Professional Documents
Culture Documents
net/publication/303847597
CITATIONS READS
18 1,241
3 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Mojtaba Sharifi on 18 September 2017.
Saeed Behzadipour 1
Department of Mechanical Engineering,
Djawad Movaffaghian Research Center in Neuro-Rehab Technologies,
Sharif University of Technology
Azadi St., P.O. Box: 11155-9567, Tehran, Iran
behzadipour@sharif.edu
Hassan Salarieh
Department of Mechanical Engineering, Sharif University of Technology
Azadi St., P.O. Box: 11155-9567, Tehran, Iran
salarieh@sharif.edu
ABSTRACT
A bilateral nonlinear adaptive impedance controller is proposed for the control of multi-DOF teleoperation
systems. In this controller, instead of conventional position and/or force tracking, the impedance of the
nonlinear teleoperation system is controlled. The controller provides asymptotic tracking of two
impedance models in Cartesian coordinates for the master and slave robots. The proposed bilateral
controller can be used in different medical applications such as tele-surgery and tele-rehabilitation where
the impedance of the robot in interaction with human subject is of great importance. The parameters of
the two impedance models can be adjusted according to the application and corresponding objectives and
requirements. The dynamic uncertainties are considered in the model of the master and slave robots. The
stability and the tracking performance of the system are proved via a Lyapunov analysis. Moreover, the
1
Corresponding author.
1
Journal of Dynamic Systems, Measurement, and Control
adaptation laws are proposed in the joint space for reducing the computational complexity however the
controller and the stability proof are all presented in Cartesian coordinates. Using simulations on a two
DOFs robots, the effectiveness of the proposed controller is investigated in telesurgery and
telerehabilitation operations.
1. INTRODUCTION
Teleoperation systems enable human to interact with environments that are too
inaccessible, too confined, or too hazardous. They have been widely used in different
areas such as minimally invasive tele-surgery [1], tele-rehabilitation [2, 3] and tele-
sonography [4] systems, nuclear waste site and radioactive material management [5],
outer space and undersea exploration [6, 7], and construction/forestry machines of the
haptic force feedback of the interaction point between the slave robot and the remote
environment is provided to the human operator using the master robot. The control
strategy of these systems is called bilateral as the information flows in two directions
between the operator and the remote environment [9]. In the bilateral teleoperation
robotic systems, the human operator applies force ( f h ) on the master robot to control
the position of the slave robot ( x s ) in order to perform a task in the remote
environment. If the slave robot perfectly tracks the master robot’s position trajectory
2
Journal of Dynamic Systems, Measurement, and Control
( xm ) and the master robot ideally realizes the slave-environment contact force ( f e ) to
Various control strategies have been proposed for linear teleoperation systems
(with one-DOF) [10-17]. Among them, the 4- channel architecture is the most successful
in terms of fulfilling transparency [12, 13]. However, this control scheme assumes
perfect knowledge of the linear dynamics of the master and the slave robots. In other
words, these controllers need the exact model of the robots dynamics that may be
transparency are hard to guarantee in practice. A few adaptive controllers [18, 19] have
been presented in the literature for linear master and slave models where the slave and
systems are required instead of one-DOF linear ones. Accordingly, designing a nonlinear
for researchers in recent years. However, the control for master and slave robots with
nonlinear models is more challenging in comparison with those with linear models. In
this context, some adaptive controllers [20, 21] were proposed for nonlinear master and
sensing their interaction forces applied to the system. On the other hand, there are
some works on the synchronization of positions and velocities of the master and slave
robots using PD [22] and adaptive [23] controllers considering a time delay in the
3
Journal of Dynamic Systems, Measurement, and Control
synchronization that is achieved in the absence of external force (the case of free
motion), and the simultaneous position and force tracking performances cannot be
provided. To achieve the position and force tracking performances for multi-DOF
systems, Ryu and Kwon [24] and Hung et al. [25] designed two nonlinear adaptive
controllers. As a modification on one of these controllers [24], Liu and Tavakoli [26, 27]
developed two nonlinear adaptive 4-channel controllers. In the last mentioned adaptive
controllers [26, 27] with position and force tracking objectives, the acceleration of each
robot in addition to the velocity and position signals should be measured and used in
interactions such as robot assisted rehabilitation and haptic systems. In this context, a
between the motion and the external forces or moments acting on the system, is
realized. Many interactive tasks that could not be handled well by pure position or force
[31, 32], many contributions were presented in this area, including robust impedance
control [33] and hybrid impedance control [34]. In addition, the adaptive impedance
4
Journal of Dynamic Systems, Measurement, and Control
admittance control in [37-39] have the advantage of decreasing the need for precise
mathematical modeling of the robot and environment using adaptive control scheme. It
should be noted that the above mentioned works [28-39] on the impedance and/or
admittance control were performed on a single robot manipulator and not for a
There are some works on employing the impedance control context for the
bilateral teleoperation systems. Rubio et al. [40] and Dubey et al. [41] proposed two
control algorithms with varying impedance parameters in order to increase stability and
enhance tracking performance. Love and Book [42] realized a virtual master workspace
with an impedance model, and reduced the operator’s fatigue during a task by
regulating the damping ratio of the impedance model. Moreover, Cho and Park [43]
proposed a bilateral impedance controller for linear (one-DOF) master and slave robots
using a robust control scheme and Garcia-Valdovinos et al. [44] designed an observer for
this controller [43]. In another work, Abbott and Okamura [45] developed impedance
for both master and slave robots that has only a damping element (without mass and
5
Journal of Dynamic Systems, Measurement, and Control
DOF teleoperation systems by defining two distinct desired impedance models for the
5. The signals of robots’ acceleration are not used in the controller due to
defining the reference impedance models and controller design. However, in the recent
works [26, 27, 46] on bilateral adaptive control, these signals are required because
direct position and force tracking of nonlinear system are pursued (conventional
coordinates [26, 27, 46], the adaptation laws are proposed in the joint space for
reducing the computational complexity however the controller and impedance models
It should be noted that the characteristics and applications of this new nonlinear
6
Journal of Dynamic Systems, Measurement, and Control
systems. In other words, this work is an extension of the adaptive impedance control
methods (such as [28-30, 35, 37, 38]) to bilateral control by realizing two virtual
master and slave robots are introduced and discussed in section 3.1. Control laws of the
master and slave robots are designed and presented in section 3.2. In section 3.3, the
closed-loop dynamics of the teleoperation system using the proposed bilateral adaptive
impedance controller is obtained. Moreover, the adaptation laws of the master and
slave robots are described in section 3.4. The stability and convergence of the
teleoperation system are proved in section 3.5 using the Lyapunov theorem. The
simulations of the teleoperation system and controller in telesurgery (Sec. 4.1) and
show the effectiveness of the proposed bilateral control method. Finally, the concluding
The n-DOF master and slave robot manipulators are modeled in the joint space
as follows [47-49]:
7
Journal of Dynamic Systems, Measurement, and Control
n1 nn
where q m and qs are the joint positions, M q,m (qm ) and M q,s (qs ) are the
nn
inertia or mass matrices, Cq,m (qm , qm ) and Cq,s (qs , qs ) includes the Coriolis and
n1
centrifugal terms, Gq,m (qm ) and Gq,s (qs ) are the gravity terms, Fq,m (qm ) and
n1
Fq,s (qs ) are the viscous and coulomb friction terms, and τ m and τ s n1
are the
vectors of the control torques (originated from the actuators) of the master and the
n1
slave robots, respectively. Also, τh is the external torque that the operator
n1
applies to the master robot and τe is the external torque that the environment
61
where xm and xs are the Cartesian coordinates of the master and slave robots’
61
end-effectors, respectively. f h and fe are the external interaction forces that the
operator applies to the master, and the environment applies to the slave, respectively.
Considering the subscript i m for the master robot and i s for the slave
robot, the kinematic transformations between the joint and Cartesian spaces are
expressed as:
8
Journal of Dynamic Systems, Measurement, and Control
matrices of dynamic models in the joint space (Eqs. (1) and (2)) and in the Cartesian
space ones (Eqs. (3) and (4)) for non-redundant and non-singular manipulators are:
Cx,i (qi , qi ) J i T Cq,i (qi , qi ) M q,i (qi ) J i 1 J i J i 1
(6)
Gx,i (qi ) J i T Gq,i (qi ) , Fx,i (qi ) J i T Fq,i (qi )
f i J i T τ i , f h J m T τ h , f e J s T τ e
respectively. The matrices expressed in dynamic model equations (1) to (4) have the
Property 1. Inertia matrices M q,i (qi ) and M x,i (qi ) are symmetric and positive definite.
Property 2. Matrices M q,i (qi ) 2Cq,i (qi , qi ) and M x,i (qi ) 2Cx,i (qi , qi ) are skew symmetric.
Property 3. The left side of Eqs. (1) to (4) can be linearly parameterized in terms of the
M q,i (qi ) φ1,i Cq,i (qi , qi ) φ2,i Gq,i (qi ) Fq,i (qi ) Yq,i (φ1,i , φ2,i , qi , qi ) αq,i (7)
M x,i (qi ) ψ1,i Cx,i (qi , qi ) ψ 2,i Gx,i (qi ) Fx,i (qi ) Yx,i (ψ1,i , ψ 2,i , qi , qi ) αx,i (8)
where α q ,i and α x ,i are the vectors of unknown parameters of the robot when
parameterization is done in the joint space and Cartesian space, respectively. The
regressor matrices Yq ,i and Yx,i contain known functions and φ1,i , φ 2,i , ψ 1,i and ψ 2,i
9
Journal of Dynamic Systems, Measurement, and Control
signals in communication channels are shown in Fig. 2. As shown in this figure, the state
variables (position and velocity) and the interaction force signals of the master robot (in
the local site) are transmitted to the slave (remote site) by two communication channels
and the interaction force of the slave robot is transmitted to the master by another
The desired impedance model of the master robot’s end-effector expresses the
desired dynamical relationship between the human operator and master robot that is
where xmodm is the response position vector of the master impedance model, and x 0
is the position that corresponds to the natural length of the virtual stiffness element
( k m ). In other words, x 0 is the desired position that the master robot should hold in
interaction force that the operator applies to the master robot and fe is the interaction
force that the environment applies to the slave robot. The force scaling factor kf in Eq.
(9) used for the transmitted interaction force f e from the slave to the master (in the
10
Journal of Dynamic Systems, Measurement, and Control
local site). mm , cm and k m are the desired virtual mass, damping, and stiffness
form.
According to Eq. (9), when the master robot reaches its objective of tracking the
master impedance model response ( xmodm ), the operator senses the environment
other words, the human operator’s perception is that he/she is interacting with the
Fig. 3.
It should be mentioned that the master impedance model (9) is constructed and
solved in the local site to obtain the desired trajectory ( xmodm ) of the master robot.
the environment and the slave should be transferred from the remote site to the local
site using a communication channel to be used in the master impedance model (9).
The desired impedance model of the slave robot’s end-effector also represents
the desired dynamics between the environment and slave robot during the tracking of
11
Journal of Dynamic Systems, Measurement, and Control
where ms , cs and k s are the desired virtual mass, damping, and stiffness coefficients of
the slave impedance model in physical interaction with environment. Also, x mod s is the
error vector of slave impedance response with respect to the master position that is
defined as:
where kp is the scaling factor corresponds to the transmitted position and velocity
vectors of the master to the slave. It should be mentioned that the position ( x m ),
velocity ( x m ) and acceleration ( x m ) of master robot are required to obtain the desired
slave impedance model response ( x mod s , x mod s , x mod s ), which will be used in the
1 c k
xm est fh kf fe m xm m (xm x0 ) (12)
mm mm mm
In Eq. (12), the master robot acceleration is estimated in terms of the interaction forces,
master position and velocity that are available signals according to the master
impedance model (9). Also, when the master robot trajectory ( x m ) asymptotically
converges to the master impedance model response ( x modm ), Eq. (12) has a higher
12
Journal of Dynamic Systems, Measurement, and Control
accuracy can be improved by specifying the same initial position and velocity for the
It is known that the slave impedance model (10) is constructed and solved in the
remote site to obtain the desired trajectory ( xmod s ) of slave robot. According to Eq.
(10), the master robot’s position, velocity and acceleration ( x m , x m and x m ) are
required in the slave impedance model. However, using the estimation of master robot
acceleration (12), the measurement of the interaction force ( f h ) between the operator
and master together with the master robot’s position and velocity ( x m and x m ) should
be transferred from the local site to the remote site using two communication channels
k s ), the tracking error of slave ( xmod ) with respect to master is decreased (the slave
s
becomes more accurate in tracking of the master trajectory). On the other hand, by
decreasing these impedance parameters in (10), the slave becomes more flexible in its
interaction with environment and consequently has higher tracking error ( xmods ).
Accordingly, two impedance models are employed in this methodology. The first
one is the master reference model (9) that the operator perceives, and the second one
is the slave reference model (10) that is the flexibility of the slave robot in its physical
tracking and/or exact force tracking by adjustment of the parameters in two impedance
13
Journal of Dynamic Systems, Measurement, and Control
models (9) and (10). In master impedance model (9), by choosing small values for the
impedance parameters ( mm , cm and k m ), the left side of Eq.(9) is small due to the
boundedness of position signals xmodm and x0 . Accordingly, the right side of Eq. (9)
is not obtained and the operator (with f h ) senses the impedance elements ( mm , cm
Also, due to the boundedness of the right side of Eq. (10) that is fe , the left
side of this equation is also bounded. Consequently, by utilizing large values for the
slave impedance parameters ( ms , cs and k s ) in the left side of Eq. (10), the desired
slave position error with respect to the scaled master trajectory ( xmods xmods kp xm )
approximately obtained using large impedance parameters for the slave. On the other
hand, if the parameters of the slave impedance are decreased, the flexibility of the slave
with respect to the master trajectory is increased and the tracking error ( x mod s )
becomes large.
14
Journal of Dynamic Systems, Measurement, and Control
impedance parameters should be decreased (for force reflecting) and the slave
more important in comparison with the slave flexibility (compliance) in the environment
the slave robot should have higher compliance in its interaction with the patient
telerehabilitation systems that the slave robot moves the patient limbs with a limited
flexibility, the patient is permitted to deviate from the initially designed (master)
trajectory using the slave impedance model (10). This performance is approximately
required in telesonography systems where the slave robot should show a flexible
Fig. 5 and their details are given in this section. As mentioned earlier, there are two
impedance models (9) and (10) that their responses are tracked by adaptive controllers
of master and slave robots. The dynamic models of master and slave robots are
considered to have fully parametric uncertainties. Also, modeling the human operator
and the remote environment is not required because their interaction forces ( f h , f e )
15
Journal of Dynamic Systems, Measurement, and Control
Now, the adaptive control laws designed for the master and slave robots are
presented. For this purpose, the sliding surfaces used in the bilateral controller are
where xm x m x modm and x s x s x mods are the vectors of the master and slave position
error with respect to the responses of their impedance models (9) and (10). 1,m and
1,s are positive constant parameters that guarantee the stability of the sliding surfaces
(13). Now, we introduce the master and slave reference velocities as additional
variables:
Also, the master and slave reference accelerations are obtained from Eq. (14) as:
where xmodm is obtained from master impedance model (9). Also, due to Eq. (11),
xmods xmods kp xm where xmods is obtained from the slave impedance model (10).
However, the exact master acceleration xm is not available and it is estimated by xm est
in Eq. (12). This acceleration estimation may has an error as xm xm est xm which is
16
Journal of Dynamic Systems, Measurement, and Control
xmods xmod kp xm est should be used instead of its real value xmod in Eq. (16) to
est s s
xr,s est xmods 1,s xs xmods kp xm est 1,s xs
est
(17)
Now, the nonlinear adaptive control laws of the master and slave robots for
tracking their impedance models are defined in Cartesian coordinates, respectively, as:
f s Mˆ x,s (q s ) xr ,s est 2,s S s Cˆ x ,s (q s , q s ) xr ,s
(20)
Gˆ x, s (q s ) Fˆx,s (q s ) fe s sgn(S s )
Accent denotes the estimated values of matrices, vectors and scalars. 2,m and 2,s
are two positive constants that guarantee the stability and convergence of the
teleoperation system, which will be determined using the Lyapunov theorem. s in the
control law of the slave robot (20) is a positive constant parameter, and it will be shown
that s sgn(Ss ) provides the robustness of the bilateral controller against the bounded
17
Journal of Dynamic Systems, Measurement, and Control
The actual control input of robots applied in the joint space by motor torques is
τm J T
M 2, m
ˆ (q ) xr ,m Sm Cˆ x,m (qm , qm ) xr ,m
x,m m (21)
m
Gˆ (q ) Fˆ (q ) f
x,m m x ,m m h
τs Js T
Mˆ (q ) x
x, s s ˆ
r , s est 2, s S s Cx, s (q s , q s ) xr , s
(22)
ˆ ˆ
Gx,s (q s ) Fx,s (q s ) fe s sgn(S s )
With the purpose of employing parameterization in the joint space, the master and
slave input torques ( τ m and τ s ) are expressed in terms of joint space matrices and
τ m Mˆ q,m J m 1 xr ,m 2,m S m
(23)
Cˆq,m Mˆ q,m J m 1 J m J m 1 xr ,m Gˆ q,m Fˆq,m J mT f h
τ s Mˆ q , s J s 1 xr ,s est 2,s S s
(24)
Cˆq, s Mˆ q, s J s 1 J s J s 1 xr ,s Gˆ q,s Fˆq,s J sT fe J sT s sgn(S s )
The above relations for the motor input torques can be written as:
18
Journal of Dynamic Systems, Measurement, and Control
where v1,m , v2,m , v1,s and v2,s are known vectors and do not contain any estimated
parameters of the system’ dynamics. These vectors are obtained for the master and
v1,m J m1 xr,m 2,m Sm J m1J m J m1xr,m , v2,m J m1 xr,m (27)
v1,s J s 1 xr,s est 2,s Ss J s 1J s J s 1xr,s , v2,s J s 1 xr,s (28)
joint space (Eqs. (25) and (26)) are linearly parameterized as:
where Yq,m αˆ q,m and Yq, s αˆ q, s are linear parameterizations of the first four terms of Eq.
(25) and (26), respectively. Also, α q ,m and α q ,s are the vectors of actual parameters of
the master and slave robots, respectively, and αˆ q ,m and αˆ q , s are their estimations. Yq,m
and Yq,s are the joint space regressor matrices for the master and slave robots,
respectively. As a result, the linear parameterization for each robot is expressed as:
where v1,i and v2,i were obtained in Eqs. (27) and (28) (using i m and i s the above
19
Journal of Dynamic Systems, Measurement, and Control
In this section, the closed-loop dynamics of the master and slave robots using
the proposed bilateral adaptive impedance controller is obtained. For this purpose, the
control laws (19) and (20) are replaced in the teleoperation system dynamics (3) and (4)
that yields:
M x ,m x m xr ,m 2,m M x ,m S m
Mˆ x ,m M x ,m x r ,m 2,m S m
Cˆ
(32)
x ,m Cx , m x r , m Cx , m S m
Gˆx ,m Gx ,m Fˆ x ,m Fx ,m
M x , s x s xr , s est 2, s M x , s S s
Mˆ x, s M x, s x r ,s 2, s S s
Cˆ
(33)
x,s Cx , s x r , s C x , s S s
According to Property 3 and using Eqs. (27), (28) and (31), the above equations can be
rewritten as:
M x,s x s xr ,s est 2,s M x,s S s Cx,sS s J s T Yq,s αq,s s sgn(Ss ) (35)
20
Journal of Dynamic Systems, Measurement, and Control
where αq,m αˆ q,m αq,m and αq,s αˆ q,s αq,s are the error vectors of master and slave
parameter estimations, respectively. Using Eq. (18) for the estimated value of the slave
M x,s xs xr ,s kp M x,s xm 2,s M x,s Ss Cx,sSs J s T Yq,s αq,s s sgn(Ss ) (36)
Substituting the time derivative of Eq. (15) in Eqs. (34) and (36) yields the final error
M x,s Ss 2,s M x,s Ss Cx,s Ss J s T Yq,s αq,s kp M x,s xm s sgn(Ss ) (38)
Now, the estimation error for each of the master and slave robots is defined. The
master or slave robot dynamics equation in the joint space is expressed as follows:
M q,i (qi ) qi Cq,i (qi , qi ) qi Gq,i (qi ) Fq,i (qi ) τq,i (39)
21
Journal of Dynamic Systems, Measurement, and Control
where in this paper, for the master: i m and τq,m τ m τ h , and also for the slave:
i s and τq,s τ s τ e . Then, the above dynamics passes through a first-order filter
W ( s) (40)
s
where w(t ) e t is its impulse response in time domain. Both sides of Eq. (39) are
The first term in the left side of (41) is obtained using the partial integration, as:
t
w(t r ) M
0 q ,i qi dr
t d
w(t r ) M q ,i qi 0 w(t r ) M q ,i qi dr
t
0 dr (42)
t
w(0) M q ,i qi w(t ) M q ,i (qi (0))qi (0) w(t r ) M q ,i qi w(t r ) M q ,i qi dr
0
where Wq,i is the filtered version of Yq ,i in Eq. (7) with φ1,i qi and φ2,i qi .
Moreover, due to the filtering and partial integration (42), the elements of matrix Wq,i
are only functions of the available measurements of q i and q i , and the acceleration
22
Journal of Dynamic Systems, Measurement, and Control
( q i ) is not required. Also, Eq. (43) is the linear parameterization of the filtered dynamics
t
w(t r ) τ
0 q ,i (r ) dr y i (t ) (44)
Equation (43) can be written for αˆ q,i (the estimated value of α q,i ) as:
ei yˆ i (t ) yi (t ) (46)
Now, the bounded gain forgetting (BGF) composite adaptation laws of master
where Ri nn
(for master i m or slave i s ) is a positive definite weighting matrix
that indicates the cost of estimation error ei (having the parameters information) in the
23
Journal of Dynamic Systems, Measurement, and Control
Ri ai I nn (50)
where I nn is the unity matrix with the same dimensions as Ri , and ai is a positive
constant. Also, Pi (t ) in (48) and (49) is the adaptation gain that is a uniformly positive
d 1
Pi (t ) i (t ) Pi 1 Wq ,iTWq,i (51)
dt
where 0,i and k0,i are two positive constants specifying the maximum forgetting rate
and the upper bound of the gain matrix norm ( Pi ), respectively. The gain update rule
i (t ) 1,i [49].
It should be mentioned that Eqs. (48) and (49) express the adaptation laws in
joint space to reduce the complexity of calculations; however the closed-loop dynamics
In the following Lyapunov stability proof, it will be shown that the tracking error
estimation.
24
Journal of Dynamic Systems, Measurement, and Control
V
2
1 T
Sm M x,m S m STs M x,s S s αqT,m Pm (t )1 αq,m αqT, s Ps (t )1 αq, s (53)
The Lyapunov function terms are proposed in two different coordinates, the first two
terms (tracking errors) are in Cartesian coordinates while the last two terms
because the adaptation laws are given in joint space to reduce the complexity of their
calculations in comparison with previous works [26, 27, 46] that present the adaptation
law in Cartesian space. Since the inertia matrix M x ,i and the adaptation gain matrix Pi
are uniformly positive definite, the Lyapunov function is positive definite ( V 0 ). The
1 1
V STm M x ,m S m M x ,m S m STs M x ,s S s M x ,s S s
2 2
1
αˆ q ,mT Pm (t ) 1 α q ,m α q ,mT
2
d
dt
Pm (t ) 1 α q ,m (54)
1
αˆ q , sT Ps (t ) 1 α q , s α q , sT
2
d
dt
Ps (t ) 1 α q , s
25
Journal of Dynamic Systems, Measurement, and Control
where αq,i αˆ q,i , because αq,i αˆ q,i αq,i and α q ,i is the constant vector of actual
parameters; thus αq,i 0 . Substituting the final error dynamics of master and slave
S s kp M x , s x m s sgn(S s )
T
According to Property 2 of robot dynamics that M x,i 2Cx,i is skew symmetric, Eq. (55)
simplifies to:
S mT J m T Yq ,m α q ,m αˆ q ,mT Pm (t ) 1 α q ,m α q ,mT Pm (t ) 1 α q ,m
1 d
2 dt
(56)
S sT J s T Yq , s α q , s αˆ q , sT Ps (t ) 1 α q , s α q , sT Ps (t ) 1 α q ,s
1 d
2 dt
S sT kp M x , s x m s sgn(S s )
Then, using the bounded gain forgetting (BGF) composite adaptation laws (48) and (49),
we have:
26
Journal of Dynamic Systems, Measurement, and Control
Also, using the exponentially forgetting least-squares update rule (51) for the adaptation
After that, substituting the relation of the estimation error (47) and Eq. (50), we have:
acceleration xm , the positive constant parameter s should satisfy the following
component-wise inequality:
27
Journal of Dynamic Systems, Measurement, and Control
T
vector, i.e.: if r =[r1 , r2 , ... , ri ] is a vector with i elements: r = max (r1 , r2 , ... , ri ) . It
should be noted that the actual parameters in M x,s are uncertain in this adaptive
however to have the stability property, the maximum estimated value of M x,s xm
should be bounded and its bound, as mentioned in Eq. (54), should be less than s . In
other words, s should be chosen as large as needed (using a trial and error method)
such that the stability of the system is provided; which means inequality (60) is satisfied.
Then, the time derivative of Lyapunov function (59) can be written as:
By choosing ai 1 2 and noting that M x ,i and Pi are positive definite and according to
definite.
Using the Lyapunov theorem and Barbalat’s lemma [49], and according to the
system and the convergence to Sm 0 and Ss 0 are proved. Since the dynamics of
28
Journal of Dynamic Systems, Measurement, and Control
master and slave sliding surfaces S m and S s are stable (see Eq. (13)), the convergence
proved as t .
Also, as mentioned before for the update rules (51) and (52), Pi (t ) k0,i I and the
Under this condition, the convergence of i (t ) αq,iT Pi (t )1 αq,i 0 leads to αq,i 0 ,
estimation ( αq,i 0 ). A positive constant 0,i is defined as 0,i min 22,i , 1,i for each
of master ( i m ) and slave ( i s ). Also, by defining 0 min 0,m , 1,s and using Eqs.
(53) and (61), it can be written for the Lyapunov function that:
V (t ) 0 V (t ) 0 (63)
Thus, it is concluded that the value of Lyapunov function converges to zero as:
estimation error αq,i to zero is proved due to the definition of Lyapunov function (48).
Before illustrating the simulation results, it should be mentioned that the “sgn”
function in the slave control law (20) or (24) may lead to undesired discontinuities and
29
Journal of Dynamic Systems, Measurement, and Control
4. SIMULATIONS
robot with one Revolute and one Prismatic joint, as shown in Fig. 6. Each of the master
and slave robots is equipped with a force sensor to measure the externally applied force
and moment. The robots along with the proposed bilateral controller are modeled in
Simulink-Matlab software.
qm 1 l2 m and q s 1 l2 s for master and slave, respectively, where 1 is the
T T
revolute joint position and l z is the prismatic joint position. The Cartesian position of
the end-effector for each robot is defined as xi x z i . The Jacobian matrix of the
T
l l sin(1 ) cos(1 )
Ji 1 2
l1 l2 cos(1 )
(64)
sin(1 ) i
Employing the Lagrange method, the dynamic equations of each robot are
obtained as:
30
Journal of Dynamic Systems, Measurement, and Control
I I m l
1 2
2
1 1g
m2 l1 l2 l2 g 1 2m2 l1 l2 l2 g l21
2
(65)
m1l1g m2 l1 l2 l2 g g cos(1 ) c11 1 sgn(1 ) 1
where the subscript 1 denotes link 1 and joint 1 (revolute), and subscript 2 is used for
are the inertia and dimensional properties of link 1 and 2 of the robots. Moreover, c1
and c2 are the viscous friction coefficients of joints 1 and 2, respectively. Similarly, 1
and 2 are the coulomb friction coefficients of joints 1 and 2, respectively. 1 and 2
are the applied torques on the revolute ( 1 ) and prismatic ( l2 ) joints, respectively. By
arranging Eqs. (65) and (66) in the standard form of dynamics presented in the joint
space, the matrices and vectors in Eqs. (1) and (2) are obtained as:
I I m l 2 m l l l 2 0
1 2 1 1g 2 1 2 2g
M q ,i (qi )
0 m2 i
m2 l1 l2 l2 g l2 m2 l1 l2 l2 g 1
Cq ,i (qi , qi ) (67)
m2 l1 l2 l2 g 1 0
i
m1l1g m2 l1 l2 l2 g g cos(1 )
Gq ,i (qi )
m2 g sin(1 ) i
c11 1 sgn(1 )
1
Fq ,i (qi ) , i int
2 i
i
c2l2 2 sgn(l2 )
i
31
Journal of Dynamic Systems, Measurement, and Control
where for the master int m h and for the slave int s e . Moreover, the inertia matrix
Now, the linear parameterization of the robot dynamics in the joint space that
αq M11 m2 m2 l1 l2 g m1 gl1g m2 g l1 l2 g
T
m2 g c1 c2 1 2 (69)
Considering the known vectors v1 v11 v12 and v2 v21 v22 in Eq. (31), the
T T
Y1,2 l2 2 v11 l2 l2 v21 1 v22 , Y2,2 v12 l21 v21 ,
Y1,3 2 l2 v11 l2 v21 1 v22 , Y2,3 1 v21 ,
Y1,4 cos(1 ) , Y2,4 0,
Y1,5 l2 cos(1 ) , Y2,5 sin(1 ) ,
Y1,6 cos(1 ) , Y2,6 0, (70)
Y1,7 1 , Y2,7 0,
Y1,8 0, Y2,8 l2 ,
Y1,9 sgn(1 ) , Y2,9 0,
Y1,10 0, Y2,10 sgn(l2 ) ,
32
Journal of Dynamic Systems, Measurement, and Control
The parameters of robots’ kinematics and dynamics (67), control laws ((19) and
(20)) and adaptation laws ((48) and (49)) that are used in simulations are given in Table
1. The parameters of the control laws are adjusted such that each controller has its
minimum tracking errors ( x m , x s ). Similarly, the parameters of the adaptation laws are
chosen using trial and error such that the minimum RMS of estimation errors
( em Wq ,mαq ,m , e s Wq , s αq , s ) is obtained.
simulate the teleoperation system, the interaction forces should be produced according
to the dynamics models of the human operator and the environment. These models are
presented by the following general forms [20, 21, 26, 27, 46, 51, 52]:
fe M e xs Ce xs Ke (xs x0 ) (72)
where f h is the active force of human operator’s hand generated by the muscles.
matrices corresponding to the mass (inertia), damping and stiffness of the operator and
environment, respectively. These dynamic parameters were presented in [26, 27, 46] for
a human operator (surgeon) and soft tissue environment that are listed in Table 2.
33
Journal of Dynamic Systems, Measurement, and Control
[28, 29]):
impedance model (9) are chosen to be small (to obtain force tracking performance) and
the parameters of the slave impedance model (10) are chosen to be large (to obtain
position tracking performance). Also, the position and force scaling factors are
considered to be kp 1 and kf 1 in order to provide the exact tissue force for the
surgeon and to have equal trajectories during the operation. The values of these
impedance parameters and scaling factors are given in Table 3 for telesurgery
application. Moreover, using these parameters, the master impedance model (9) has
two negative real poles ( 2 , 2 ), and the slave impedance model (10) has two negative
real poles ( 20 , 20 ). In other words, according to the standard expression of linear
second order systems, the master impedance model has a natural frequency of
2 rad/sec and a damping ratio of 1 . Also, the slave impedance model has a natural
Employing the proposed bilateral impedance controller, the following results are
obtained. The interaction force between the surgeon (operator) and the master robot
( f h ), together with the interaction force between the slave robot and the soft tissue
34
Journal of Dynamic Systems, Measurement, and Control
shown in Fig. 8. As it is seen in Fig. 8, the maximum value of force error is 4.1 N (less
than 10% of the maximum interaction force, i.e. 42.5 N in Fig. 7) and it occurred at the
initial instant. Also, the mean value of force error is 0.22 N (0.5% of the maximum
interaction force, i.e. 42.5 N in Fig. 7). It should be mentioned that a small force error is
seen throughout the motion (Fig. 8). This remaining small force error ( fh kf fe ) is
( mm xmodm cm xmodm km (xmodm x0 ) ) that determines the difference between the scaled
Figure 9 shows the position responses of master and slave impedance models
together with the master and slave robots’ trajectories in x direction. The initial
position of the master robot is considered to be xm0 xm0 , zm0 0.62, 0.62 m that has
T T
an initial error with respect to the initial value of the master impedance model response
xmodm0 xmodm0 , zmodm0 0.6, 0.6 m . Also, the initial position of the slave robot is
T T
considered to be xs 0 xs 0 , zs 0 0.64, 0.64 m that has an initial error with respect to
T T
35
Journal of Dynamic Systems, Measurement, and Control
teleoperation system, which is the result of choosing large values for the slave
and accelerations of the master and slave end-effectors and their impedance models are
shown in Fig. 9. The absolute position tracking error between master and slave robots
errors of the master and slave robots with respect to their corresponding impedance
In Fig. 10, the initial (maximum) value of the absolute tracking error between the
master and slave robots is 2.83 102 m , i.e. 58% of the motion range of master robot in
x direction ( 4.85 102 m due to Fig. 9). However, the mean value of the tracking error
between the master and slave is 7.60 104 m , i.e. 1.5% of the range of master motion in
x direction that is 4.85 102 m (Fig. 9). This tracking error has two sources: 1. the
desired slave error with respect to master xmods (xmods kp xm ) that is tried to be small
by choosing large parameters for the slave impedance model (10), and 2. the slave
tracking error ( x s x s xmods ) that converges to zero and originated from the
performance of slave controller, which is shown in Fig. 10. Also, the master tracking
master controller in asymptotic convergence of the tracking error. Moreover, the mean
values of the master and the slave absolute tracking errors with respect to their
36
Journal of Dynamic Systems, Measurement, and Control
corresponding impedance models (9) and (10) are 7.24 104 m and 7.37 104 m ,
bilateral adaptive controller of master and slave in asymptotic tracking, Fig. 11 shows
impedance controller can achieve their objectives. In other words, the force tracking
small values for the master impedance parameters ( mm , cm and k m ) in Eq. (9). Moreover,
choosing large values for the slave impedance parameters ( ms , cs and k s ) in Eq. (10).
These two objectives (position and force tracking performances) are obtained by
defining a desired impedance objective for each of master and slave robots (one
objective for each robot). However, in previous bilateral nonlinear adaptive controllers
(such as [26, 27]), the position and force tracking performances are followed
In addition, the actual and estimation values of the master acceleration ( x m and
x m est ) together with their difference (estimation error xm xm est xm ) in x direction
37
Journal of Dynamic Systems, Measurement, and Control
of Cartesian coordinates are shown in Fig. 12. The acceleration estimation error is
similar in z direction and it is not presented here for the sake of brevity. The initial
acceleration of the master robot is specified such that to be approximately the same as
its corresponding impedance model (9), by applying the zero forces at the initial
moment ( t 0 ) of motion. As a result, the master robot has no significant error in its
acceleration at the initial moment in comparison with its impedance model (9).
However, as shown in Fig. 12, the acceleration estimation error becomes non-zero after
applying the interaction forces and having non-zero response of master impedance
model, and due to the initial position error of the master robot with respect to its
zero during the first 0.1 sec of motion as a result of convergence of master robot
trajectory to its impedance model response (Fig. 9). This performance is in accordance
with the presented discussion for obtaining the master estimated acceleration x m est by
Eq. (12). During this initial short time duration (about 0.1 sec) of motion, the
acceleration estimation has a bounded error that its upper bound is about 1.7 m s 2 in
each direction of Cartesian space (Fig. 12). This upper bound should be estimated
designer to have a better initial estimation. Based on the inequality (60) and the
control law (20) is adjusted on 3.5 as mentioned in Table 1. Due to Figs. 9-11, it is seen
that the slave robot can successfully achieve its desired trajectory ( x mod s ) and also the
38
Journal of Dynamic Systems, Measurement, and Control
slave control law (20) (which is replaced by s tanh(400 Ss ) to avoid chattering) can
provided the robustness against the bounded estimation error of the master
acceleration ( x m ).
master acceleration estimation decreases by specifying the same initial position and
velocity for the master robot and its corresponding impedance model which results in
decreasing the initial master position and velocity errors. Under this condition, the
accuracy of Eq. (12) in estimating the actual master acceleration using the dynamics of
its impedance model increases and the transient value of xm xm est xm decreases.
Fig. 13. This figure shows the estimation of ten dynamic parameters introduced in Eq.
(69) for the master (Fig. 13 a) and slave (Fig. 13 b) robots. All of the model parameters of
adaptation laws (48) and (49) of both robots, an arbitrary initial guess
αˆ q initial [0.058 0.53 0.43 15.58 5.2 5.3 15 6 2.6 1.2]T is used for the vector of
αq actual [0.875 2 0.5 9.81 19.62 4.905 3 15 1.5 2.5]T (in SI units).
As it is seen in Fig. 13, the estimation process has almost settled after about
seven seconds. As mentioned before in Sec. 3.4 about the BGF composite adaptation
39
Journal of Dynamic Systems, Measurement, and Control
laws and in the stability proof (Sec. 3.5), when the persistent excitation conditions
m (t ) 1,m 0 and s (t ) 1,s 0 are satisfied for both master and slave, the
0,i 3.5 , k0,i 100 and Pi initial 33 I in the adaptation laws (48)-(52), the value of
i (t ) for the master ( i m ) and slave ( i s ) during the motion is obtained and shown
in Fig. 14.
As it is clear in Fig. 14, i (t ) is always positive during the whole motion and
laws (48) and (49) can estimate the actual parameters of the master and slave robots as
their convergence is shown in Fig. 13. This can also be realized from the comparison of
the final estimations of the parameters (Fig. 13) and their actual values:
telesonography applications an adjustable flexibility is desired for the slave robot in its
and ks ) of the slave impedance model (10) should be decreased in comparison with the
in (10), the slave robot becomes more flexible in its interactions with the environment
40
Journal of Dynamic Systems, Measurement, and Control
telerehabilitation system that the slave robot moves the patients’ limbs, it can have a
limited flexibility and the patient is permitted to deviate from the initially designed
In this case, the dynamic models of therapist (human operator) and the patient
where f h and f e are respectively the active forces of therapist and patient generated
[26, 27, 46] where the human arm parameters are presented in Table 4.
Also, the active force of therapist (operator) and patient (environment) are considered
[28, 29]):
parameters of the master impedance model (9) are chosen to be small (to obtain force
tracking performance). Moreover, to have a flexible interaction between the slave robot
41
Journal of Dynamic Systems, Measurement, and Control
and the environment, values of the slave impedance parameters in (10) should be
chosen to be as small as needed for its flexibility. Also, the position and force scaling
interaction force on the therapist hand and to follow the exact trajectory of the master
robot by the slave. The values of these impedance parameters and scaling factors are
givrn in Table 5 for telerehabilitation applications. Similar to the previous section, using
these parameters, the master impedance model (9) has two negative real poles
( 2 , 2 ), and the slave impedance model (10) has two negative real poles ( 20 , 20 ).
Accordingly, the master impedance model has a natural frequency of 2 rad/sec and a
damping ratio of 1 , and the slave impedance model has a natural frequency of
should have flexibility in physical interactions, the slave impedance parameters in Table
application). Also, similar to the previous section, the master impedance parameters are
Sec. 3.1.
forces between the therapist (operator) and the master robot ( f h ), and the interaction
force between the patient (environment) and the slave robot ( kf fe ) in z direction are
42
Journal of Dynamic Systems, Measurement, and Control
therapist has a full haptic force feedback from the patient force. The force tracking error
( fh kf fe ) in z direction is shown in Fig. 16. As seen in Fig. 16, the mean value of the
force error is 0.42 N (less than 0.5% of the maximum interaction force, i.e. 86 N in Fig.
14). The source of the small force error in Fig. 16 is the small master impedance
The position, velocity and acceleration responses of master and slave impedance
models, and the master and slave robots’ trajectories in z direction are shown in Fig.
17. According to Fig. 17, the position tracking performance is weak, which is the result
provide flexibility, as discussed in Sec. 3.1. The absolute position tracking error between
master and slave robots ( x s kp xm ) in x z plane is shown in Fig. 18. Also, the
2
absolute position tracking error of the master and slave robots with respect to their
In Fig. 18, the absolute tracking error between the master and slave robots are
considerable due to realizing a flexibility for the patient that can deviate from the
master trajectory. Therefore, the main part of this tracking error is the desired slave
43
Journal of Dynamic Systems, Measurement, and Control
error with respect to the scaled master position xmod xmod kp xm . This error becomes
s s
larger in comparison with the previous simulations of telesurgery operation (Sec. 4.1) as
a result of choosing smaller parameters for the slave impedance model (10), as
discussed in Sec. 3.1. Also, the master and slave tracking errors with respect to their
and converge to zero that indicates a suitable performance of master and slave adaptive
controllers.
master robot trajectory tracks its impedance response ( xm xmodm , xm xmodm and
xm xmodm as shown in Figs. 9 and 17). Therefore, the master robot behaves similar to
its desired impedance model (9) in response to the operator and environment forces
position, velocity and acceleration of the master ( x m , x m and x m ) specify the force
tracking error (fh kf fe ) . On the other hand, the human operator can enforce his
force ( f h ) according to Eq. (78). However, by increasing the absolute values of master
44
Journal of Dynamic Systems, Measurement, and Control
Therefore, the bound and mean values of force tracking error are related to the
selected master impedance parameters and the generated master trajectory by the
interaction forces. In the presented simulations, the master impedance parameters are
chosen to be small (as listed in Tables 3 and 5) to decrease the force tracking error while
the master impedance model stability is also guaranteed. Under this condition, for the
common motions of master and consequently slave robots (such as in Figs. 9 and 17),
the force tracking error (as shown in Figs. 8 and 16) is small in comparison with the
mentioned previously, the mean value of the force error in Figs. 8 and 16 is around 0.5%
surgery is of great importance since it may have significant impact on the performance
of the bi-lateral control system. To investigate such condition for the proposed
increased to ke 10000 N/m (as assumed previously in [53]). The other parameters of
the human operator and tissue are considered the same as those presented in Table 2
for models (71) and (72). Also, the components of f h as the active force of the human
operator’s hand in Eq. (71) are considered to be fhx fhz 30 30 cos(t ) N . The initial
positions of the robots and their impedance models are the same as those in Sec. 4.1. To
45
Journal of Dynamic Systems, Measurement, and Control
impedance models (9) and (10) are adjusted similar to Sec. 4.1 (listed in Table 3) for
As seen in Fig. 19, when the position of the slave end-effector reaches the
surface of the hard tissue that is located on ze 0.66 m , the contact starts ( t 0.88 sec )
and the environment interaction force ( f e ) becomes non-zero. As shown in Fig. 19, the
force tracking error temporarily becomes larger at the starting moments of this
intermittent hard contact and then decreases. However, this force error is small after
and before the contact when the salve robot has a free (non-contact) motion based on
previous discussions on Eq. (78). The velocity and acceleration values have sudden
changes at the starting time of the contact which are not shown for the sake of brevity.
Also, the estimation error of the master acceleration that converges to zero before the
contact, has bounded jumps (with the absolute maximum value of 1.8 m sec2 ) at the
start of contact ( t 0.88 sec ) as illustrated in Fig. 18. However, the controller is robust
against this bounded acceleration error as mentioned previously in sections 3.2, 3.5 and
4.1. Accordingly, the proposed controller provides the stability of the teleoperation
system during such hard intermittent contact; however, bounded transient force
tracking error and acceleration estimation error may occur at the start of this kind of
contact.
5. CONCLUSIONS
46
Journal of Dynamic Systems, Measurement, and Control
asymptotic tracking of two desired impedance models for master and slave robots in
and slave robots. The master impedance model determines the haptic sense of the
operator in its physical interaction, and the slave impedance model establishes the
that are position and force tracking are followed simultaneously for each robot. As
discussed in Sec. 1, reaching these two objectives for each DOF of robot (in Cartesian
coordinates) with one control input is difficult. However, in the proposed nonlinear
adaptive impedance controller, one control objective (realizing the impedance model)
can be provided for each DOF with one control input. Moreover, unlike some previous
The system dynamics, impedance models, control laws and Lyapunov stability
proof are all presented in Cartesian coordinates. However, the BGF composite
adaptation laws are defined in the joint space to decrease the computational
complexity, especially for higher DOF robots. Accordingly, the terms of the Lyapunov
47
Journal of Dynamic Systems, Measurement, and Control
The presented simulations on a two DOFs master and slave robots show the
models.
application, the parameters of the two impedance models can be chosen according to
the requirements and objectives. For example, in tele-surgery applications, the slave
(compliance). In this case, the slave impedance parameters should be increased as much
slave robot should have higher compliance in its interaction with the patient
parameters of the master impedance model can be chosen according to the haptic
communication channels of the proposed teleoperation system will affect the stability
proof and transparency analysis. These issues may be the subjects of some future works
REFERENCES
[1] Tavakoli, M., Aziminejad, A., Patel, R. V., and Moallem, M., 2006, "Methods and
Mechanisms for Contact Feedback in a Robot-Assisted Minimally Invasive
48
Journal of Dynamic Systems, Measurement, and Control
49
Journal of Dynamic Systems, Measurement, and Control
[17] Polushin, I. G., Liu, P. X., Lung, C.-H., and On, G. D., 2010, "Position-Error Based
Schemes for Bilateral Teleoperation with Time Delay: Theory and Experiments," Journal
of Dynamic Systems, Measurement, and Control, 132(3), pp. 031008 (11 pages).
[18] Lee, H. K., and Chung, M. J., 1998, "Adaptive Controller of a Master–Slave System
for Transparent Teleoperation," Journal of Robotic Systems, 15(8), pp. 465-475.
[19] Shi, M., Tao, G., and Liu, H., 2002, "Adaptive Control of Teleoperation Systems,"
Journal of X-Ray Science and Technology, 10(1-2), pp. 37-57.
[20] Wen-Hong, Z., and Salcudean, S. E., 2000, "Stability Guaranteed Teleoperation: An
Adaptive Motion/Force Control Approach," Automatic Control, IEEE Transactions on,
45(11), pp. 1951-1969.
[21] Malysz, P., and Sirouspour, S., 2009, "Nonlinear and Filtered Force/Position
Mappings in Bilateral Teleoperation with Application to Enhanced Stiffness
Discrimination," Robotics, IEEE Transactions on, 25(5), pp. 1134-1149.
[22] Lee, D., and Spong, M. W., 2006, "Passive Bilateral Teleoperation with Constant
Time Delay," Robotics, IEEE Transactions on, 22(2), pp. 269-281.
[23] Chopra, N., Spong, M. W., and Lozano, R., 2008, "Synchronization of Bilateral
Teleoperators with Time Delay," Automatica, 44(8), pp. 2142-2148.
[24] Ryu, J.-H., and Kwon, D.-S., 2001, "A Novel Adaptive Bilateral Control Scheme
Using Similar Closed-Loop Dynamic Characteristics of Master/Slave Manipulators,"
Journal of Robotic Systems, 18(9), pp. 533-543.
[25] Hung, N. V. Q., Narikiyo, T., and Tuan, H. D., 2003, "Nonlinear Adaptive Control
of Master–Slave System in Teleoperation," Control Engineering Practice, 11(1), pp. 1-10.
[26] Liu, X., and Tavakoli, M., 2012, "Adaptive Control of Teleoperation Systems with
Linearly and Nonlinearly Parameterized Dynamic Uncertainties," Journal of Dynamic
Systems, Measurement, and Control, 134(2), pp. 021015 (10 pages).
[27] Liu, X., and Tavakoli, M., 2011, "Adaptive Inverse Dynamics Four-Channel Control
of Uncertain Nonlinear Teleoperation Systems," Advanced Robotics, 25(13-14), pp.
1729-1750.
[28] Sharifi, M., Behzadipour, S., and Vossoughi, G., 2014, "Nonlinear Model Reference
Adaptive Impedance Control for Human–Robot Interactions," Control Engineering
Practice, 32(pp. 9-27.
[29] Sharifi, M., Behzadipour, S., and Vossoughi, G. R., 2014, "Model Reference
Adaptive Impedance Control in Cartesian Coordinates for Physical Human–Robot
Interaction," Advanced Robotics, 28(19), pp. 1277-1290.
[30] Sharifi, M., Behzadipour, S., and Vossoughi, G. R., 2012, "Model Reference
Adaptive Impedance Control of Rehabilitation Robots in Operational Space," eds., pp.
1698-1703.
[31] Hogan, N., 1985, "Impedance Control: An Approach to Manipulation: Part I---
Theory," Journal of Dynamic Systems, Measurement, and Control, 107(1), pp. 1-7.
[32] Kazerooni, H., Sheridan, T., and Houpt, P., 1986, "Robust Compliant Motion for
Manipulators, Part I: The Fundamental Concepts of Compliant Motion," Robotics and
Automation, IEEE Journal of, 2(2), pp. 83-92.
50
Journal of Dynamic Systems, Measurement, and Control
[33] Lu, Z., and Goldenberg, A. A., 1995, "Robust Impedance Control and Force
Regulation: Theory and Experiments," The International Journal of Robotics Research,
14(3), pp. 225-254.
[34] Anderson, R. J., and Spong, M. W., 1988, "Hybrid Impedance Control of Robotic
Manipulators," Robotics and Automation, IEEE Journal of, 4(5), pp. 549-556.
[35] Lu, W. S., and Meng, Q. H., 1991, "Impedance Control with Adaptation for Robotic
Manipulations," Robotics and Automation, IEEE Transactions on, 7(3), pp. 408-415.
[36] Chien, M., and Huang, A., 2004, "Adaptive Impedance Control of Robot
Manipulators Based on Function Approximation Technique," Robotica, 22(04), pp. 395-
403.
[37] Abdossalami, A., and Sirouspour, S., 2009, "Adaptive Control for Improved
Transparency in Haptic Simulations," Haptics, IEEE Transactions on, 2(1), pp. 2-14.
[38] Seraji, H., 1994, "Adaptive Admittance Control: An Approach to Explicit Force
Control in Compliant Motion," eds., 4, pp. 2705-2712.
[39] Tee, K. P., Yan, R., and Li, H., 2010, "Adaptive Admittance Control of a Robot
Manipulator under Task Space Constraint," eds., pp. 5181-5186.
[40] Rubio, A., Avello, A., and Florez, J., 1999, "Adaptive Impedance Modification of a
Master-Slave Manipulator," eds., 3, pp. 1794-1799.
[41] Dubey, R. V., Tan Fung, C., and Everett, S. E., 1997, "Variable Damping Impedance
Control of a Bilateral Telerobotic System," Control Systems, IEEE, 17(1), pp. 37-45.
[42] Love, L. J., and Book, W. J., 2004, "Force Reflecting Teleoperation with Adaptive
Impedance Control," Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE
Transactions on, 34(1), pp. 159-165.
[43] Cho, H. C., and Park, J. H., 2005, "Stable Bilateral Teleoperation under a Time
Delay Using a Robust Impedance Control," Mechatronics, 15(5), pp. 611-625.
[44] Garcaa-Valdovinos, L. G., Parra-Vega, V., and Arteaga, M. A., 2007, "Observer-
Based Sliding Mode Impedance Control of Bilateral Teleoperation under Constant
Unknown Time Delay," Robotics and Autonomous Systems, 55(8), pp. 609-617.
[45] Abbott, J. J., and Okamura, A. M., 2007, "Pseudo-Admittance Bilateral
Telemanipulation with Guidance Virtual Fixtures," The International Journal of Robotics
Research, 26(8), pp. 865-884.
[46] Liu, X., Tao, R., and Tavakoli, M., 2014, "Adaptive Control of Uncertain Nonlinear
Teleoperation Systems," Mechatronics, 24(1), pp. 66-78.
[47] Craig, J. J., 2005, Introduction to Robotics: Mechanics and Control,
Pearson/Prentice Hall,
[48] Spong, M. W., and Hutchinson, S., 2005, Robot Modeling and Control, Wiley,
[49] Slotine, J. J. E., and Li, W., 1991, Applied Nonlinear Control, Prantice-Hall, NJ,
Englewood Cliffs.
[50] Khatib, O., 1987, "A Unified Approach for Motion and Force Control of Robot
Manipulators: The Operational Space Formulation," Robotics and Automation, IEEE
Journal of, 3(1), pp. 43-53.
[51] Sirouspour, S., and Setoodeh, P., 2005, "Adaptive Nonlinear Teleoperation Control
in Multi-Master/Multi-Slave Environments," eds., pp. 1263-1268.
51
Journal of Dynamic Systems, Measurement, and Control
[52] Speich, J. E., Shao, L., and Goldfarb, M., 2005, "Modeling the Human Hand as It
Interacts with a Telemanipulation System," Mechatronics, 15(9), pp. 1127-1142.
[53] Hashtrudi-Zaad, K., and Salcudean, S. E., 1996, "Adaptive Transparent Impedance
Reflecting Teleoperation," eds., 2, pp. 1369-1374.
channels
Fig. 2 The signal transmissions in the proposed bilateral controller with three
communication channels
Fig. 3 The haptic sense of the human operator using the master impedance
model (9)
Fig. 4 The concepts of the master and slave impedance models in the proposed
Fig. 6 The two-DOF robot manipulator with revolute and prismatic joints used
Fig. 7 Interaction forces between the surgeon and master ( f h ), and between
Fig. 9 Position, velocity and acceleration of the master and slave robots
52
Journal of Dynamic Systems, Measurement, and Control
Fig. 10 The absolute position tracking errors in x z plane: the master error
with respect to its impedance model xm xmod (blue dashed line), the
m 2
slave error with respect to its impedance model x s x mod (red dash-
s 2
line).
Fig. 12 The actual and estimation values of the master acceleration ( x m and
Fig. 13 The estimation of ten dynamic parameters for (a) master and (b) slave
Fig. 14 The value of the master and slave forgetting factors m (t ) and s (t )
Fig. 15 Interaction forces between the therapist and the master robot ( f h ), and
Fig. 17 Position, velocity and acceleration of the master and slave robots
Fig. 18 The absolute position tracking errors in x z plane: the master error
53
Journal of Dynamic Systems, Measurement, and Control
line).
tissue, the contact starts at t 0.88 sec when the slave end-effector reach
54
Journal of Dynamic Systems, Measurement, and Control
Table 1 The parameter values of the robots, control laws and adaptation laws
tissue environment
Table 3 Master and slave impedance parameters with scaling factors for
telesurgery application
(environment)
Table 5 Master and slave impedance parameters together with scaling factors for
telerehabilitation application
55
Journal of Dynamic Systems, Measurement, and Control
56
Journal of Dynamic Systems, Measurement, and Control
Fig. 2 The signal transmissions in the proposed bilateral controller with three
communication channels
57
Journal of Dynamic Systems, Measurement, and Control
Fig. 3 The haptic sense of the human operator using the master impedance model (9)
58
Journal of Dynamic Systems, Measurement, and Control
Fig. 4 The concepts of the master and slave impedance models in the proposed bilateral
control strategy
59
Journal of Dynamic Systems, Measurement, and Control
60
Journal of Dynamic Systems, Measurement, and Control
Fig. 6 The two-DOF robot manipulator with revolute and prismatic joints used for the
61
Journal of Dynamic Systems, Measurement, and Control
50
f
Interaction Forces : f h and f e (N)
h
40
f
e
30
20
10
-10
0 2 4 6 8 10 12 14
Time (sec)
Fig. 7 Interaction forces between the surgeon and master ( f h ), and between the slave
62
Journal of Dynamic Systems, Measurement, and Control
2
Force Tracking Error: f h-k f f e (N)
1.5
0.5
-0.5
-1
0 2 4 6 8 10 12 14
Time (sec)
63
Journal of Dynamic Systems, Measurement, and Control
0.67
0.66
Position in x direction (m)
0.65
0.64
0.06
0.04
0.02
-0.02
-0.04
0 2 4 6 8 10 12 14
Time (sec)
0.2
Acceleration in x direction (m/s )
2
0.1
-0.1
-0.2
0 2 4 6 8 10 12 14
Time (sec)
Fig. 9 Position, velocity and acceleration of the master and slave robots together with
their impedance model responses in x direction
64
Journal of Dynamic Systems, Measurement, and Control
0.03
0.015
0.01
0.005
0
0 2 4 6 8 10 12 14
Time (sec)
Fig. 10 The absolute position tracking errors in x z plane: the master error with
65
Journal of Dynamic Systems, Measurement, and Control
0.2
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec)
66
Journal of Dynamic Systems, Measurement, and Control
0.5
-0.5
-1
-1.5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (sec)
Fig. 12 The actual and estimation values of the master acceleration ( x m and x m est )
together with their difference (estimation error xm xm est xm ) in x direction, where
67
Journal of Dynamic Systems, Measurement, and Control
20
q,m 1
18
q,m 2
Estimated Parameters of Master Robot
16
q,m 3
14
q,m 4
12
q,m 5
10
q,m 6
8
q,m 7
6
q,m 8
4
q,m 9
2
q,m 10
0
0 2 4 6 8 10 12 14
Time (sec)
(a)
20
q,s 1
18
q,s 2
Estimated Parameters of Slave Robot
16
q,s 3
14
q,s 4
12
q,s 5
10
q,s 6
8
q,s 7
6
q,s 8
4
q,s 9
2
q,s 10
0
0 2 4 6 8 10 12 14
Time (sec)
(b)
Fig. 13 The estimation of ten dynamic parameters for (a) master and (b) slave
68
Journal of Dynamic Systems, Measurement, and Control
3.5
Forgetting Factors of Adaptation Laws
3 m
s
2.5
1.5
0.5
0
0 2 4 6 8 10 12 14
Time (sec)
Fig. 14 The value of the master and slave forgetting factors m (t ) and s (t )
69
Journal of Dynamic Systems, Measurement, and Control
100
Interaction Forces : f h and f e (N)
f
80 h
f
e
60
40
20
-20
0 2 4 6 8 10 12 14
Time (sec)
Fig. 15 Interaction forces between the therapist and the master robot ( f h ), and
70
Journal of Dynamic Systems, Measurement, and Control
4
Force Tracking Error: f h- k f f e (N)
-1
-2
-3
-4
0 2 4 6 8 10 12 14
Time (sec)
71
Journal of Dynamic Systems, Measurement, and Control
0.9
Position in z direction (m)
0.85
0.8
0.75
0.2
-0.2
-0.4
0 2 4 6 8 10 12 14
Time (sec)
1.5
Acceleration in z direction (m/s )
2
0.5
-0.5
-1
-1.5
0 2 4 6 8 10 12 14
Time (sec)
Fig. 17 Position, velocity and acceleration of the master and slave robots together with
their impedance model responses in z direction
72
Journal of Dynamic Systems, Measurement, and Control
0.035
0.03
Position Tracking Error (m)
0.025
0.02
0.005
0
0 2 4 6 8 10 12 14
Time (sec)
Fig. 18 The absolute position tracking errors in x z plane: the master error with
73
Journal of Dynamic Systems, Measurement, and Control
0.67
0.66
Position in z direction (m)
0.65
0.63
Master Imp. Model
0.62 Master Robot
Slave Imp. Model
0.61
Slave Robot
0.6
0 1 2 3 4 5 6 7
Time (sec)
2
Force Tracking Error: f h-k f f e (N)
-2
-3
0 1 2 3 4 5 6 7
Time (sec)
5
Acceleration estimation (m/s )
2
-5 Contact
Actual Master Acceleration
Estimated Master Acceleration
Estimation Error of Master Acceleration
-10
0 0.2 0.4 0.6 0.8 1 1.2
Time (sec)
Fig. 19 Position and force tracking performance together with master acceleration
estimation during the intermittent contact with a hard tissue, the contact starts at
t 0.88 sec when the slave end-effector reach to the surface of tissue ( ze 0.66 m ) and
74
Journal of Dynamic Systems, Measurement, and Control
Table 1 The parameter values of the robots, control laws and adaptation laws
m1 4 kg , m2 2 kg am as 1 1,m 1,s 1
I1 0.083 kg.m2 ,
0,m 0,s 5 2,m 2,s 60
I 2 0.0415 kg.m2
l1 0.5 m , l2 0.5 m k0,m k0,s 100 s 3.5
l1g 0.25 m , l2 g 0.25 m Pm initial Ps initial 33 I66
g 9.81 m/s2
c1 3 N.m.s , c2 15 N.s/m
1 1.5 N.m , 2 2.5 N
75
Journal of Dynamic Systems, Measurement, and Control
Table 2 Dynamic parameters of a human operator (surgeon)’s arm and soft tissue
environment
Parameters of the Surgeon Parameters of the Soft Tissue
(Operator) (Environment)
mh 3.25 kg me 1 kg
ch 20 N.s/m ce 40 N.s/m
kh 300 N/m ke 1500 N/m
76
Journal of Dynamic Systems, Measurement, and Control
Table 3 Master and slave impedance parameters with scaling factors for the telesurgery
application
Master impedance Slave impedance Position and force
parameters parameters scaling factors
mm 1 kg ms 1000 kg kp 1
cm 4 N.s/m cs 40000 N.s/m kf 1
km 4 N/m ks 400000 N/m
77
Journal of Dynamic Systems, Measurement, and Control
mh 3.25 kg me 3.25 kg
ch 20 N.s/m ce 20 N.s/m
kh 300 N/m ke 300 N/m
78
Journal of Dynamic Systems, Measurement, and Control
Table 5 Master and slave impedance parameters together with scaling factors for the
telerehabilitation application
Master impedance Slave impedance Position and force
parameters parameters scaling factors
mm 0.5 kg ms 10 kg kp 1
cm 2 N.s/m cs 400 N.s/m kf 1
km 2 N/m ks 4000 N/m
79