Atashzar FS, Polushin GL, Patel RV. Networked teleoperation with non-passive environment: application to tele-rehabilitation. Paper
presented at: 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS); 2012; Vilamoura, Portugal.

How to cite this article: Wang Z, Chen Z, Zhang Y, Yu X, Wang X, Liang B. Adaptive finite-time con-
trol for bilateral teleoperation systems with jittering time delays. Int J Robust Nonlinear Control. 2018;1–24.
https://doi.org/10.1002/rnc.4423