Improving Tracking Performance of Nonlinear Uncertain Bilateral Teleoperation Systems With Time-Varying Delays and Disturbances

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2019.2962663, IEEE/ASME
Transactions on Mechatronics
JOURNAL OF IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. XX, NO. XX, XX XXXX 1
Improving Tracking Performance of Nonlinear

Uncertain Bilateral Teleoperation Systems with
Time-Varying Delays and Disturbances
Henghua Shen, Student Member, IEEE, and Ya-Jun Pan, Senior Member, IEEE
Abstract—In this paper, an adaptive non-singular terminal on the human operator as well as the environment that cannot
sliding mode (ANTSM) method is proposed for the motion be satisfied in all applications. In the literature of teleoperative
tracking control of a bilateral teleoperation system. Efforts in stability, work has been done based on the dynamic and kine-
this paper seek to improve the position tracking performance
of nonlinear systems subject to time-varying network delays, matic models to relax the strong limitations on the human and
parametric uncertainties, and unknown external disturbances environments [7]. Moreover, controllers have been developed
and frictions. Another issue addressed in this paper is the with adaptiveness and robustness to relieve the reliance on
common delay-induced phase shift of tracking profiles in many the precise information of the model and other undesired
control methods, which is greatly reduced by introducing a novel disturbances/frictions acting on robots [8], [9].
mixed-type of feedback signals in the ANTSM control design.
Furthermore, the proposed adaptive control design with two Unfortunately, the delay-induced phase shift (or horizontal
online-estimated compensatory bounds removes the requirement shift) still remains as a major limitation in network-delayed
of exact knowledge of network delays and disturbance bounds as systems. In many methods proposed in the field of networked
a prior. In the master side, a force predictor is used to estimate the control systems, the delay-induced phase shift are caused by
current environmental force for the reference signal generator. the use of protocol that compares an agent’s own current
Therefore, the direct transmission of force signals is avoided.
By comparing with the existing model-based and model-free state with the delayed states from the network transmission,
methods, numerical simulation results with six degrees of freedom which is known as feedback signals without self-delays. Con-
(6 DOFs) manipulators illustrate the merits of the developed sequently, the systems tend to track the delayed information,
robust and adaptive controllers. Experimental results with two which gives rise to the position tracking errors. To solve this
Phantom Omni devices are also provided to demonstrate the issue, it is usually assumed that the human motions are given
effectiveness and the significant performance improvements of
the proposed controllers. in a move-and-wait fashion or operating with a constant speed
[10], but this cannot always be met in practical applications.
Index Terms—Nonlinear Bilateral Teleoperation, Non-Singular
Alternatively, protocols given by the difference between the
Terminal Sliding Mode, Adaptive Control, Time-Varying Net-
work Delays, Phase Shift, Mixed-Type Feedback. delayed states and the agent’s own state manually delayed by
an exact/inexact estimate of the communication delay tend
to improve the synchronization performance. This type of
I. INTRODUCTION
protocol is known as feedback signals with exact/inexact self-
Teleoperation systems are the interactive manipulations of delays. However, using this type of feedback signals can cause
a remote system where the master information is transmitted the control system to become unstable, especially when the
to control systems in a remote site, and the fed-back environ- communication delays are long.
mental forces provide the situational awareness for the human Motivated by the contrary effects of the feedback signals
operator. Bilateral teleoperation has broad industrial applica- with and without self-delays, J. Klotz, et. al [11] proposed
tions, such explosive removal [1], underwater exploration and a model-free control method where a mixture of different
inspection [2], and medical surgery [3]. As the information is types of feedback signals normalized by a set of weighting
shared in real-time via network channels, one of the inevitable coefficients are utilized. This method improves the tracking
effects is the transmitting delay which makes the remote robots accuracy and meanwhile maintains the stability. However, the
sluggish to the dynamic motions from the master side. self-delayed signals are difficult to be predetermined, as the
Teleoperation systems are prone to instabilities due to network delays are usually nonuniform, asymmetric, and not
various system nonlinearities and long delays that decrease the exactly known. Therefore, some studies in recent years have
stability region/margin on the S-plane [4]. A popular method focused on using self-delayed feedback signals with inexact
of ensuring the stability is the use of time-domain passivity delay information, such as the upper delay bounds or the
control approaches [5], [6]. By considering the energy content nominal delays. Using the upper delay bounds artificially
of input-output ports of a system, an artificial damping is enlarges the delays via a receiver buffer whose size is not less
applied to dissipate the energy and therefore ensures the than the maximum network delay. However, the performance
stability. However, this approach imposes passivity constraints and stability of the systems degrade as longer delays decrease
H. Shen and Y. Pan are with the Department of Mechanical Engineering, the stability region/margin on the S-plane. Alternatively, in
Dalhousie University, Halifax, Canada, B3H 4R2. Corresponding author: Y. practice, transmission delays can be modeled in a stochastic
Pan, email: [email protected]. form with a known nominal delay plus an unknown but
1083-4435 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2019.2962663, IEEE/ASME
bounded delay variations. In this way, the delay-induced errors II. P RELIMINARIES
can mostly be compensated, while the errors induced by the A. Finite Time Convergence Criteria
delay variations are regarded as bounded disturbances [12].
Considering a nonlinear dynamic system:
To deal with the effect of external disturbances, the sliding
mode control (SMC) method has been extensively studied to ẋ1 = x2 ,
(1)
improve the robustness of both linear and nonlinear systems. ẋ2 = f (x) + b(x)u + d(x),
Moreover, to deal with the unknown conditions, such as
where x = [x1 x2 ]T ∈ R2 is the system state, f (x) and
parameter uncertainties, frictions, and unknown disturbances,
b(x) are smooth nonlinear functions, and d(x) is the external
Z. Man, et al [13] proposed an adaptive terminal sliding mode
disturbance with a known bound, i.e., |d(x)| ≤ D and D ≥
(TSM) method to estimate the boundary of the uncertainty
0. A sliding manifold s is constructed from the user-define
and the disturbance and therefore the prior information of the
position error ep and the velocity error ev [15],
bounds is not required. The non-singular terminal sliding mode
(NTSM) [14], [15] was developed so that the sliding surface s = ep + β(ev )α , (2)
as well as the zero errors can be realized in a finite time and
where β > 0, and α = (2h + 1)/(2h − 1) with h = 1, 2, ...,
meanwhile avoid singularities. However, the ANTSM control
such that 1 < α < 2. The NTSM controller u designed from
design remains as a challenging issue for networked systems in
the sliding surface s enables the finite-time convergence and
the presence of time-varying delays, parametric uncertainties
disturbance insensitiveness. In the finite-time stability analysis
and disturbances with unknown bounds.
using the Lyapunov method, the following lemma provides a
This work presents a novel control scheme that can improve
sufficient condition to determine the convergence time.
the tracking performance of teleoperation systems described
by nonlinear second-order Euler–Lagrange (EL) equations of Lemma 1. [17] For a non-Lipschitz continuous non-linear
motion subject to time-varying network delays, parametric system ẋ = f (x), suppose there exists a continuous function
uncertainties, and unknown external disturbances and frictions. V (x) defined on a neighborhood of the origin such that the
Compared with the literature in [7], [11], [13]-[16], this work following conditions hold: (1) V (x) is positive definite; (2)
has the following contributions. There exist real numbers c > 0 and 0 < γ < 1 such that
1) Given a time-varying master trajectory delayed by time- V̇ (x) + cV γ ≤ 0. Then the origin is locally finite-time stable,
varying latencies, the mixed-type feedback signals without and the settling time satisfies
self-delay and with inexact self-delay are crafted in the V (x0 )1−γ
ANTSM control design to reduce the phase shift effect and t(x0 ) ≤ .
c(1 − γ)
meanwhile maintain the stability without imposing passivity
constraints on the operator and environment. A sufficient
B. Types of Feedback Signals
condition is developed for the accuracy improvement by using
the mixed-type feedback signals. In the control design of networked systems with time-
2) Two online-estimated compensatory bounds in the slave varying delays, there are three types of feedback signals
ANTSM controller are established to compensate the state commonly used [18].
errors induced by the combined effects of time-varying delays 1) Feedback without self-delay:
and the use of the mixed-type feedback signals. The errors ei (t) = xi (t) − xj (t − Tij (t)). (3)
caused by unknown delay variations are treated as disturbances
that can be well dealt with by the developed ANTSM control. 2) Feedback with identical (or exact) self-delay:
3) Both numerical simulations and experimental tests are ei (t) = xi (t − Tij (t)) − xj (t − Tij (t)). (4)
provided to prove the feasibility and effectiveness of the
proposed method for the systems with dynamic uncertainties, 3) Feedback with different (or inexact) self-delay:
unknown external disturbances, and unknown frictions. Simu-
ei (t) = xi (t − T̂ij (t)) − xj (t − Tij (t)), (5)
lation results in comparison with model-free and model-based
control methods present the performance improvement of the where T̂ij (t) is the inexact estimation of the time-varying
proposed control approach. The operating performance of the delay Tij (t) between the agents i and j.
developed method is demonstrated experimentally with two Controllers designed with the feedback signals without self-
haptic devices communicating via a local area network (LAN). delay are superior in maintaining the stability but the tracking
The rest of the paper is organized as follows. Section accuracy is degraded. In contrast, controllers designed from the
II provides relevant notations and preliminaries. In Section feedback signals with exact/inexact self-delay help improve
III, the control problems are formulated with the systems the tracking accuracy, but the systems are prone to be unstable,
described by EL dynamics. The main results of the proposed especially when the network delays are long.
ANTSM control design are presented in Section IV. Section In order to keep equations as compact as possible, the
V and Section VI present the results of the numerical sim- following notations are used throughout the article:
ulations and experimental tests, respectively, to verify and 1) m and s denote the variables of the master and slave
validate improving performance and practical feasibility of the manipulators, respectively, and the subscripts of ∗m,s is used
proposed theories. Finally, Section VII summarizes the results when referring to the same variable for the master and slave;
and outlines the future work. 2) The argument of time-dependent signals will be omitted
−T jo −1 ˙ −1 −T jo
unless otherwise specified, e.g., x ≡ x(t); Jm,s Mm,s Jm,s Jm,s Jm,s , gm,s = Jm,s gm,s , where the super-
3) k ∗ k1 denotes the `-1 norm for a matrix/vector, and | ∗ | is script “jo” denotes the matrices/vector in joint space.
the absolute value for a scalar;
Property 2. The Jacobian matrix, Jm,s = ∂xm,s /∂qm,s ,
4) Simplify the trigonometry as s1 = sin(q1 ), c1 = cos(q1 ),
bridges the velocities in different spaces as ẋm,s = Jm,s q̇m,s .
s2 = sin(q2 ), c2 = cos(q2 ), s3 = sin(q3 ), c3 = cos(q3 ),
And the acceleration of the end effector can be obtained by
s23 = sin(q2 +q3 ), c23 = cos(q2 +q3 ), c2∗23 = cos(2q2 +q3 );
the time-differentiation as ẍm,s = J˙m,s q̇m,s + Jm,s q̈m,s .
5) Phase shift defines how far the a wave’s position is shifted
horizontally from others in the similar waveform (https : Property 3. [21] As Mm,s has linearly independent columns,
+ +
//en.wikipedia.org/wiki/P hase( waves)). Mm,s Mm,s = I, where Mm,s represents the Pseudo-inverse
of Mm,s .
III. P ROBLEM F ORMULATION Assumption 1. The human/operator force and the environ-
For the master-slave manipulators, the end effectors are mental force are measurable.
modeled by the nonlinear Euler-Lagrange formulations as
Assumption 2. The following terms in the dynamic models are
Mm ẍm + Cm ẋm + gm + fnm = um + fh , (6) bounded by unknown positive constants as: kMm,s k1 ≤ km,s 1
,
Ms ẍs + Cs ẋs + gs + fns = us − fe , (7) kCm,s xm,s + gm,s k1 ≤ km,s + km,s kxm,s k1 + km,s kẋm,s k21 ,
2 3 4
5
kdm,s k1 ≤ km,s , and kffm,s k1 ≤ km,s
6 1
kẋm,s k1 , where km,s ∼
where qm,s ∈ Rk×1 , q̇m,s ∈ Rk×1 , q̈m,s ∈ Rk×1 denote 6
km,s are the unknown constants determining the bounds of the
the joint position, velocity, and acceleration, respectively. according parameters.
xm,s ∈ R3×1 , ẋm,s ∈ R3×1 , and ẍm,s ∈ R3×1 denotes those
in operational space. k is the link number of a manipulator. Assumption 3. The uncertain parametric matrices/vectors,
Note that k may be different for the master and the slave in the Mm,s , Cm,s , and gm,s , can be divided into a known nominal
case of heterogeneous robots. Mm,s = Mm,s (qm,s ) ∈ R3×3 part plus an uncertainty varying from the nominal part, that
is the symmetric and uniformly positive definite inertia matrix. is, Mm,s = M̄m,s + ∆Mi , Cm,s = C̄m,s + ∆Cm,s , and
Cm,s = Cm,s (qm,s , q̇m,s ) ∈ R3×3 is the Coriolis and gm,s = ḡm,s + ∆gm,s , where M̄m,s ,C̄m,s , and ḡm,s are
Centrifugal loading matrix, and gm,s = gm,s (qm,s ) ∈ R3×1 is nominal matrices/vector. ∆Mm,s , ∆Cm,s , and ∆gm,s are
the gravitational loading vector. The control input is denoted unknown modeling uncertainties.
by um,s ∈ R3×1 . The human/operator force is denoted by Assumption 4. The time-varying network delays are bounded
fh ∈ R3×1 , and the environmental force fe ∈ R3×1 can be and can be divided into two parts: a known and constant
modeled in a general form [19] as nominal delay T̄ plus an unknown but bounded deviation δ(t),
→
− i.e., T (t) = T̄ + δ(t). δ(t) is changing sufficiently slowly so
fe = Be ẋs + Ke xs + Ce 1 = Θe Φe , (8)
that kδ̇(t)k1 ≤ 1.
where Be ∈ R3×3 , Ke ∈ R3×3 , and Ce ∈ R3×3 are
damping, stiffness and constant respectively, and Θe = According to Assumption 3, the models in (6) and (7) can
→
− be rewritten as
[Be Ke Ce ] ∈ R3×9 . 1 ∈ R3×1 is a vector of all ones.
T T → −T T
Φe = [ẋs xs 1 ] ∈ R9×1 denotes the regressor. M̄m,s ẍm,s + C̄m,s ẋm,s + ḡm,s = um,s + ρm,s ± fh,e , (10)
fnm,s = fnm,s (q̇m,s ) ∈ R3×1 contains the disturbances and
where ρm,s = fnm,s − (∆Mm,s ẍm,s + ∆Cm,s ẋm,s + ∆gm,s ).
frictions, that is, fnm,s = ffm,s + dm,s , where dm,s ∈ R3×1
Based on Assumption 2, ρm,s is bounded by
denotes the unknown external disturbances caused by set point
changes, low frequency load disturbances, etc. ffm,s ∈ R3×1 kρm,s k1 ≤ Θm,s Φm,s , (11)
denotes the equivalent frictions exerted on the end-effectors
where Θm,s represents the unknown parameters consisting of
resultant from the joint frictions. The friction model in this 1 6
km,s ∼ km,s , and Φm,s is the known scalar given by Φm,s =
paper takes into account the effects of the Coulumb friction,
1 + kxm,s k1 + kẋm,s k21 > 0.
static friction, viscous friction and Stribeck effect, and is given
Control Objective: When the teleoperation system formu-
as a function of the angular velocity q̇m,s [20] as
n lated by (10) is subject to parametric uncertainties, unknown
ffm,s = Jm,s γm,s 1 2
[tanh(γm,s q̇m,s ) − tanh(γm,s3
q̇m,s )] frictions, external disturbances, and time-varying delays, the
o objective is to design an ANTSM controller in operational
4 5 6
+γm,s tanh(γm,s q̇m,s ) + γm,s q̇m,s , (9) space to reduce the motion tracking errors between the master
and the slave in the sense of kxm (t) − xs (t)k1 ≤ , where the
p
where γm,s (p = 1 , 2 , . . . , 6 ) are unknown coefficients, bound ≥ 0 is desired to be as small as possible while the
3×k
Jm,s ∈ R is the Jacobian matrix. tracking stability is well maintained.
In addition, the dynamic models of the systems in (6) and
(7) satisfy the following properties and assumptions. IV. M AIN R ESULTS
Property 1. [19] For non-redundant manipulators in a non- In this section, an ANTSM control method is developed
singular configuration, the relationships of Mm,s , Cm,s , gm,s based on the control objective and the subsequent stability
matrices/vector in operational space with those in joint space analysis. The complete control system is illustrated in Fig.1,
−T jo −1 −T jo −1
are: Mm,s = Jm,s Mm,s Jm,s , Cm,s = Jm,s Cm,s Jm,s − where the master is controlled to follow the reference signal
where epm,s ∈ R3×1 is the position error and the velocity error
T1 (t ) T
xm is denoted by evm,s ∈ R3×1 for the subsequent design. T1 (t)
Master
is the time-varying network delays from the master side to
ANTSM Controller the slave side. Consider Assumption 4, we have T1,2 (t) =
um (Slave) T̄ + δ1,2 (t). xs (t − T̄ ) is the end effector position of the
fh ANTSM Control us slave manually delayed by T̄ . Then the position and velocity
(Master) fe
Slave
tracking errors are related by an auxiliary self-delayed velocity
error, εsd
s ∈ R
3×1
, given as εsd v p
s = es − ės . Furthermore, we
xr , xr , xr
xs
have
T
Reference fˆe Force ˆe Force

εsd
s = ẋs (t) − ẋs (t − T̄ ) − ẋm (t − T1 (t))(1 + δ̇1 (t)). (17)
T2 (t )
generator Predictor Estimator
Let the compensatory bound of self-delay-induced errors be
Fig. 1: Control architecture. Bsd

s = kẋs (t) − ẋs (t − T̄ )k1 + 2kẋm (t − T1 (t))k1 , (18)
such that kεsd
s k1 ≤ Bsd
holds. To facilitate the subsequent
s
generated from the human input and the predicted environ- analysis, another compensatory bound Bas is given by
mental force, while the slave controller aims at reducing the
Bas = 2||ẍm (t − T1 (t))||1 , (19)
phase shift effect between the master and slave trajectories.
Note that the time-derivative as well as low-pass filters (in to compensate the error introduced by time-varying delays
order to reduce the resultant noises and high frequencies) may when the acceleration signals are acquired by time derivative.
apply to the position signals when the velocity and acceleration
information are required [22]. C. Controller Development
A. Reference Signal Generator Based on the principle in (2), the sliding surfaces of sm,s ∈
To avoid the propagation of the force signal via network R3×1 for the control design are given by
channels, the environment force is transformed by a pair of sm = epm + βm (evm )αm , ss = eps + βs (evs )αs . (20)
force estimator/predictor so that a non-power signal of Θe in
(8) is estimated and transmitted. The adaptive controllers are designed as
In the master side, the environmental force is predicted by n −(ev )(2−αm )
m
um = C̄m ẋm + ḡm − fh + M̄m + ẍr
f̂e = Θ̂e (t − T2 (t))Φm , (12) αm βm
o
M+
where T2 (t) denotes the network delays from the slave side −M̄m Θ̂m Φm sgn(sm ) − κm sgn(sm ) , (21)
→
−
to master side. Φm = [ẋTm , xTm , 1 T ]T . Θ̂e is the estimate of n −(ev )(2−αs )
s
Θe given by the environmental force predictor [23] us = C̄s ẋs + ḡs + M̄s + ẍm (t − T1 (t))
αs βs
˙
Θ̂e = −HΦe (ΦTe Θ̂e − fe ),
1
(13)

− Bsd v (1−αs )
s diag(es ) + Bas sgn(ss )
d −1 αs βs
H = Φe ΦTe , H(0) = h0 I3×3 , h0 > 0. (14) o
dt −M̄sM + Θ̂s Φs sgn(ss ) − κs sgn(ss ) + fe , (22)
With the environmental forces predicted at the current time
step, a target impedance behavior of the master manipulator M+ +
where κm,s > 0, M̄m,s ≥ kM̄m,s k1 and Θ̂m,s is the estimate
is provided as follows to generate the reference signals: of Θm,s provided in Theorem 1. sgn(∗) represents the sign
Mr ẍr + Cr ẋr + gr = ksc1 fh − ksc2 f̂e , (15) function. Because (1 − αs ) < 0, (evs )(1−αs ) is operated
element-wisely as (r = 1, 2, 3)
3×1 3×1 3×1
where xr ∈ R , ẋr ∈ R , and ẍr ∈ R denote the v
reference position, velocity, and acceleration for the motion v (1−αs ) (es (r))(1−αs ) , if evs (r) 6= 0,
(es ) (r) =
control design. ksc1 and ksc2 are the scaling factors for the 0, otherwise.
human operator force and the estimated environmental force, Theorem 1. Consider the teleoperation systems as in (6) and
respectively. Mr ∈ R3×3 , Cr ∈ R3×3 , gr ∈ R3×1 are the (7), the compensatory bounds updated by (18) and (19), and
target parameters of the master behavior. the adaptive law given by
˙
(
B. Mixed-Type Feedback Signals and Compensatory Bounds Θ̃ m = −αm βm ksTm diag[(evm )(αm −1) ]k1 M̄m
M+
Φm ,
For the master and the slave control systems, the mixed-type ˙
Θ̃ T
= −α β ks diag[(e ) v (αs −1)
]k M̄ M+
Φ ,
s s s s s 1 s s
feedback signals for the subsequent control design are given
the proposed ANTSM controllers in (21) and (22) can guar-
as
 p antee the system stability and the sliding surface sm,s = 0 in
 em = xm − xr ,
 (20) to be reached in a finite time ts that satisfies
 v
em = ẋm − ẋr , √
(16) 1
 eps = xs (t − T̄ ) − xm (t − T1 (t)), 6Vs (xs (0)) 2

 v ts ≤ . (23)
es = ẋs (t) − ẋm (t − T1 (t)), 3αs βs κs k(evs )(αs −1) k1
See the proof in Appendix A. TABLE I: Dynamic parameters of the Phantom Omni device
Parameter Value Parameter Value
Proposition 1. During the convergence towards sm,s = 0, the
l1 (m) 0 l2 (m) 0.135
velocity error evm,s = 0 can be proved to be not an attractor l3 (m) 0.130 θ5 (kg · m) 9.1 × 10−3
for sm,s 6= 0 and Θ̃m,s 6= 0. θ1 (kg · m2 ) 3.7 × 10−3 θ6 (kg · m) 5.2 × 10−3
θ2 (kg · m2 ) 7.0 × 10−3 θ7 (N · m · s/rad) 0.096
See the proof in Appendix B. θ3 (kg · m2 ) 8.0 × 10−3 θ8 (N · m · s/rad) 0.145
θ4 (kg · m2 ) 0.4 × 10−3 θ9 (N · m · s/rad) 0.055
Theorem 2. After sm,s = 0 is satisfied, the ANTSM controllers
in (21) and (22) can ensure the finite-time synchronization
to the time-varying reference signal with bounded tracking with f = pi/10, and the virtual environmental force as
errors. By selecting the control gains to satisfy the following [0.025 0.040 0.030]T N . The resultant assigned velocity is
sufficient condition upper-bounded by Bv = 0.250 m/s. The random time-varying
1 1 network delay is given in a bound as T1 (t) = T2 (t) ∈
(4αs βs ) 1−αs < Bv T̄ 1−αs , (24)
[0.298, 0.302] sec with a nominal delay of T̄ = 0.300 sec.
using the mixed-type feedback signal is superior to the utiliza- The control gains are chosen as αm,s = 5/3 and βm,s = 1
tion of feedback without self-delays in terms of the tracking so as to satisfy the condition (24) in Theorem 2. κm,s =
1 2
accuracy. Bv denotes the reference velocity bound. 8. The frictional force coefficients are γm,s = 3, γm,s =
3 4 5 6
6, γm,s = 1, γm,s = 2, γm,s = 4, γm,s = 1. Assume that
See the proof in Appendix C.
the parameter uncertainties are half of the true values, that is,
Proposition 2. Utilizing the mixed-type feedback signal in the ∆Mm,s = 12 Mm,s , ∆Cm,s = 12 Cm,s , and ∆gm,s = 12 gm,s .
slave control design improves the velocity tracking between the Mr = diag[2 2 1.4], Cr = diag[1.5 2 1.5], gr = [2.5 2 1.6]T ,
master and the slave in the sense that and ksc1 = ksc2 = 1 as in (15). In practice, the sign function
in the sliding mode control method is usually replaced by a
kẋs (t − T̄ )k1 → kẋm (t − T1 (t))k1 ,
saturation function to reduce the chattering effect,
which explains the reduction of the delay-induced phase shift s
k ∆ k1 sgn(s), ksk1 < ∆,
in the position profiles between the master and slave. sat(s) = (25)
sgn(s), otherwise,
See the proof in Appendix D. where ∆ is the boundary layer width, and ∆ = 0.08.
Proposition 3. After ss = 0 is satisfied, provided that the
reference signal is constant with zero speed, the tracking errors A. Comparison of Different Types of Feedbacks
epm,s tend to converge to zero, that is, epm,s (t) → 0 as t → ts . This comparison is to demonstrate the effects of different
See the proof in Appendix E. types of feedback signals in (3)-(5), and illustrate the improv-
ing performance as claimed in Theorem 2. The controllers
V. N UMERICAL S IMULATIONS used in this comparison study are based on the schemes in
The numerical simulations were performed on the master-
slave manipulators represented by the mathematical model of a Reference Master Slave
3-link (k = 3) Phantom Omni haptic device with the following (a)
0.2
(b)
inertia and Coriolis matrices, and the gravity vector [22]. 0.2
    0
h11 0 0 0 0.1
-0.2
M =  0 h22 h23  , g = θ5 gc2 + θ6 gc23  , 0 10 20 30 40 0 10 20 30 40
0 h32 h33 θ6 gc23
(c) (d)
  0.2
−(a1 q̇2 ) −a1 q̇1 −a2 q̇1 0.2
C =  a1 q̇1 −a3 q̇3 −a3 (q̇2 + q̇3 ) , 0
0.1
a2 q̇1 a3 q̇2 0 -0.2
0 10 20 30 40 0 10 20 30 40
where h11 = θ1 + θ2 c22 + θ3 c223 + 2θ4 c2 c23 , h22 = θ2 + θ3 +
(e) (f)
2θ4 c3 , h23 = θ3 + θ4 c3 , h33 = θ3 , a1 = θ2 c2 s2 + θ3 c23 s23 + 0.2
0.2
θ4 c2∗23 , a2 = θ3 c23 s23 + θ4 c2 s23 , a3 = θ4 s3 . The relevant 0
parameters to calculate M , C, and g are given in Table I [24]. 0.1
-0.2
0 10 20 30 40 0 10 20 30 40
Unless otherwise specified, parameters were selected as fol-
lows for simulations. The initial joint positions of manipulators
were selected as (unit: rad), qs (0) = [0.700 0.600 − 0.500]T , Fig. 2: Comparisons of the tracking performance (along x and
qm (0) = [−0.300 0.300 − 0.800]T , and the initial joint y directions). Case I: (a) and (b) using feedback without self-
velocities are zero for all manipulators. The reference tra- delays (T = 0.3 sec); Case II: (c) and (d) using feedback
jectory of the end effector is assigned by setting the human without self-delays as in (4) (T = 0.02 sec); Case III: (e) and
force as [0.030sin(f t) + 0.100 0.070 0.040sin(f t)]T N (f) using the mixed-type feedback as in (5) (T = 0.3 sec).
TABLE II: Steady-state errors using three different controllers

Constant Reference Time-Varying Reference
Control Method Mean SD Max Min Mean SD Max Min
(×10−3 m) (×10−3 m) (×10−3 m) (×10−3 m) (×10−3 m) (×10−3 m) (×10−3 m) (×10−3 m)
P-like [25] 1.933 0.522 2.869 0.962 -0.947 33.933 74.104 -66.246
NTSM [16] -2.822 1.839 0.078 -4.808 -2.615 4.400 6.643 -10.613
ANTSM -1.284 1.143 -0.105 -3.179 -0.952 1.959 4.167 -3.684
SD: Standard Deviation.
(21) and (22) with different types of feedback signals. Fig.2 TABLE III: Comparisons of three control methods
shows the position tracking of three cases (along the x and y Features
Control Method
directions for example), namely, Feedback Adaptive Robust Force Trans.
Case I: (a) and (b) present the results of using feedback P-like [25] (3) N N WF
NTSM [16] (3) N Y DT
without self-delays as in (3) with T = 0.3 sec; ANTSM (3)&(5) Y Y Pdt
Case II: (c) and (d) are simulated by utilizing feedback Force Trans.: Force Transmission Method, N: Not considered, Y: Consid-
without self-delays as in (4) with T = 0.02 sec; ered, WF: Wave Form, DT: Direct Transmission, Pdt: Force Prediction,
Case III: (e) and (f) are simulated with the mixed-type
feedback as in (5) and T = 0.3 sec.
In Case II, the nominal delay interval was chosen as is able to accurately estimate the environmental force in
0.02 sec because the controllers fail to maintain the stability Case 1 and Case 2. Although suffering from the unknown
when T̄ > 0.02 sec in this simulation. Nevertheless, Fig.2 (c) or inexact information of system parameters, external noise
and (d) present accurate position tracking in x and y directions. bounds, time-varying network delays, and frictional forces.
In Fig.2 (a), stability can be guaranteed when the nominal Fig.5 and Fig.6 demonstrate that the developed controller
delay increases to 0.3 sec, but a phase shift induced by time still provides good tracking performance in both cases. By
delays is inevitable and therefore results in a lower tracking comparing the trajectories of the master and the slave with
accuracy. In Fig.2 (e) where the controllers use the mixed-type the reference trajectories, respectively, it can be seen that
feedback signal, the phenomenon of phase shift is eliminated, the tracking errors are within small bounds, where the er-
and meanwhile the tracking stability is well maintained. ror bound is [−3.179, −0.105] × 10−3 m in Case 1 and
By comparing the trajectories along x-direction and y- [−3.684, 4.167] × 10−3 m in Case 2. In addition, the end
direction for the three cases, when the reference trajectory effector positions of the master and slave can synchronize to
tends to be constant, it is clear that controllers using any the reference signals in a finite time (in about 5 sec).
of the three types of feedback signals can provide good
tracking performance in terms of the accuracy. This is in C. Comparisons with existing control methods
correspondence with Proposition 3 and explains the wide
To further illustrate the performance improvement using
adoption of the wait-and-move operation in many time-delayed
the proposed control method, comparisons with a model-
applications to avoid the consideration of phase shift.
free P-like control method in [25] and a model-based non-
singular terminal sliding mode (NTSM) control method in
B. Tracking Performance Illustration [16] are analysed. The P-like control method is based on the
By adopting the mixed-type feedback signals in the slave passivity theory and free from the requirement of the dynamic
controller, the performance of the bilateral teleoperation sys- parameters. Therefore, the parameter uncertainty does not
tem is demonstrated with two cases: affect the tracking performance. The NTSM method is model-
Case 1: The master-slave position tracking with constant based and provides high tracking accuracy when the dynamic
reference signals. parameters are exactly known, and it is robust to bounded
Case 2: The master-slave position tracking with time-varying external disturbances as well. The major features of the
reference signals. compared control methods are listed in Table.III, including the
The random network delays and external disturbance are utilization of feedback types, adaptive ability, robust ability,
depicted in Fig.3 (a) and (b). Fig.3 (c) and (d) present the and the way of the force transmission.
frictional forces in the two cases. In Case 1, as the reference The tracking error comparisons of the three controllers
trajectories are given as constant signals, the joints eventually are presented in Table.II. The results show that when the
settle down and stay put, which results in zero frictional reference is given as a constant signal, all three cases are
forces, while in Case 2, the non-zero frictions exist throughout able to track the reference with small error bounds. However,
the operation. The similar phenomenon can be found in the when given a time-varying reference, using P-like controller
compensatory bounds as well, as shown in Fig.4 (c)-(f). During results in large tracking errors (e.g., with a standard deviation
the operation, the compensatory bounds is evident during of 33.933 × 10−3 m). The NTSM method also presents
the transient phase, and then they rapidly reduce and vary the increasing tracking errors, although the errors vary less
within a small bound. Fig.4 (a) and (b) illustrate the good significantly than those using the P-like controller. In contrast,
adaptive ability of the proposed controller. From Fig.3 (e) the proposed ANTSM scheme can provide better tracking
and (f), it is shown that the environmental force predictor accuracy regardless the behavior of the reference.
Master Slave
(a) (b)
0 0
-0.5 -0.5
-1 -1
0 10 20 30 0 10 20 30
(c) (d)
2 2
1 1
0 0
0 10 20 30 0 10 20 30
(e) (f)
40 40
20 20
0 0
0 10 20 30 0 10 20 30
Fig. 3: (a) Network delays; (b) External disturbances; Fric- Fig. 4: (a) Estimation parameters Θ̂ in Case 1; (b) Estimation
tional forces in (c) Case 1 and (d) Case 2; Environmental parameters in Case 2; Compensatory bound Bsd s (c) in Case
force estimations for (e) Case 1 and (f) Case 2 where the solid 1 and (d) in Case 2; Compensatory bound Bas (e) in Case 1
lines and dash lines represent the true environment forces and and (f) in Case 2.
estimate forces, respectively.
Reference Master Slave Reference Master Slave

0.3 0.1 0.1
0.2
0.2 0 0
0.1
0.1 -0.1 0 -0.1
0 10 20 30 0 10 20 30 0 10 20 30 0 10 20 30
0.2 0.2 0.2 0.2
0 0 0 0
-0.2 -0.2 -0.2 -0.2

0 10 20 30 0 10 20 30 0 10 20 30 0 10 20 30
0.2 0.5 0.2 0.5
0 0 0 0
-0.2 -0.5 -0.2 -0.5

0 10 20 30 0 10 20 30 0 10 20 30 0 10 20 30
Fig. 5: Position tracking and the tracking errors of the master- Fig. 6: Position tracking and the tracking errors of the master-
slave teleoperation with constant reference signals. slave teleoperation with time-varying reference signals.
Fig. 8: (a) Measured one-way network delays in a LAN; (b)

Mean value of delays and the standard deviation.
https://www.3dsystems.com) served as the master and slave

to test practical feasibility and the performance of the devel-
oped ANTSM controller. The haptic device has joint position
encoder with a resolution of approximately 4 × 10−4 rad.
Fig. 7: The experimental testbed. The experimental setup is shown in Fig.7. Two devices are
run in two different computers connected through a wireless
VI. E XPERIMENTAL R ESULTS local area network (LAN). The control software used to imple-
ment the measurement and network communication is Quanser
In this section, experiments were conducted on a pair of
QUARC 2.6, running in Matlab R2017a (64-bit) environment.
6-DoF Phantom Omni haptic devices (with 3 actuated joints,
level of chattering effect can still be observed. By observing

the slave manipulator, during the steady-state phase, the two
compensatory bounds are bounded within a small range, i.e.,
Bsd a 2
s ≤ 0.004 m/s and Bs ≤ 0.150 m/s , as depicted in
Fig.10 (c) and (d). Fig.10 (e) demonstrates the estimation
parameters for the master and slave systems. As the trajectories
are time-varying, the estimation parameters do not converge to
a constant. Fig.10 (f) depicts good ability of the force predictor
to estimate the virtual environment forces.
VII. C ONCLUSION
This paper has presented a bilateral teleoperation system
in the presence of time-varying network delays. The ANTSM
controllers using the mixture of feedback signals as well as
the compensatory bounds estimated online to reduce the phase
Fig. 9: Position tracking and the tracking error of the master-
shift and therefore ensure the improved tracking synchroniza-
slave system with time-varying reference signal.
tion of the end-effectors when the manipulators are subjected
to parametric uncertainties, unknown frictions and external
disturbances. The simulations and experiments have shown the
effectiveness of the ANTSM in teleoperation systems.
A PPENDIX A
P ROOF OF T HEOREM 1
In the proof of Theorem 1 as well as the proof in subsequent
appendices, we take the slave system as an example in the
analysis and the proof for the master follows the similar steps.
Proof. Substituting the control inputs into the dynamics in
(10), we have the closed-loop dynamics as
ẍs − ẍm (t − T1 (t)) = M̄i+ ρs − M̄sM + Θ̂s Φs sgn(ss )
(ev )(2−αs ) 1
− s − Bsd diag(evs )(1−αs ) sgn(ss )
αs βs αs βs s
Fig. 10: Filtered control inputs for (a) the master and (b) −Bas sgn(ss ) − κs (t)sgn(ss ). (26)
the slave; The compensatory bounds: (c) Bsd a
s and (d) Bs ; (e)
Estimation parameters; (f) The estimated environmental forces The left-hand side of (26) has the relationship as ẍs −
(solid lines) and the given environmental forces (dash lines). ẍm (t − T1 (t)) = ėvs + εas , where εas = ẍm (t − T1 (t))(1 −
δ̇1 (t)) whose upper-bound is given by (19). Multiplying
αs βs diag[(evs )(αs −1) ] at both sides of (26) leads to

The Simulink model is capable of running in real time at an ṡi = αs βs diag[(evs )(αs −1) ] εas − Bas sgn(ss ) + M̄i+ ρs
update rate of up to 1000 Hz. The measured one-way delays
of User Datagram Protocol (UDP) send/receive are shown in −M̄sM + Θ̂s Φs sgn(ss ) − κs (t)sgn(ss )
Fig.8, where the average network delay of T̄ = 0.042 sec
+εsd sd
s − Bs sgn(ss ). (27)
and the standard error of σ = 0.004 sec are obtained by
computing the normal distribution probability density function The stability of the closed-loop system can be shown by
(ND-pdf). The control gains were selected as αm = αs = 7/5, defining the following Lyapunov function
βm = βs = 1, and κm = κs = 8. 1 1
Tests with time-varying reference trajectories were per- Vs = sTs ss + Θ̃2s , (28)
2 2
formed and the tracking performance is presented in Fig.9 (a)-
(c), which show the good tracking synchronization between the where Θ̃s is the estimating error of Θs , i.e., Θ̃s = Θs − Θ̂s ,
˙ ˙ .
master and the slave. Accordingly, in Fig.9 (d)-(f), the small and Θ̂s = Θ̃s
and bounded tracking errors can be achieved within about Using (27) and taking derivative of Vs give
[−0.005 0.005] m along each direction. For safety’s sake, a ˙
V̇s = sTs ṡs + Θ̃s Θ̃ s
saturation block in the Simulink model is used in joint space
˙

to constrain the inputs such that τm,s ∈ [−0.200 0.200] N m, = Θ̃i Θ̃i + αs βs diag[(evs )(αs −1) ] sTs εas − Bas sTs sgn(ss )
as shown in Fig.10 (a) and (b) for the master and the slave,
+M̄i+ sTs ρs − M̄sM + Θ̂s Φs sTs sgn(ss )
respectively. The chattering effect in the control input signals
is reduced by using the saturation function, although a certain −κs (t)sTs sgn(ss ) + sTs εsd sd T
s − Bs ss sgn(ss ).
As sTs sgn(ss ) ≥ 0, and according to the definitions of the Assume that with proper controller, the position error with-
compensatory bounds, we have the following inequalities: out self-delay can idealy converge to zero, that is, ēps → 0 as
sTs εsd sd T T a a T
s − Bs ss sgn(ss ) ≤ 0, and ss εs − Bi ss sgn(ss ) ≤ 0, t → ∞ [25] or in a finite time [16]. It is also assumed that the
Then, V̇s becomes approximation of T̄ ẋm (t) = xm (t) − xm (t − T̄ ) holds. From
˙ + α β diag[(ev )(αs −1) ] kM̄ + k sT Θ Φ
(18) and (43), we obtain kevs k1 = kεsd s k1 ≤ 4Bv , and then
V̇s ≤ Θ̃s Θ̃ s s s s s 1 s s s the bound of ε̄sm (t) and εsm (t − T̄ ) are given as follows

−M̄sM + Θ̂s Φs sTs sgn(ss ) − κs (t)sTs sgn(ss ) . kε̄sm (t)k1 = T1 (t)kẋm (t)k1 , (34)
αs
Let the adaptive law be kεsm (t − T̄ )k1 ≤ βs (4Bv ) + δ1 (t)kẋm (t − T1 (t))k1 . (35)
˙ = −α β M̄ M + sT diag[(ev )(αs −1) ]Φ .
Θ̃ (29) Throughout the entire manipulation, it is reasonable to as-
s s s i s s s
sume that the inequality in (35) is equivalent to the expression:
Then V̇i becomes
kεsm (t)k1 ≤ βs (4Bv )αs + δ1 (t)kẋm (t)k1 . (36)
V̇s ≤ −αs βs κs kss k1 diag[(evs )(αs −1) ] ≤ 0. (30)
When evs 6= 0, the finite time stability of the delayed systems To facilitate the description of the performance improvement
can be realized. According to Lemma 1, the sliding surface in terms of the tracking error, an auxiliary error is defined as
ss = 0 is reachable in a finite time ts , Esp = kεsm (t)k1 − kε̄sm (t)k1
√ 1
6Vs (xs (0)) 2 ≤ βs (4Bv )αs − T̄ kẋm (t)k1 , (37)
ts ≤ . (31)
3αs βs κs k(evs )(αs −1) k1
which can be interpreted as follows:
Following the similar steps, the stability of the master system 1) when Esp > 0, using the mixed-type feedback signals
can be proved. This concludes the proof of Theorem 1. results in larger tracking error than using the feedback signals
without self-delays.
A PPENDIX B 2) when Esp = 0, using the mixed-type feedback signals
P ROOF OF P ROPOSITION 1 have the same tracking error as using the feedback signals
Proof. Taking the slave system as an example, provided that without self-delays.
ss 6= 0 and Θ̃i 6= 0 when evs = 0, the following simplified 3) when Esp < 0, using the mixed-type feedback signals
expression can be obtained after substituting the control input provides smaller tracking error than using the feedback signals
in (22) into the dynamics in (7): without self-delays.
Provided the velocity bound Bv (as given in (43)) and
ėvs = M̄i+ ρs − εas − (M̄sM + Θ̂s Φs + Bas + κs (t))sgn(ss ).
nominal network delay interval T̄ , from (37), the control gains
Consider the properties of the vector norm: (1) x ∈ Rm×1 and are chosen to satisfy the following sufficient condition
y ∈ Rm×1 (m = 1, 2, · · · ), kxk1 = 0 if and only if x = 0;
(2) |kxk1 − kyk1 | ≤ kx − yk1 , we have βs (4Bv )αs − T̄ kẋm (t)k1 < 0, (38)
kėvs k1 ≥ kM̄s+ ρs − M̄sM + Θ̂s Φs sgn(ss )k1 such that Esp < 0 holds. Rewriting (38) give the following
sufficient condition for the control gain selection
−kκs (t))sgn(ss )k1 − kεas + Bas sgn(ss )k1 . 1 1
(4αs βs ) 1−αs < Bv T̄ 1−αs . (39)
Based on the definitions of the compensatory bounds and
This concludes the proof of Theorem 2.
adaption bounds, we have kėvs k1 > 0 and therefore ėvs 6= 0.
This shows that evs = 0 is not an attractor for ss 6= 0 and
Θ̃s 6= 0 [16]. This concludes the proof of Proposition 1 A PPENDIX D
P ROOF OF P ROPOSITION 2
A PPENDIX C Proof. When the sliding surface ss = 0 is reached, (20)
P ROOF OF T HEOREM 2 becomes eps + βs (evs )αs = 0 whose time derivative gives
Proof. Denote the feedback signal without self-delay as 1
ėvs = diag[(evs )1−αs ](−evs + εsd
s ). (40)
ēps = xs (t) − xm (t − T1 (t)) αs βs
= xs (t) − xm (t) + xm (t) − xm (t − T1 (t)) Based on the definition of αs , (evi )1−αs ≥ 0 is always
= ε̄sm (t) + [xm (t) − xm (t − T1 (t))], (32) satisfied. Therefore, (40) can be interpreted by the element-
wise inequalities as (p = 1, 2, 3),
where ε̄sm (t) = xs (t) − xm (t) is the instant tracking error v
between the master and slave. From (16) and (20), the self- ės (p) ≤ 0 when −evs (p) + εsd
s (p) ≤ 0,
(41)
delayed position error becomes ėvs (p) ≥ 0 when −evs (p) + εsd
s (p) ≥ 0.
eps = xs (t − T̄ ) − xm (t − T1 (t)) The monotonicity in (41) implies the reachability of a switch-

= εsm (t − T̄ ) + [xm (t − T̄ ) − xm (t − T1 (t))] ing surface given by
= −βs (evs )αs . (33) abs(evs ) = abs(εsd v sd
s ), with sgn(es ) = sgn(εs ). (42)
As the switching surface in (42) is reached, using (16) and [14] M. Zhihong, A. P. Paplinski, and H. R. Wu, “A robust mimo terminal
(17), we obtain sliding mode control scheme for rigid robotic manipulators,” IEEE
transactions on automatic control, vol. 39, no. 12, pp. 2464–2469, 1994.
[15] Y. Feng, X. Yu, and Z. Man, “Non-singular terminal sliding mode control
evs − εsd
s = ẋs (t − T̄ ) − ẋm (t − T1 (t))δ̇1 → 0. (43) of rigid manipulators,” Automatica, vol. 38, no. 12, pp. 2159–2167,
2002.
As a result, it is straightforward that kẋs (t − T̄ )k1 → kẋm (t − [16] S. Khoo, L. Xie, and Z. Man, “Robust finite-time consensus tracking
T1 (t))k1 ≤ Bv , where Bv denotes the velocity bound. This algorithm for multirobot systems,” IEEE/ASME transactions on mecha-
concludes the proof of Proposition 2. tronics, vol. 14, no. 2, pp. 219–228, 2009.
[17] S. P. Bhat and D. S. Bernstein, “Finite-time stability of continuous
autonomous systems,” SIAM Journal on Control and Optimization,
vol. 38, no. 3, pp. 751–766, 2000.
A PPENDIX E [18] U. Munz, A. Papachristodoulou, and F. Allgower, “Delay robustness
P ROOF OF P ROPOSITION 3 in non-identical multi-agent systems,” IEEE Transactions on Automatic
Control, vol. 57, no. 6, pp. 1597–1603, 2012.
Proof. Provided that ẋm (t) = 0 is given by the master side, [19] Z. Chen, Y.-J. Pan, and J. Gu, “Integrated adaptive robust control
and consider Proposition 2, ẋs (t) = 0 and Bv = 0 can be for multilateral teleoperation systems under arbitrary time delays,”
achieved after the sliding surface ss = 0 is reached. Therefore, International Journal of Robust and Nonlinear Control, vol. 26, no. 12,
pp. 2708–2728, 2016.
based on (35), the tracking error can be closed in the sense of [20] C. Makkar, G. Hu, W. G. Sawyer, and W. E. Dixon, “Lyapunov-based
εsm (t) = xs (t) − xm (t) = 0. This concludes the proof. tracking control in the presence of uncertain nonlinear parameterizable
friction,” IEEE Transactions on Automatic Control, vol. 52, no. 10,
pp. 1988–1994, 2007.
ACKNOWLEDGMENT [21] A. Ben-Israel and T. N. Greville, Generalized inverses: theory and
applications, vol. 15. Springer Science & Business Media, 2003.
This research was supported in part by the Natural Sciences [22] S. Forbrigger, “Prediction-based haptic interfaces to improve trans-
and Engineering Research Council of Canada (NSERC), the parency for complex virtual environments,” 2017.
[23] K. Yoshida and T. Namerikawa, “Stability and tracking properties in
Nova Scotia Graduate Scholarship (NSGS), and the China predictive control with adaptation for bilateral teleoperation,” in 2009
Scholarship Council (CSC). American Control Conference, pp. 1323–1328, IEEE, 2009.
[24] S. Forbrigger and Y.-J. Pan, “Improving haptic transparency for un-
certain virtual environments using adaptive control and gain-scheduled
R EFERENCES prediction,” IEEE transactions on haptics, vol. 11, no. 4, pp. 543–554,
2018.
[1] M. W. Carey, E. M. Kurz, J. D. Matte, T. D. Perrault, and T. Padir, [25] E. Nuño, L. Basañez, R. Ortega, and M. W. Spong, “Position tracking
“Novel eod robot design with dexterous gripper and intuitive teleoper- for non-linear teleoperators with variable time delay,” The International
ation,” in World Automation Congress 2012, pp. 1–6, IEEE, 2012. Journal of Robotics Research, vol. 28, no. 7, pp. 895–910, 2009.
[2] P. Ridao, M. Carreras, E. Hernandez, and N. Palomeras, “Underwater
telerobotics for collaborative research,” in Advances in telerobotics,
pp. 347–359, Springer, 2007.
[3] M. Lirici, V. Papaspyropoulos, and L. Angelini, “Telerobotics in
medicine and surgery,” Minimally Invasive Therapy & Allied Technolo-
gies, vol. 6, no. 5-6, pp. 364–378, 1997.
[4] Y. Tipsuwan and M.-Y. Chow, “Control methodologies in networked Henghua Shen (S’16) received the B.E. degree in
control systems,” Control engineering practice, vol. 11, no. 10, Mechanical Engineering from Fuzhou University in
pp. 1099–1111, 2003. 2012, the M.E. degree in Mechanical Engineering
[5] U. Ahmad and Y.-J. Pan, “A time domain passivity approach for from Fuzhou University in 2015. Currently, he is
asymmetric multilateral teleoperation system,” IEEE Access, vol. 6, pursuing his PhD degree in Mechanical Engineer-
pp. 519–531, 2017. ing at Dalhousie University, Halifax, Nova Scotia,
[6] K. Haninger and M. Tomizuka, “Robust passivity and passivity re- Canada. His research interests are in the area of
laxation for impedance control of flexible-joint robots with inner-loop computer vision, robotics and networked control
torque control,” IEEE/ASME Transactions on Mechatronics, vol. 23, systems. He is a student member of IEEE, IEEE
no. 6, pp. 2671–2680, 2018. Control Systems Society.
[7] H. Shen, Y.-J. Pan, and B. He, “Teleoperation of multiple cooperative
slave manipulators using graph-based non-singular terminal sliding-
mode control,” in Robotics and Biomimetics (ROBIO), 2017 IEEE
International Conference on, pp. 1430–1435, IEEE, 2017.
[8] Y.-C. Liu and M.-H. Khong, “Adaptive control for nonlinear teleopera-
tors with uncertain kinematics and dynamics,” IEEE/ASME Transactions
on Mechatronics, vol. 20, no. 5, pp. 2550–2562, 2015. Ya-Jun Pan (S’00-M’03-SM’11) is a Professor in
[9] H. Jin, Z. Liu, H. Zhang, Y. Liu, and J. Zhao, “A dynamic parameter the Dept. of Mechanical Engineering at Dalhousie
identification method for flexible joints based on adaptive control,” University, Canada. She received the B.E. degree
IEEE/ASME Transactions on Mechatronics, vol. 23, no. 6, pp. 2896– from Yanshan Univ. (1996), the M.E. degree in
2908, 2018. Mechanical Engineering from Zhejiang Univ. (1999)
[10] E. Nuno, “Consensus of euler-lagrange systems using only position and the Ph.D degree in Electrical and Computer En-
measurements,” IEEE Transactions on Control of Network Systems, gineering from the National University of Singapore
2016. (2003). She held post-doctoral positions of CNRS in
[11] J. R. Klotz, S. Obuz, Z. Kan, and W. E. Dixon, “Synchronization of the Laboratoire d’Automatique de Grenoble (GIPSA
uncertain euler–lagrange systems with uncertain time-varying communi- Lab), France and the Dept. of Electrical and Com-
cation delays,” IEEE transactions on cybernetics, vol. 48, no. 2, pp. 807– puter Engineering at the Univ. of Alberta, Canada.
817, 2018. Her areas of research are on the robust nonlinear control, cyber physical
[12] Y.-J. Pan, H. Marquez, and T. Chen, “Stabilization of remote control systems, intelligent transportation systems and collaborative autonomous
systems with unknown time varying delays by lmi techniques,” Interna- vehicles. She is currently the Associate/Technical Editors for IEEE Trans. on
tional Journal of Control, vol. 79, no. 07, pp. 752–763, 2006. Industrial Electronics, IEEE/ASME Trans. on Mechatronics, and IEEE Trans.
[13] M. Zhihong, M. O’day, and X. Yu, “A robust adaptive terminal sliding on Cybernetics. She is a Fellow of ASME, a senior member of IEEE, and a
mode control for rigid robotic manipulators,” Journal of Intelligent and Registered P.Eng. in Nova Scotia, Canada.
Robotic systems, vol. 24, no. 1, pp. 23–41, 1999.

Improving Tracking Performance of Nonlinear Uncertain Bilateral Teleoperation Systems With Time-Varying Delays and Disturbances

Uploaded by

Copyright:

Available Formats

Improving Tracking Performance of Nonlinear Uncertain Bilateral Teleoperation Systems With Time-Varying Delays and Disturbances

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Improving Tracking Performance of Nonlinear Uncertain Bilateral Teleoperation Systems With Time-Varying Delays and Disturbances

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

Improving Tracking Performance of Nonlinear

Reference fˆe Force ˆe Force

Fig. 1: Control architecture. Bsd

TABLE II: Steady-state errors using three different controllers

Reference Master Slave Reference Master Slave

0.2 0.2 0.2 0.2

-0.2 -0.2 -0.2 -0.2

0.2 0.5 0.2 0.5

-0.2 -0.5 -0.2 -0.5

Fig. 8: (a) Measured one-way network delays in a LAN; (b)

https://www.3dsystems.com) served as the master and slave

level of chattering effect can still be observed. By observing

eps = xs (t − T̄ ) − xm (t − T1 (t)) The monotonicity in (41) implies the reachability of a switch-

You might also like