2011 50th IEEE Conference on Decision and Control and
European Control Conference (CDC-ECC)
Orlando, FL, USA, December 12-15, 2011
Control of Nonlinear Bilateral Teleoperation Systems Subject to
Disturbances
Alireza Mohammadi, Mahdi Tavakoli and Horacio J. Marquez
Abstract— Teleoperation systems, consisting of a pair of
master and slave robots are subject to different types of
disturbances such as joint frictions, varying contact points, unmodeled dynamics and unknown payloads. Such disturbances,
when unaccounted for, cause poor teleoperation transparency
and even instability. This paper presents a novel nonlinear
bilateral control scheme, based on the concept of disturbance
observer based control, to counter these disturbances and their
negative effects on the teleoperation systems. The proposed
disturbance observer based bilateral control law is able to
acheive global asymptotic force tracking, and global exponential
position and disturbance tracking in the presence of various
disturbances. The minimum exponential convergence rate of
the position and the disturbance tracking errors can be tuned
by the controller parameters. Simulations are presented to show
the effectiveness of the proposed control scheme.
I. INTRODUCTION
Every teleoperation system consists of a master robot and
a slave robot. The master interacts with a human operator
and the slave interacts with a remote environment. If force
feedback from the slave side to the master side is provided,
the system is called a bilateral teleoperation system to distinguish it from a unilateral teleoperation system. A bilateral
teleoperation system is said to be transparent if the slave
robot precisely follows the position of the master robot and
the master robot faithfully displays the slave-environment
contact force to the human operator.
The most successful control scheme in achieving a fully
transparent teleoperation system is the 4-channel architecture
[1], [2], which is mostly suitable for teleoperation systems
with fixed linear models. Physical robots, however, are
nonlinear systems subject to various disturbances, such as
unknown dynamics, joint frictions, unknown payloads, etc.
[3]. Such disturbances, when unaccounted for, cause poor
teleoperation transparency and even instability. One way
to suppress these disturbances, is to employ disturbance
observers [4]. Disturbance observer based control has been
used in applications such as independent robot joint control
[5], and robot joint friction estimation and compensation [6],
[7].
A considerable part of the existing literature on disturbance
observer design for robotic applications uses linearized models or linear system techniques [8], [9]. In order to overcome
This work was supported by the Natural Sciences and Engineering
Research Council of Canada (NSERC).
A. Mohammadi was with the Department of Electrical & Computer
Engineering, University of Alberta, Edmonton, AB T6G 2V4 Canada. He is
now with the Edward S. Rogers Sr. Department of Electrical & Computer
Engineering, University of Toronto, Toronto, ON M5S 3G4 Canada. Email:
the linear disturbance observer limitations for the highly
nonlinear and coupled dynamics of robotic manipulators,
Chen et al. proposed a nonlinear disturbance observer for
nonlinear robotic manipulators and designed it such that
no acceleration measurement was needed [10]. However,
the closed-loop stability of the overall system including the
disturbance observer and the controller was not investigated.
The investigation of the stability and performance of a
master-slave teleoperation system under disturbance observer
based control is even more challenging and not studied
either. While a distrubance observer based controller was
designed for bilateral teleoperation systems in [11], the
master and the slave robots were considered to be linear
with only one DOF. This serves as the motivation to look
for a disturbance observer based control law for nonlinear
and n-DOF teleoperation systems.
This paper1 addresses the problem of disturbance observer
based control of nonlinear teleoperation systems. A disturbance observer based control law will be proposed and
incorporated into the framework of the 4-channel teleoperation architecture. Under the proposed control law, full transparency and exponential disturbance and position tracking
are achieved.
The organization of this paper is as follows. Section II
introduces the nonlinear model of the teleoperation systems
and 4-channel bilateral control architecture. Section III proposes a novel disturbance observer based controller for nonlinear teleoperation systems subject to various disturbances.
The teleoperation system transparency is also addressed in
this section. Finally, simulations in section IV show the
efficiency of the proposed control scheme as compared with
the case where no disturbance observer is employed.
II. NONLINEAR MODEL OF A TELEOPERATION
SYSTEM
The dynamical models investigated in this paper and our
proposed control law in section III will be in the Cartesian
space. This enables us to make a teleoperation system
transparent without requiring the master and the slave robots
to have similar kinematics and dynamics.
A. Model of a teleoperation system in the Cartesian space
The dynamic equations describing the motions of the endeffectors of n-DOF master and slave robots, which interact
with the human operator and the remote environment, in the
presence of disturbances can be written as
alireza.mohammadi@mail.utoronto.ca
M. Tavakoli is with the Department of Electrical & Computer Engineering, University of Alberta, Edmonton, AB T6G 2V4 Canada. Email:
Mxm (qm )ẍm + Nxm (qm , q̇m ) =
Mxs (qs )ẍs + Nxs (qs , q̇s ) =
tavakoli@ece.ualberta.ca
H. J. Marquez is with the Department of Electrical & Computer Engineering, University of Alberta, Edmonton, AB T6G 2V4 Canada. Email:
marquez@ece.ualberta.ca
978-1-61284-799-3/11/$26.00 ©2011 IEEE
1765
1A
fm + fh + dm
fs − fe + ds
more complete version of this paper can be found in [12].
(1)
(2)
where qm , qs , xm , xs , Mxm (qm ), Mxs (qs ), Nxm (qm , q̇m )
and Nxs (qs , q̇s ) are the vectors of joint positions, the position/orientation (pose) vectors of the end-effectors in the
Cartesian space, the inertia matrices in the Cartesian space,
and the vectors of Coriolis/centrifugal and gravity forces in
the Cartesian space. The subscripts m and s refer to the
master and the slave robots, respectively. Also, fm , fs ∈ R6×1
are the control forces applied to the master and the slave
end-effectors in the Cartesian space, fh , fe ∈ R6×1 are the
forces exerted to the master and the slave end-effectors by the
human operator and the remote environment, and dm , ds ∈
R6×1 are the disturbance forces exerted to the master and the
slave end-effectors in the Cartesian space. These disturbance
forces represent the lumped effect of all disturbances acting
on the master and the slave robots such as joint frictions,
varying contact points, unmodeled dynamics and unknown
payloads in the Cartesian space.
B. 4-channel control architecture
C1
C2
C3
C4
=
=
=
=
Zs + Cs
I + C6
I + C5
−Zm − Cm
(3)
III. DISTURBANCE OBSERVER BASED CONTROL
OF TELEOPERATION SYSTEMS
In this section, first the concept of disturbance observer
based control is introduced. Next, control laws are developed
based on this concept and incorporated into the 4-channel
architecture for a nonlinear teleoperation system.
A. Disturbance observer based control concept
Figure 2 depicts a disturbance observer which is used for
attenuating disturbances acting on a single robot. The vector
d represents the lumped disturbance, which deteriorates
the tracking performance of the robot control system. The
disturbance observer role is to estimate this disturbance as
closely as possible. The estimated disturbance d̂ is then
subtracted from the control signal f to cancel out or minimize
the effect of the disturbance.
Fig. 2. Disturbance observer used for attenuating disturbances acting on
a single robot.
Fig. 1.
4-channel teleoperation control architecture.
The 4-channel control architecture is shown in Figure 1.
Exogenous signals f∗h and f∗e are exerted by the human operator and the remote environment, respectively. The signals xm ,
xs , fh , fe , fm , fs , dm and ds are as defined in (1)–(2). Position
information is exchanged between the master and the slave
via the position channels C1 and C4 . Force information is
exchanged through the force channels C2 and C3 . In addition,
Cm and Cs are local master and slave (position) controllers.
Lastly, C5 and C6 provide the master and the slave with
local force feedback from the human operator and the remote
environment, respectively. In the conventional design of 4channel controllers, it is assumed that dm = 0 and ds = 0.
We are, however, going to deal with the case where these
disturbances are not zero.
In a fully transparent teleoperation system we have xs = xm
and fh = fe . If the master and the slave are represented by
LTI impedances Zm (s) and Zs (s), the 4-channel teleoperation
system of Figure 1 becomes fully transparent in the absence
of disturbances and delays if the controllers are chosen as
[13]
B. Proposed teleoperation control laws
Extending the idea of disturbance observer based control
of a single robot to a master-slave teleoperation system, we
will design a disturbance observer for each of the master
and the slave robots in order to estimate and cancel out the
disturbances.
Assume that the master and the slave disturbances are
estimated to be d̂m and d̂s , respectively. We propose the
following nonlinear control laws for the master and slave
robots described by (1) and (2), respectively:
fm
fs
= Mxm (qm )[−Cm xm − C2 fe − C4 xs + C6 fh + fh ]
+Nxm (qm , q̇m ) − fh − d̂m
(4)
= Mxs (qs )[−Cs xs + C1 xm + C3 fh − C5 fe − fe ]
(5)
+Nxs (qs , q̇s ) + fe − d̂s
where Cm , Cs , C1 , . . . , and C6 are some LTI controllers
used in the above nonlinear control laws. Note the use of
disturbance estimates d̂m and d̂s in the proposed control laws.
The disturbance observer based control laws (4) and (5),
when applied to the master and slave described by (1)–(2),
result in the following closed-loop equations for the two
robots:
1766
ẍm
ẍs
= −Cm xm − C2 fe − C4 xs + C6 fh
+fh + M−1
xm (qm )∆dm
= −Cs xs + C1 xm + C3 fh − C5 fe
−fe + M−1
xs (qs )∆ds
Ψxs (qs )∆ds
∆ẍ + Kv ∆ẋ + K p ∆x = Ψxm (qm )∆dm −Ψ
(6)
Cm
Cs
= Kmv s + Kmp
= Ksv s + Ksp
where
(7)
where ∆dm = dm − d̂m and ∆ds = ds − d̂s are the master and
the slave disturbance estimation errors, respectively.
Remark 1. When ∆dm = 0 and ∆ds = 0, i.e., under ideal
disturbance tracking, the closed-loop system equations (6)
and (7) describe an n-DOF teleoperation system without
disturbances, similar to the one shown in Figure 1 with
the master and slave robots represented by identity inertia
matrices, i.e., Zm (s) = s2 I and Zs (s) = s2 I. ♦
Let us choose the master and the slave local position
controllers in (4) and (5) to be of proportional-derivative
type:
(8)
= Cm f
= Cs f
(9)
where Cm f and Cs f are constant force reflection gain matrices. We choose the other controllers in (4) and (5) according
to (3) to satisfy the full transparency conditions:
C1
= s2 I + Ksv s + Ksp
C4
C5
C6
= −(s2 I + Kmv s + Kmp )
= Cs f − I
= Cm f − I
(10)
Remark 2. In order to implement C1 and C4 in (10), we
need to measure or compute the acceleration of the master
and the slave robots. We can omit the acceleration terms if
good low-frequency transparency is enough in the desired
application. However, requiring good transparency over both
low and high frequencies justifies using accelerometers [2].
♦
Using (8), (9) and (10) in the master and the slave closedloop dynamics (6) and (7) result in
∆ẍ = −Kmv ∆ẋ − Kmp ∆x + Cm f ∆f + M−1
xm (qm )∆dm
(11)
−1
∆ẍ = Ksv ∆ẋ + Ksp ∆x − Cs f ∆f − Mxs (qs )∆ds
(12)
where ∆x = xm − xs is the position tracking error and ∆f =
fh − fe is the force tracking error. Assume that matrices Cm f ,
−1
−1
Cs f and C−1
m f + Cs f are invertible. Multiplying (11) by Cm f
−1
and (12) by Cs f and adding them together, we can find the
dynamic equation governing the position tracking error
Kv
−1 −1
−1
−1
= (C−1
m f + Cs f ) (Cm f Kmv + Cs f Ksv )
Kp
−1 −1
−1
−1
(C−1
m f + Cs f ) (Cm f Kmp + Cs f Ksp )
(14)
=
Ψ xm (qm ) =
Ψ xs (qs ) =
−1 −1 −1 −1
(C−1
m f + Cs f ) Cm f Mxm (qm )
−1 −1 −1 −1
(C−1
m f + Cs f ) Cs f Mxs (qs )
(15)
(16)
(17)
C. Proposed disturbance observers
In this section, we will design the disturbance observers
in a way that full transparency and disturbance tracking are
acheived under the control laws (4) and (5) with choices
in (8), (9) and (10). Our proposed disturbance observers
designed in the Cartesian space given the master and slave
dynamics (1)–(2) are
where Kmv , Kmp , Ksv and Ksp are constant gain matrices.
Also, let us choose the force reflection gains in (4) and (5)
to be
C2
C3
(13)
d̂˙ m
= −Lm d̂m + Lm [Mxm (qm )ẍm + Nxm (qm , q̇m )
ΨTxm (qm )(∆ẋ + γ∆x)
(18)
−fh − fm ] +Ψ
d̂˙ s
= −Ls d̂s + Ls [Mxs (qs )ẍs + Nxs (qs , q̇s )
ΨTxs (qs )(−∆ẋ − γ∆x)
+fe − fs ] +Ψ
(19)
where γ is an arbitrary positive constant. Also, Lm and Ls
are constant gain matrices.
Note that in the 4-channel teleoperation control architecture, it is assumed that fh and fe are measured and the same
measurements will be used in (18) and (19). The proposed
nonlinear disturbance observers in (18) and (19) also need
acceleration measurements. As mentioned before, the need
for full transparency in a wide frequency range justifies
using accelerometer – again, the same measurements will be
needed in (18) and (19). Equations (1) and (2), along with
(18) and (19), result in the following disturbance estimation
error dynamics:
d̂˙ m
d̂˙
s
ΨTxm (qm )(∆ẋ + γ∆x)
= Lm ∆dm +Ψ
(20)
ΨTxs (qs )(−∆ẋ − γ∆x)
Ls ∆ds +Ψ
(21)
=
Thoroughout the paper, we assume that the rate of change
of the lumped disturbance is neglibgible in comparsion with
the estimation error dynamics, i.e. ḋm ∼
= 0 and ḋs ∼
= 0.
This assumption is not overly restrictive and is commonly
encountered in the robotics literature (see, for example, [10]).
Thus, the following disturbance estimation error dynamics
result from (20) and (21) for the master and the slave,
respectively:
∆ḋm
∆ḋs
ΨTxm (qm )(∆ẋ + γ∆x)
= −Lm ∆dm −Ψ
ΨTxs (qs )(−∆ẋ − γ∆x)
= −Ls ∆ds −Ψ
(22)
(23)
Remark 3. The terms Ψ Txm (qm )(∆ẋ + γ∆x) and
Ψ Txs (qs )(∆ẋ + γ∆x) in (18) and (19) are new and do
not exist in the nonlinear disturbance observer proposed
1767
by [10]. These new terms will help us to guarantee global
force tracking, global exponential position tracking and
disturbance tracking in our teleoperation system. ♦
The following two theorems state the main results of this
paper.
Theorem 1: Consider the teleoperation system subject to
disturbances described by (1) and (2). The master and the
slave disturbance observers are given in (18) and (19). Then
the disturbance observer based control laws given in (4)
and (5) with choices in (8), (9) and (10), guarantee global
asymptotic stability of the disturbance tracking error, the
position tracking error, and the force tracking error if the
following conditions hold:
1) Lm = LTm > 0 and Ls = LTs > 0 are constant symmetric
and positive definite matrices,
2) Kv given by (14) exists and is a constant symmetric
and positive definite matrix satisfying Kv > γI,
3) K p given by (15) exists and is a constant symmetric
and positive definite matrix,
4) ḋm ∼
= 0 and ḋs ∼
= 0, i.e., the rates of change of disturbances acting on the master and the slave robots
are negligible in comparison with the estimation error
dynamics (20) and (21).
Proof: Proof can be found in [12].
The previous theorem addresses the case when we have
slow-varying disturbances. In tha case of fast-varying disturbances, the tracking errors will be globally uniformly
ultimately bounded [12]. The next theorem states that the
disturbance tracking and the position tracking errors of the
teleoperation system can converge exponentially to zero
under certain conditions.
Theorem 2: Consider the teleoperation system subject to
disturbances described by (1) and (2). The master and the
slave disturbance observers are given in (18) and (19). Under
the control laws (4) and (5) with choices in (8), (9) and
(10), the disturbance tracking and position tracking errors
converge exponentially to zero
provided that the conditions
of Theorem 1 hold and γ ≤ λmax (K p + γKv ) where λmax (.)
represents the maximum eigenvalue of a matrix.
Proof: Proof can be found in [12].
Remark 4. As it is shown in Theorem 2, the minimum exponential rate of convergence for the disturbance
tracking and the position tracking errors to the origin is
Γ2 )
min (Γ
Γ
equal to λλmax
Γ1 ) . On the other hand, we have λmax (Γ 1 ) =
(Γ
κ1 = max{1, λmax (K p + γKv )} and λmin (Γ2 ) = κ2 =
min{λmin (Kv − γI), γλmin (K p ), λmin (Lm ), λmin (Ls )}. Thus,
one can simply determine the minimum rate of convergence
by computing κκ21 . ♦
IV. SIMULATION STUDY
In this section, computer simulations will illustrate the effectiveness of the proposed control scheme. Both the master
and the slave robots are considered to be planar two-link
manipulators with revolute joints. The Cartesian dynamics
are [14]
Mx (q) =
Vx (q, q̇) =
m2 + ms21
0
0
m2
2
Vx1 (q, q̇) Vx2 (q, q̇)
T
Gx (q) =
where
m1 g cs21 + m2 gs12
m2 gc12
T
Vx1
= −(m2 l1 c2 + m2 l2 )q̇21 − m2 l2 q̇22 −
c2
(2m2 l2 + m2 l1 c2 + m1 l1 2 )q̇1 q̇2
s2
Vx2
= m2 l1 s2 q̇21 + l1 m2 s2 q̇1 q̇2
Also, the forward kinematics and the Jacobian matrix are
h(q) =
x1
x2
T
=
J(q) =
l1 c1 + l2 c12
l1 s2
l1 c 2 + l 2
0
l2
l1 s1 + l2 s12
T
where l1 and l2 are the lengths of the links, and m1 and m2
are the point masses of the links. Also, we have s1 = sin(q1 ),
s2 = sin(q2 ), c1 = cos(q1 ) and c2 = cos(q2 ).
In this simulation study, the remote environment and the
human operator’s hand are modeled as linear impedances
with mass, damping and stiffness terms. We take Ze (s) =
(me s2 + be s + ke )I and Zh (s) = (mh s2 + bh s + kh )I, where I
is the identity matrix. In the simulations, the human hand
parameters are chosen as in [15]. The remote environment
is chosen to be a relatively stiff and damped medium,
modeled by dampers and springs. These lead us to mh =
11.6kg, bh = 17Nsm−1 , kh = 243Nm−1 and me = 0.0kg, be =
800Nsm−1 , ke = 1000Nm−1 .
The friction torques acting on the joints of the robots are
generated based on the model in [16]. For the i − th joint of
the robot, i = 1, 2, we have the frictions modeled as
τi f riction
−q̇2i
)]
v2si
−q̇2
+Fsi sgn(q̇i ) exp( 2 i ) + Fvi q̇i
vsi
= Fci sgn(q̇i )[1 − exp(
where Fci , Fsi , Fvi are the Coulomb, static, and viscous
friction coefficients, respectively. The parameter vsi is the
Stribeck parameter. In the simulations, the friction coefficients and the Stribeck parameter for the master and the slave
are chosen as follows [3]:
Fci
= 0.49, Fsi = 3.5, Fvi = 0.15, vsi = 0.189, i = 1, 2
In the simulations, we take the actual dynamic parameter
values of the master and the slave robots to be
m1m = 2.3kg, m2m = 2.3kg, l1m = 0.5m, l2m = 0.5m
m1s = 1.5kg, m2s = 1.5kg, l1s = 0.5m, l2s = 0.5m
Assuming a maximum of ±25% uncertainty in these parameters, we take the approximate values of these master and the
slave parameters to be randomly deviated from the actual
values:
1768
m̂1m = 2.82kg, m̂2m = 2.36kg, lˆ1m = 0.62m, lˆ2m = 0.56m
m̂1s = 1.16kg, m̂2s = 1.29kg, lˆ1s = 0.38m, lˆ2s = 0.52m
The controllers Cm , Cs , C1 , . . . , and C6 are chosen as in (8),
(9) and (10). The controllers and disturbance observer gains
and initial conditions are chosen as
γ =1
Kmv = 10I , Ksv = 10I
Kmp = 20I , Ksp = 20I
,
Cs f = I
Cm f = I
Lm = 20I , Ls = 20I
d̂0m = 0
,
d̂0s = 0
Kv = 5I
, K p = 10I
The above gains satisfy the conditions in Theorem 1 and the
Theorem 2. Note that the master and the slave controller and
observer gains have been chosen to be equal– this choice is
not necessary but is one that results in identical closed-loop
master and slave dynamics (11) and (12) when ∆dm = 0 and
∆ds = 0.
To start the simulations, there is no need to apply exogenous forces to the master and the slave, i.e., we take f∗h = 0
and f∗e = 0. Instead, we choose different values for initial
joint positions of the master and the slave while assuming
that both robots are initially at rest. We take the initial joint
position vectors to be
Fig. 4. Force tracking of the teleoperation system with conventional scheme
(no disturbance observer).
Figures 5 and 6 show the distrubance tracking of the
disturbance observer at the master and the slave sides,
repectively. As it can be seen in Figure 3, the disturbance
tracking error is small when there are no fast changes in the
disturbances. In fact, the estimated disturbances tend towards
the actual disturbances in the steady state.
q0m = [30◦ , 135◦ ]T , q0s = [0, 90◦ ]T
Figures 3 and 4 show the position and force tracking
response of the teleoperation system when no disturbance
observer is used. Because of the friction forces and the
dynamic uncertainties present in the master and the slave,
the control law without using disturbance observers fails to
achieve good position and force tracking. As can be seen in
Figure 3, from t ≈ 2sec we have ∆ẋ ≈ 0 and we have got
a constant offset in position tracking errors from t ≈ 2sec ,
when no disturbance observer is used.
Fig. 5.
Disturbance tracking at the master side.
Figures 7 and 8 show the position and force tracking
response of the teleoperation system when disturbance observer is used. As it can be seen, good force and position
tracking is achieved. Although we have disturbances with relatively fast dynamics our proposed control scheme achieves
full transparency. We expect the minimum exponential convergence rate, as mentioned in Remark 5, to be equal to
Fig. 3. Position tracking of the teleoperation system with conventional
scheme (no disturbance observer).
κ2
=
κ1
min{λmin (Kv − γI), γλmin (K p ), λmin (Lm ), λmin (Ls )}
max{1, λmax (K p + γKv )}
4
=
15
As it can be seen from the disturbance tracking and position
1769
Fig. 6.
Disturbance tracking at the slave side.
Fig. 8.
Force tracking of the teleoperation system with disturbance
observer-based controller.
tracking plots, the slowest tracking speed belongs to ∆x2 . In
this case we have
ln(
0.14
κ2
4
∆x2 (1sec )
) ≈ ln(
) = 0.33 ≥
=
∆x2 (0sec )
0.1
κ1
15
Fig. 7. Position tracking of the teleoperation system with disturbance
observer-based controller.
V. CONCLUSION
In this paper, a novel control scheme has been proposed
for nonlinear teleoperation systems subject to dynamic uncertainties and disturbances. The designed nonlinear disturbance
observer-based controller is incorporated into the 4-channel
bilateral teleoperation control framework. The proposed control scheme is able to suppress the disturbances present in the
teleoperation system. Transparency, in terms of asymptotic
convergence of the position and force tracking errors to zero
and exponential convergence of disturbance tracking and
position tracking errors to zero are achieved. The minimum
exponential convergence rate can be adjusted to a desired
value by proposed controller parameters. Simulations are
done to show the effectiveness of the proposed approach.
R EFERENCES
[1] K. Hashtrudi-Zaad and S. E. Salcudean, “Transparency in time
delay systems and the effect of local force feedback for transparent
teleoperation,” IEEE Trans. Robot. Autom., vol. 18, no. 1, pp. 108–
114, 2002.
[2] D. A. Lawrence, “Stability and transparency in bilateral teleoperation,”
IEEE Trans. Robot. Autom., vol. 9, pp. 624–637, Oct. 1993.
[3] N.V.Q. Hung, H.D. Tuan, T. Narikiyo, and P. Apkarian, “Adaptive
control for nonlinearly parameterized uncertainties in robot manipulators,” IEEE Trans. Cont. Syst. Technol., vol. 16, no. 3, pp. 458–468,
2008.
[4] K. Ohnishi, M. Shibata, and T. Murakami, “Motion control for
advanced mechatronics,” IEEE/ASME Trans. Mechatron., vol. 1, no.
1, pp. 56–67, Mar. 1996.
[5] K. S. Eom, I. H. Suh, and W. K. Chung, “Disturbance observer
based path tracking control of robot manipulator considering torque
saturation,” in Proc. Int. Conf. Advanced Robot., 1997, pp. 651–657.
[6] S. Sankaranarayanan and F. Khorrami, “Model independent friction
compensation,” in Proc. American Control Conference, 1998, pp. 463–
467.
[7] G. G. Zhang and J. Furusho, “Control of robot arms using joint torque
sensors,” in Proc. IEEE Int. Conf. Robot. Autom., 1997, pp. 3148–
3153.
[8] S. Kim, J. Y. Lee, and D. K. Sung, “A shifted gamma distribution
model for long-range dependent internet traffic,” IEEE Commun. Lett.,
vol. 7, no. 3, pp. 124–126, Mar. 2003.
[9] S. Komada, N. Machii, and T. Hori, “Control of redundant manipulators considering order of disturbance observer,” IEEE Trans. Ind.
Electron., vol. 47, no. 2, pp. 413–420, Apr. 2000.
[10] W. H. Chen, D. J. Ballance, P. J. Gawthrop, and J. O’Reilly, “A
nonlinear disturbance observer for robotic manipulators,” IEEE Trans.
Ind. Electron., vol. 47, no. 4, pp. 932–938, Aug. 2000.
[11] Y. Matsumoto, S. Katsura, and K. Ohnishi, “An analysis and design
of bilateral control based on disturbance observer,” in Proc. IEEE Int.
Conf. Ind. Technol., 2003, pp. 802–807.
[12] A. Mohammadi, M. Tavakoli, and H. J. Marquez, “Disturbance
observer based control of nonlinear haptic teleoperation systems,” IET
Control Theory Appl., 2011, In Press.
[13] M. Tavakoli, A. Aziminejad, R.V. Patel, and M. Moallem, “Highfidelity bilateral teleoperation systems and the effect of multimodal
haptics,” IEEE Trans. Syst., Man Cybern. B, bern., vol. 37, no. 6, pp.
1512–1528, Dec. 2007.
[14] J. J. Craig, Introduction to Robotics: Mechanics and Control, Upper
Saddle River, NJ: Pearson Prentice Hall, 2005.
[15] K. Kosuge, Y. Fujisawa, and T. Fukuda, “Control of mechanical system
with man-machine interaction,” in Proc. IEEE/RSJ Int. Conf. Intell.
Robots Syst., 1992, pp. 87–92.
[16] B. Armstrong-Hélouvry, P. Dupont, and C. Canudas de Wit, “A survey
of models, analysis tools and compensation methods for the control of
machines with friction,” Automatica, vol. 30, no. 7, pp. 1083–1138,
1994.
1770