Proceedings of the 17th World Congress

The International Federation of Automatic Control

Seoul, Korea, July 6-11, 2008

Force Tracking Impedance Control with

Variable Target Stiffness
K. Lee and M. Buss ∗
∗
Institute of Automatic Control Engineering,
Technische Universität München, Munich, 80290 Germany
(E-mail: [email protected])

Abstract: In this paper, a novel force tracking impedance control strategy is presented in which
target stiffness is varied on-line to regulate the desired contact force without any knowledge
of the environment. Humans can control contact force by adjusting their arm stiffness. The
contact force can be either increased by making one’s arm stiffer or decreased by reducing the
arm stiffness. Furthermore, humans can keep the force tracking error within a certain range
without any knowledge of environmental parameters as long as how much force they exert
on the object is known to them. Analogously, the proposed control scheme achieves a contact
force regulation control by adjusting the target stiffness of the impedance control. The new force
tracking impedance control scheme does not require estimating environment stiffness or locations
since the controller is adapted only based on the previous force tracking error between the desired
and real contact force. Stability of the proposed scheme is discussed with a quadratic Lyapunov
function. Extensive simulation studies with a 7 degree of freedom (DoF) robot manipulator
using full arm dynamics are conducted to demonstrate the validity of the proposed scheme.

1. INTRODUCTION be achieved since it controls the interaction force indirectly

by regulating the dynamic relationship between the EEF
As robot manipulators have evolved and their applications position and the interaction force. Therefore, the inability
have been broadened, interaction (compliance) control of of the impedance control strategy to achieve force track-
robot manipulators with stiff environments or objects has ing control has been considered as a major disadvantage
become a key component for the success of many manipu- compared to hybrid position/force control.
lation tasks. Typical applications of the compliance control As a remedy to this issue, many efforts have been made to
are deburring, assembling, grinding, and surface finishing achieve force tracking control within the impedance con-
tasks, etc. Due to the central importance of such compliant trol framework. Some researchers introduce a sliding mode
motions for robot manipulators, vast investigations have type of control based on the impedance control scheme
been made on this challenge over the past two decades re- to achieve force tracking characteristics for autonomous
sulting in two basic classes: hybrid position/force control, robots as well as teleoperation systems, see e.g. Hace et al.
e.g. Raibert et al. (1981); Khatib. (1987), and impedance (1997); Cho et al. (2001); Iwasaki et al. (2003). Seraji et al.
control, e.g. Hogan. (1985, 1988). (1997) present an adaptive control scheme that generates
The hybrid position/force control is suitable where the a desired trajectory in order to regulate a desired contact
environments are well structured and their geometrical in- force either based on the estimation of the environmental
formation is previously known, since this strategy allows to stiffness and location or without the estimation. Chan
control positions in unconstrained directions and interac- et al. (1991) proposes a variable structure control scheme
tion forces in constrained directions explicitly as required based on precise environmental knowledge (stiffness and
in many applications. However, it requires to decompose position of objects) for the robust impedance control un-
the task space of the manipulator into two subspaces: a der parametric uncertainties and external disturbances.
position space and a force space corresponding to those In practical cases, however, the location and stiffness of
end-effector (EEF) directions of which either the position environment are usually unknown and difficult to estimate
or the interaction force is to be controlled, respectively. accurately. Furthermore, an inaccurate estimation of these
Therefore, it may not be a promising approach for un- parameters usually leads to a large force tracking error.
structured and dynamically changing environments. Jung et al. (2004) propose an adaptive impedance control
in which null stiffness is assigned for the constrained mo-
Alternatively, the impedance control strategy provides
tion control to guarantee a zero force tracking error com-
compliant manipulator motions in a unified framework for
bined with an adaptive control feature which makes the
both constrained and unconstrained directions. The core
system robust to the uncertainties in both robot dynamics
of the impedance control is to regulate dynamic response
and environment parameters.
of the EEF to interaction forces by establishing a suitable
virtual mass-spring-damper system on the EEF. Although However, considering human force control capabilities
the impedance control scheme can provide a unified frame- when exerting a certain force on a digital scale, neither null
work for both constrained and unconstrained motion con- stiffness nor variation of the desired trajectory is natural
trol problems, an explicit interaction force control can not and intuitive. While one’s hand is in contact with the en-

978-3-902661-00-5/08/$20.00 © 2008 IFAC 6751 10.3182/20080706-5-KR-1001.2474

17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

vironment, the desired trajectory of the hand or the exact For simplicity, we consider a one dimensional case as shown
knowledge of environment stiffness is not required. On the in Fig. 2-(a), in which an EEF of a robot manipulator only
other hand, one cannot exert contact forces with null arm just contacts with a wall without generating a contact
stiffness. Thus, suitable force tracking impedance control force. It can be easily extended to the entire task space
schemes should consider the fact that humans control the without loss of generality. In this case, (1) is reduced to
force exerted on an object by adapting their arm stiffness. mẍ + bẋ + k(xd − xc ) = f. (2)
The objective of this study is to develop an intuitive and
Further, assuming good tracking performance of the inner
anthropomorphic force tracking control scheme within the
position control loop, the compliant trajectory xc can be
impedance control framework. The main idea is to adapt
reached by the EEF (xc = x). Then the contact force at
the target stiffness of the impedance controller according
the steady state
to force tracking errors. Since the adaptation of the target
stiffness depends only on the previous force tracking error, f = k(xd − xc ) (3)
the knowledge of the environment stiffness is not required. can be calculated. Representing the environment by a
Thus, the proposed force tracking impedance control is linear spring model with stiffness ke , the contact force can
simple and robust to environmental parameter variation. be alternatively expressed by
This paper is structured as follows. Section 2 reviews
f = ke (x − xe ). (4)
briefly the conventional impedance control and a novel
force tracking impedance control scheme with variable Substituting x (= xc ) from (4) into (3), the steady state
stiffness is developed. Stability and force tracking errors contact force can be rewritten as
of the proposed control scheme are discussed in Section 3. kke
Intensive dynamic simulations are presented in Section 4 f= (xd − xe ). (5)
k + ke
to demonstrate the efficiency of the proposed scheme for
Since the wall position xe and stiffness ke are environment
various environments including variations of the environ-
parameters, (5) reveals two possible strategies to control
mental parameters. Section 5 describes the conclusions
the contact force; by tuning either 1) the desired trajec-
drawn from this work.
tory xd or 2) the target stiffness k. However, a desired
trajectory planning that guarantees force tracking in the
2. FORCE TRACKING IMPEDANCE CONTROL
constrained space is not intuitive. Moreover, it can be
A position based impedance control is one of the most typ- noted that small numerical errors in the calculation of xd
kke
ical compliance control schemes which is briefly reviewed will be multiplied by the resulting stiffness ( k+k e
) leading
in the Section 2.1. It is modified with variable stiffness for to significant force tracking errors. On the contrary, the
a force tracking impedance control scheme in Section 2.2. contact force controllability can be improved by controlling
the target stiffness that affects the contact force in har-
kke
2.1 Position based impedance Control monic mean fashion ( k+k e
) incorporating the environment
stiffness ke .
A position based impedance control scheme consists of
an inner position control loop and an outer indirect force One interesting thing to be noticed here is negative target
control loop, see Fig. 1. In free space motion (zero contact stiffness. When the desired trajectory xd is set to penetrate
forces F ), the desired trajectory Xd is identical to the the wall location xe far enough, i.e., reaching xd generates
compliant trajectory Xc since no compliant motions are a contact force greater than the desired contact force fd ,
necessary. In constrained motions, however, a nonzero then a positive stiffness modifies xd in the same direction of
contact force modifies the desired trajectory in the outer the contact force f by (3). However, when the penetration
impedance control loop resulting in the compliant desired of xd in xe is not deep enough, i.e., even reaching xd
trajectory, which is to be tracked by the inner motion can only produce a contact force f smaller than the
control loop. One of the common formulations for the desired force fd , the positive stiffness will not help at
target impedance is all. In this case, one needs a negative stiffness which can
modify the desired trajectory xd in the opposite direction
M Ẍdc + B Ẋdc + KXdc = F, (1) to the contact force. Fig. 2-(b) shows a modification of
where M , B, K are n×n constant diagonal mass, damping, the desired trajectory versus target stiffness. From the
and stiffness matrices of the target impedance for an n- above observations, a force tracking impedance control is
dimensional task space; F and Xdc denote the contact
force and the corresponding desired trajectory modifica-
tion (Xd − Xc ) by the impedance controller, respectively.
Forward
Kinematics

Impedance Motion Robot/

Control Control Environment wall

a) b)

Fig. 1. Position based impedance control strategy: Xd , Xc , Fig. 2. Robot EEF contacting with a wall (a), and stiffness
and X are the desired, compliant, and EEF trajectory, versus desired trajectory modification ∆x at a given
respectively. F is the measured force. contact force 10[N ] (b)

6752
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

Impedance Filter
Forward On the other hand, as mentioned in the previous section,
Kinematics
it is not possible to obtain the desired contact force fd
Motion Robot/ with a positive target stiffness for some cases. Consider xd
Control Environment
does not penetrate enough in xe such that the maximum
achievable contact force f is smaller than the desired
contact force fd ; fd > f = ke (xd − xe ). In this case, the
negative target stiffness can modify the desired trajectory
xd such that the compliant trajectory xc can further
penetrate xe such that the desired force can be achieved.
The proposed scheme does not pose any limit on the sign of
Fig. 3. Position based force tracking impedance control the target stiffness so that the time varying target stiffness
with variable parameters k(t) may have a negative sign depending on the sign of
α. Since neither negative nor time varying target stiffness
developed in the next section with variable stiffness of the is widely adopted in robotics, stability of the proposed
target impedance. scheme is discussed in the next section.
2.2 Impedance Control with variable stiffness 3. STABILITY OF TIME VARYING IMPEDANCE
The idea is simple, intuitive, and anthropomorphically CONTROL
inspired. Consider a hand in contact with a wall. In
order to exert more force, humans make their arm stiffer, The proposed impedance controller with variable target
whereas they make it softer when the contact force should stiffness in (6) is a specific case of
be reduced. It means the arm stiffness is adapted to the m(t)ẍdc + b(t)ẋdc + k(t)xdc = f (14)
difference of a desired and actual contact force. In order to which can be characterized as a second order linear time
imitate this human force control capability, an impedance varying system. Thereby m(t) and b(t) are time varying
control with variable target stiffness is proposed. The positive inertia and damping coefficients, and k(t) time
corresponding impedance equation can be written as varying stiffness. Since such systems are of significant
mẍdc + bẋdc + k(t)xdc = f. (6) importance in control theory many efforts have been made
for the one dimensional case. Thereby, b denotes a constant to provide explicit stability condition, see e.g., Harris.
damping coefficient, and the stiffness k(t) is adapted to (1980); Rugh. (1996); Slotine. (1991); Gil. (2005). Many of
force tracking errors to minimize them. The resulting force those stability analyses assume slowly varying and positive
tracking impedance control scheme is illustrated in Fig. 3. definite parameters. Hence, these analyses can not apply
Denoting the force tracking error ef as to the proposed scheme and new stability analysis should
be devised.
ef = fd − f, (7)
Consider a force regulation problem (fd =const.) with a
the adaptation of k(t) is defined as more general linear time varying system (14). For obtain-
k(t) = αk0 x−1 ing a force tracking control the proposed control law (6)-
dc , (8)
(9) is applied to (14). In order to analyze the stability, we
with define the positive scalar Lyapunov function candidate
α = kf ef + kv ėf . (9) 1 1 1
V = mẋ2 + ke (x − xe )2 + kn e2f , (15)
Thereby kf and kv denote a constant proportional and 2 2 2
differential gain for the force tracking control, respectively. where the argument t is dropped. Thereby kn is a nominal
Substituting (6) into the force tracking error ef , (7) can positive stiffness which will be defined later and m is a
be written as time varying positive inertia. It is noted that the Lyapunov
ef = fd − mẍdc − bẋdc − k(t)xdc . (10) function candidate can be interpreted as a sum of kinetic
and potential energy for the motion control and the force
Combining (10) with the proposed force control law (8) tracking control. Next, we assume a constant position
and (9) yields the force tracking error dynamics tracking error δx in the inner motion control loop xc =
ef = fd − mẍdc − bẋdc − kf0 ef − kv0 ėf (11) x + δx and a constant desired trajectory xd , (14) can be
rewritten as
⇒ kv0 ėf + (kf0 + 1)ef = fd − mẍdc − bẋdc (12)
mẍ = −fd − bẋ + kf0 (ef + 1) + kv0 ėf . (16)
with kf0
= k0 kf and kf0
= k0 kv . Consequently, the steady Furthermore, due to the constant desired force fd , the
state force tracking error becomes force tracking error ef and its first time derivative can be
fd fd written with a linear environment model f = ke (x − xe )
ef,ss = 0 = . (13) as
kf + 1 k0 kf + 1
From (13), it is noted that the proposed control does not ef = fd − f = fd − ke (x − xe ), (17)
provide a zero force tracking error for none zero desired ėf = −f˙ = −ke ẋ, (18)
force. However, the presented method can guarantee to
keep steady state force tracking error below the measure- for a flat environment ẋe = 0.
ment resolution of conventional load cells. For instance, On the other hand, the time differentiation of (15) gives
the steady state force tracking error ef,ss is smaller than 1
V̇ = ṁẋ2 + mẍẋ + ke (x − xe )ẋ + kn ef ėf , (19)
10−4 N for kf =103 , k0=103 N/m, and fd = 100 N. 2

6753
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

Fig. 5. Position tracking (left) and response of contact

Fig. 4. Physical (left) and schematic view with link coor- force (right) using pure impedance control; (m, b, k) =
dinate systems (right) of the 7-DoF dual arm robot (1, 102 , 103 ), (xe , xd ) = (0.4, 0.42) m, ke = 104 N/m

and substituting (16)-(18) into (19) yields 7 DoF. Both arms are built using commercially available
1 components combined with aluminum/steel construction
V̇ = ṁẋ2 − bẋ2 − kv0 ke ẋ2 + (kf0 − kn ke )ef ẋ. (20) elements. The actuation torque is provided by DC-motors
2 coupled with harmonic drive gears. It is equipped with
If we now define the positive nominal stiffness kn as six-axis JR3 force-torque sensors mounted on its EEF.
kf0 The joint angles are measured by digital MR-encoders.
kn = , (21) A single arm weighs approx. 13.5 kg with a payload of
ke
6 kg and a maximum reach of 0.86 m. In the following
then the last term on the right-hand side of (20) is canceled simulation studies the position based impedance control is
resulting in employed as illustrated in Fig. 3 with full robot dynamics
model and a sampling time 2 ms. The desired contact force
 
0 1
V̇ = − b + kv ke − ṁ ẋ2 ≤ 0, (22) is set to 50 N and the contact force is calculated by (4).
2
The simulation studies consist of seven examples including
which is negative semi-definite as long as b and ṁ satisfy variable geometries and stiffness of environment as well
the inequality as a contact simulation with a conventional impedance
2(b + kv0 ke ) ≥ ṁ. (23) control scheme. As a desired task, the EEF is required
Since the system is not autonomous, the invariant-set to keep its orientation perpendicular to a wall and move
theorem is not applicable to show an asymptotic behavior 0.4 m upward vertically while exerting 50 N to the wall.
of the system. However, it can be provided by invoking The following impedance parameters and nominal stiffness
Barbalat’s lemma. The Lyapunov function candidate V are used throughout the simulation studies: (m, b, k0 ) =
in (15) is lower bounded, since ẋ and ef are bounded, and (1, 102 , 103 ).
V̇ is negative semi-definite as shown above. Considering Pure Impedance Control
that the acceleration of the EEF is bounded in physical For comparison the force tracking performance of a pure
systems it is reasonable to conclude that impedance control is illustrated in Fig. 5. As discussed in
1  1 Section 1 and 2 the conventional impedance control regu-
V̈ = − 2ḃ − m̈ ẋ2 − 2(b + kv0 ke − ṁ)ẋẍ (24)
2 2
is also upper bounded by assuring that (2ḃ−m̈) is bounded,
and it proves that the tracking error converges by resorting
to Barbalat’s lemma. The above stability analysis can be
readily applied to (6). The stability of the proposed force
tracking control scheme is thus proved.

4. SIMULATION STUDIES

The proposed force tracking impedance control presented

in Sections 2 and 3 is simulated using a dual-arm redun-
dant manipulator. The dual-arm redundant manipulator
was developed at the Institute of Automatic Control En-
gineering (LSR) of the Technische Universität München.
The physical construction and a schematic view with co-
ordinate systems are shown in Fig. 4. Each arm consists
of two spherical joints with 3 DoF at shoulder and wrist, Fig. 6. Position tracking (left) and response of contact
each, and one revolute joint at the elbow, which results in force (right); (xe , xd ) = (0.4, 0.42) m, ke = 104 N/m

6754
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

Fig. 7. Position tracking (left) and response of contact Fig. 9. Position tracking (left) and response of contact
force (right); (xe , xd ) = (0.4, 0.401) m, ke = 104 N/m force (right); xd = 0.44 m, ke = 104 N/m
lates dynamic response of the EEF corresponding to the not placed deep enough into xe , even reaching xd can
contact force. Therefore, it is intrinsically inappropriate for generate contact force f which is smaller than fd . Since
force tracking control and the force tracking error in this the positive target stiffness can only modify the desired
simulation is greater than 30 N. When the environment trajectory in the same direction with the contact force f
stiffness ke is exactly known, although it is practically reducing the penetration depth xc − xe , the desired force
unrealistic, the difference between the desired and actual can not be achieved. In this case, the negative stiffness
contact forces in Fig. 5 can be compensated by replanning modifies the desired trajectory to opposite direction of the
the desired trajectory xd . contact force such that the compliant trajectory xc will
penetrate further into xe . Thus, the desired contact force
Force Control with Constant xe and ke fd can be achieved. Fig. 7 illustrates such kind of force
In this simulation, the proposed scheme guarantees po- tracking performance when the desired trajectory is set to
sition tracking in unconstrained (x-) direction while the xd = 0.401 m which is slightly greater than xe = 0.4 m.
desired force is achieved in constrained (y-) direction as In this case, reaching xd can only generate contact force
illustrated in Fig. 6. At the initial configuration, the EEF small than the desired one (f < fd ). However, negligible
makes a just contact (f = 0) with a flat wall placed at force tracking errors could be obtained by the proposed
xe = 0.4 m. The force tracking control gains kf and kv are scheme with negative stiffness and the force tracking error
set to 103 and 26, respectively, and the target impedance is identical to that of the previous simulation. Further, it is
parameters are selected as M = I, B = 102 · I, and observed that the impact force during the transient phase
K0 = 103 · I. In this simulation, the force tracking error at from the free space motion to the constrained motion is
the steady state ef,ss is identical to the force tracking error much smaller than that of the previous simulation. In this
which is calculated by (13): ef,ss = k0 kffd+1 < 5 × 10−5 . simulation, the target impedance parameters are set to
Since it is far smaller than the measurement resolution M = I, B = 102 · I, and K0 = 103 · I, and the gains
of conventional load cells, it can be practically treated as kf = 103 and kv = 26 for the force tracking controller.
zero.
Variable Environmental Stiffness
Force Control with Negative Stiffness In this simulation, the environment stiffness ke changes
As discussed in Section 3, if the desired trajectory xd is abruptly from 2×104 N/m to 104 N/m at t = 5 s. Since the

Fig. 8. Position tracking (left) and response of contact Fig. 10. Position tracking (left) and response of contact
force (right); (xe , xd ) = (0.4, 0.41) m force (right); xd = 0.41 m, ke = 104 N/m

6755
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

framework imitating human force tracking capabilities.

The impedance control scheme adapts its target stiffness
according to the previous force tracking error resulting
in a second order linear time varying system. The pro-
posed control scheme utilizes even negative stiffness to
achieve force tracking control. Stability analysis is given
for more general second order linear time varying systems
where variable target inertia, damping, and stiffness are
employed. Extensive simulations have been conducted for
various situations including uneven environment surfaces
and abrupt changes of environment stiffness. Simulation
results prove the force tracking capability of the proposed
control scheme for various situations without knowing the
environment stiffness.

REFERENCES
Fig. 11. Position tracking (left) and response of contact
force (right); xd = 0.42 m M. Raibert and J. Craig. Hybrid Position/Force Control
of Manipulators. ASME J. of Dyn. Sys. Meas. Control,
proposed force tracking scheme adapts the target stiffness 102(2), 126-133, 1981.
only based on the previous force tracking error, the same O. Khatib. A Unified Approach for Motion and Force
steady state force tracking error is achieved for both cases Control of Robotic Manipulators. IEEE J. Robot.
with different environment stiffness, see Fig. 8. The target Automat, 3(1), 43-53, 1987.
impedance parameters are selected as M = I, B = 102 · I, N. Hogan. Impedance Control: An Approach to Manip-
and K0 = 103 · I, and force tracking gains as kf = 103 and ulator. ASME J. of Dyn. Sys. Meas. Control, 107(1),
kv = 26. Due to the change of the environment stiffness 1-24, 1985.
ke , the contact force f is perturbed at t = 5 s but it has N. Hogan. On the Stability of Manipulators Performing
settled to the desired contact force in a very short time. Contact Tasks. IEEE J. of Robot. Automat., 4(6), 677-
This simulation results show the proposed force tracking 686, 1988.
control is robust to sudden changes of the environmental T. Lasky and T. Hsia. On Force-Tracking Impedance
stiffness while guaranteeing negligible force tracking errors. Control of Robot Manipulators. Proc. IEEE Int. Conf.
Variable Environmental Geometry Robot. and Automat, Los Alamitos, CA, 274-280, 1991.
In practical cases, the wall is not flat, hence the force H.-C. Cho, J.-H. Park, K.-H. Kim, and J.-O. Park. Sliding-
tracking control should cope with surface variation of Mode-Based Impedance Controller for Bilateral Tele-
the environment. In this simulation, the proposed force operation under Varying Time-Delay. Proc. IEEE Int.
tracking control scheme is applied to two different uneven Conf. Robot. and Automat, Seoul, Korea, 21-26, 2001.
environment surfaces which have a triangular type of M. Iwasaki, N. Tsujiuchi, and T. Koizumi. Adaptive force
indent and burr. The results of this simulation are shown control for unknown Environment using sliding mode
in Fig. 9 for the triangular type of indent, and Fig. 10 controller with variable hyperplane. JSME Int. J. Series
for the burr type of environment surface, respectively. C, Mech. Syst. machine elements and manufacturing,
For both cases, it is indicated that the proposed scheme 46(3), 967-972, 2003.
maintains the desired contact force with negligible force Chan. S. et al. Robust Impedance Control of Robot
tracking errors throughout the task and is thus robust to Manipulators. Int. J. of Robot. Automat., 6(4), CA,
variations in the geometry of the surface. Furthermore, it 220-227, 1991.
can be observed that the proposed force tracking control H. Seraji and R. Colbaugh. Force Tracking in Impedance
does not show any difference in steady state force tracking Control. Int. J. of Robotic Research, 16(1), 97-117, 1997.
errors between the flat part and inclined part of the wall. S. Jung, T.C. Hsia, and R.G. Bonitz Force Track-
ing Impedance Control of Robot Manipulators under
Variable Environmental Stiffness and Geometry Unknown Environment. IEEE. Trac. Control System
In this simulation, the proposed scheme is tested in the Techn., 12(3), 474-483, 2004.
presence of both abrupt change of the environment stiff- A. Hace, S. Uran, K. Jezernik, and B. Curk. Robust Sliding
ness from 2 × 104 N/m to 1 × 104 at t = 5 s and the trian- Mode Based Impedance Control. Proc. IEEE Int. Conf.
gular type of burr in the surface. The target impedance Intell. Engnrg. Sys., 77-82, 1997.
parameters are selected as M = I, B = 102 · I, and C. J. Harris and J. F. Miles. Stability of Linear Systems.
K0 = 103 · I, and force tracking gains as kf = 103 and New York, Academic Press, 1980.
kv = 26. Due to the overlap of the abrupt change of the W. J. Rugh. Linear System Theory. Upper Saddle River,
environment stiffness and geometry at t = 5 s, relatively New Jersey, Prentice Hall, 1996.
large impact force is observed, but it has settled in a very M. I. Gil. Stability of Linear Systems Governed by Second
short time to the desired contact force fd and negligible Order Vector Differential Equations. Int. Journal of
steady state force tracking errors are achieved. Control, 534-536, 78(7), 2005.
J. J. Slotine and W. Li. Applied Nonlinear Control.
5. CONCLUSION Englewood Cliffs, New Jersy, Prentice Hall, 1991.
In this paper, a simple and intuitive force tracking con-
trol scheme is presented within the impedance control

6756