Sie sind auf Seite 1von 5

Auromarico. Vol. 32, No. 3. pp.

623-447, 1996

Pergamon ooos-1098(95)00173-5
Copyright 0 1996 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
ms-10981% $15.00 +a.00

Brief Paper

A Passivity-based Approach to Force Regulation and


Motion Control of Robot Manipulators
BRUNO SICILIANOf and LUIGI VILLANIt

Key Word-Robots; force control; position control. stability; adaptive control.

Abstract-This paper deals with the design of control by Chiaverini and Sciavicco (1993). Recently a parallel
schemes for a robot manipulator in contact with a compliant regulator with gravity compensation has been proposed
surface. A passivity-based approach is adopted in the task (Chiaverini and Siciliano. 1991: Chiaverini ef al.. 1994) that
space, where the control law contains a nonlinear guarantees that the force error is driven to zero at the
model-based term and a linear term obtained as the sum of a expense of a steady-state position error along the constrained
position action and a force action. The force control action task direction. An adaptive version with respect to gravity
dominates the position control action along the constrained parameters has also been developed (Sicilian0 and Villani,
task space direction so as to achieve force regulation at the 1993a, b).
expense of a steady-state position error. Motion control This paper is devoted to derive parallel force/position
along the unconstrained task space directions is ensured. In control schemes for a robot manipulator in contact with a
the case of imperfect model compensation, the scheme is planar contact surface in a passivity framework. In
made adaptive with respect to a set of dynamic parameters. comparison with inverse dynamics controllers, passivity-
Numerical case studies are developed for an industrial robot based controllers are expected to have enhanced robustness,
manipulator. since they do not rely on the exact cancellation of nonlinear
terms (Anderson, 1989). As in typical passivity-based control
1. Introduction schemes for robot manipulators, the resulting control law is
Control strategies for a robot manipulator in contact with an composed of a nonlinear model-based term and a linear
environment can be grouped in two categories: those compensator action. A functional expression of the reference
performing open-loop force control and those performing vector is established to be used in the controller, and is
direct closed-loop force control. To the first category belong related to both the end-effector position error and the
the impedance control schemes (Hogan, 1985; Kazerooni et contact force error. This is conceptually different from
nl., 1986) that achieve indirect force control by means of previous passivity-based hybrid force/position controllers
closed-loop position control; the position error is related to (Slotine and Li, 1987; Lozano and Brogliato, 1992) where
the contact force through a mechanical impedance of each component of the reference vector is related either to a
adjustable parameters. The most common strategy belonging position error or to a force error in respect of the task space
to the second category is the hybrid position/force control selection mechanism.
(Raibert and Craig, 1981; Khatib, 1987; Yoshikawa, 1987), The force control action of the new scheme is formulated
where either a position or a force is controlled along each with an inherent integral term, and the class of position and
task-space direction. The inclusion of an integral action in force linear compensators is found that preserve dominance
the force control loop guarantees removal of the steady-state of the force control action. It is demonstrated that the
force error (Wen and Murphy, 1991). It has been verified control scheme ensures regulation of the desired contact
experimentally that the use of an integral action may provide force along the constrained direction, with tracking of the
robustness with respect to force measurement delays (Volpe unconstrained components of the desired end-effector
and Khosla, 1993; Wilfinger et al., 1994). trajectory. If uncertainty occurs on the dynamic model
Two alternative strategies still in the second category are parameters, the scheme is made adaptive by introducing a
the inner-outer position/force control (De Schutter and Van conventional parameter estimate update law, and the
Brussel, 1988), where an external force control loop is closed resulting controller achieves the same goals as in the
around the internal position control loop, and the parallel known-parameter case.
force/position control (Chiaverini and Sciavicco, 1988), An industrial robot manipulator is considered in order to
where both a position and a force are controlled along each develop numerical examples aimed at analyzing the
task space direction with dominance of the force action performance of the proposed schemes both in the known
above the position one. The parallel control strategy is parameter case and when an unknown payload mass is
suitable to manage the interaction with scarcely structured considered.
environments, which are known to represent a drawback for
hybrid controllers. The performance of a parallel controller 2. Modelling
with inverse dynamics compensation in the case of contact The dynamics of a rigid robot manipulator in contact with
with an elastically compliant frictionless surface was studied the environment is described in the task space (Khatib, 1987)
by the equation
*Received 9 September 1994; revised 18 March 1995;
revised 16 August 1995; received in final form 19 September B,(x)ii + C,(x, i)k + d,(x, i) + G(X) = II - f, (1)
1995. This paper was not presented at any IFAC meeting.
This paper was recommended for publication in revised form where x is the m x 1 vector of task variables (usually the
by Editor C. C. Hang. Corresponding author Professor end-effector location), B, is the m X m symmetric inertia
Bruno Sicihano. Tel. +39 81768 3179; Fax +39 81768 3186; matrix, C,i is the m X 1 vector of Coriolis and centrifugal
E-mail siciliano@na.infn.it. generalized forces, d,(x, ir) is the m X 1 vector of forces
t Dipartimento di Informatica e Sistemistica, Universita generated by joint friction, & is the M X 1 vector of
degh Studi di Napoli Federico II, Via Claudio 21, 80125 gravitational generalized forces, II is the m X 1 vector of
Napoli, Italy. driving generalized forces, and f is the m X 1 vector of

443
444 Brief Papers

contact generalized forces exerted by the manipulator on the modifying (10) (Sicilian0 and Slotine. 1991) or at dynamic
environment; all task-space quantities are expressed in a level by modifying (9) (Khatib, 1987).
common reference frame. For the friction forces. the relationship between task space
In this work, the case of non-redundant non-singular and joint space is
manipulators is treated. Then the vector x represents a set of
Lagrangian generalized coordinates and the matrix B, is d,(x. ir) = J ‘(q)d(q, 4). (II)
positive-definite. Further, it can be shown that B,, B;’ and
It is supposed that d contains viscous and static friction (Yao
&- are bounded as functions of x. and C, is bounded as a
and Tomizuka. 1994). i.e.
function of x and i. For simplicity, the case M = 3 is taken,
i.e. only translational motion and linear force are considered. d(q. il) = WI + D,(q) sgn (4) (12)
Preliminary results in the case of m = 6 are reported in
Gravdahl et al. (1994). where D, and D, arc positive-definite (or positive-
The environment is thought of as a frictionless, elastically semidefinite) diagonal matrices. Then the assumption
compliant plane. Hence the model of the contact force takes eTd,(x. i, e) 20 in Lemma 1 holds. In fact. using (6). (10)
on the simple form and (I 2) gives
f = K(x - x,x). (2) e’&(x. ir, e) = (4 ~ a)‘D,(i ~ u)
where x is the position of the contact point, x,, is a point of
the plane at rest, and K is the 3 X 3 constant symmetric + (i ~ a)D.(q)Isgn (4) - sgn (a)1
stiffness matrix. An expression for K is 2 0. (13)
K = km I. (3)
4. Force regulation and motion control
where k > 0 is the stiffness coefficient and n is the unit vector
To obtain a force/position controller, the error vector e
orthogonal to the contact plane, which are assumed to be
should be properly related to the force and position errors.
both constants. It is assumed that contact with the plane is
As in Lemma 1. the goal is to express the force and position
not lost.
errors as the outputs of an exponentially stable, strictly
proper filter whose input is the error vector e E -it? fl Y?
3. Passivity-based approach
(see the fundamental input/output theorem on p. 59 of
Consider the control law
Desoer and Vidyasagar, 1975).
II = h(x)c + kJx, i)r + i,(x, r) + C(x) The parallel control strategy (Chiaverini and Sciavicco,
1988) suggests relating the error vector in (6) to both a
- K,,(i - r) + f. (4) position and a force error. Along the constrained task
where 4, el, & and & are the estimates of B,, C,, d, and directions. the conflict between the position and force actions
& respectively, f is the measured contact force. r is a 3 X 1 must he managed by imposing dominance of the force action
reference vector and KD i> a symmetric. positive-definite gain above the position one.
yatrix. The matrix C, in (4) is chosen so that Let xd denote the time-varying desired end-effector
position. with x4, id,?& E _%‘Z. Let fd denote the constant
B,(x) - 2&x, ir) 6 skew-ymmetric (Slotine and Li, 1987).
desired force. Then indicate by Ax = x - xd the error between
Assuming that B,, C,, d, and & have the same functional
the actual and desired end-effector position, and by
forms as B,, C,, d, and &, the control law (4) can be written
Af = f - fd the error between the actual and desired contact
as
force.
u = Y,(x, i. r, i)6 - K&i - r) + f (5) An appropriate choice for the error vector in (6) is

where the well-known property of linearity of the dynamic e=[sltK,(s)]Ax+s ‘K&)Af (14)
model (1) with respect to the p X 1 parameter vector 8 has
been exploited (Niemeyer and Slotine. 1991). where .v is the Laplace variable, and K,(s) and K,=(S) denote
Setting the transfer functions of the position and force compensators
to be designed. It is worth pointing out that, in view of (6)
r=i-e. (6)
and (7). K,(s) and Kf(s) in (14) must have relative degree
i=ji-& (7) greater than or equal to 0 for the implementation of the
control law (5) with only measurements of position, velocity
and combining (1) with (5) gives and force.
B,(x)6 + C,(x, i)e + ix(x. i. e) + K,e = Y,(.)t? The first term on the right-hand side of (14) is the same as
in Slotine and Li (1987). The presence of an integral action in
where 2, = i, - d, and 4 = 4 - 0. the second term is motivated by the desire to achieve a null
steady-state force error at the expense of a finite steady-state
Lemma I. Consider _the system described by (8). If the position error along the constrained direction, on condition
mapping -e*YJ.)O is passive and e$(x,ir, e) ~0 then that a suitable choice of K,(s) and K,(s) is made. It is
e E 2: n 6pz. Further, if YJ.)B E .ipz, then e-t0 as t 4 =. assumed that K,(s) does not have poles at the origin.
By virtue of (2) and (3). the equation (14) can be
Proof This proof is conceptually similar to that of Theorem effectively decomposed into the component along n and the
3 by Ortega and Spong (1989). n components on the plane of contact. For this purpose, it is
advantageous that K.,(s) and K,(s) not rotate the vectors Ax
The control law (5) is expressed in the task space. The and Af respectively, i.e. K,(s) = k,(s)1 and K,(s) = k,(s)I. In
corresponding 3 X 1 vector T of joint actuating generalized this way, the analysis is reduced to the following three
forces can be computed as (one + two) equations:
,r= J’(q)u, (9) e,,=[s+k,(s)]h,,+s ‘k,(s)Aj; (l-5)
where q is the 3 X 1 vector of joint variables and J is the 3 li 3
e,, = [s + k,(s)] Ax,, -- s ‘kf(s)f,,, (16)
manipulator Jacobian matrix established by j, = J(q)cj. The
implementation of the control law (9) in terms of joint where, with obvious notation, (15) involves the normal
positions and velocities (Fossen. 1993) is based on the choice components of the force and position vectors, while (16)
involves the components on the plane. Notice that the
(r = J ‘(q)r (10)
contact model (2), (3) implies f,, = 0; hence it is reasonable to
for a non-singular Jacobian matrix. Note that in the case of a choose f,,, = 0. If no information about the plane geometry is
redundant manipulator, q is an n X 1 vector with n >3, and available. i.e. the direction of n is unknown, it is advisable to
redundancy can be solved either at kinematic level by choose f,, = 0.
Brief Papers 445

The position along the normal direction is given by Corollary. Consider the system (1) with the environment
model (2). (3), the control law (5) with parameter estimate
x,=k-‘Af,+k-‘fdn+x<,,,. (17) update law
It follows from (15) and (17) that & = -T-‘YT(.)e, (22)
Afn = k[g(s)e, + h(s)&, - k-‘fdn - dl> (18) where r is a p Xp symmetric, positive-definite matrix, and
where (6), (7) and (14). On the basis of the assumptions (Al)-(A4)
s of Theorem 1, the statements (Sl)-(S3) of the theorem hold.
g(r) = sz + k,(s)s + kk,(s) ’ (19) In addition 4, 4 E 2:.

4s + W)l The above results are based on a suitable choice of k,(s)


h(s) =sz + k,(s) + kk,(s)
(20)
and kf(s). The simplest choice is to take them as constants,
Further, (16) with f,, = 0 gives i.e. b(s) = h, and kf(s) = A2 (Sicilian0 and Villani, 1994)
with A,, A2> 0. This ensures that the transfer functions (19)
1 and (20) are always exponentially stable independently of k,
Axp =
S+ ep. although the actual value of the stiffness coefficient affects
the performance of the system. By virtue of this choice, from
Note that, since k,(s) and k,(s) have relative degree
(6) and (7), the reference vector I and its derivative become
greater than or equal to 0, both g(s) and l/[s + C(s)] have
relative degree 1, whereas h(s) is proper. In order to apply
the input/output theorem on p. 59 of Desoer and Vidyasagar r=ir-e=$-A,Ax-A2 Af do, (23)
I0
(1975), k,(s) and kf(s) must be chosen so that the transfer
functions g(s) and l/[s + k,(s)] are both exponentially stable. i=Z-B=jid-A,Ai-AzAf. (24)
In that case, from (20), h(s) is also exponentially stable.
On the other hand, the second term on the right-hand side Note that these expressions are similar in nature to the
of (18) plays the role of a disturbance on the force error reference trajectories proposed by Su et al. (1992) and
along the normal. Since h(s) possesses at least a zero at the Arimoto et al. (1992) which were derived in the joint space,
origin, the effect of a constant disturbance is rejected. Notice however. In contrast with those works, here no selection
that fs, and x0,,,are constants whereas .Q,, is not necessarily a between force-controlled and position-controlled com-
constant. ponents of I and i is accomplished by the control law (5).
The following analysis is aimed at demonstrating the Interestingly enough, the control law (5) with (23) and (24)
convergence features of the-closed-loop system. Initially, the can be regarded as an extension of the impedance controller
known-parameter case, i.e. 8 = 0, is analyzed. in Kelly ef al. (1989) in that a desired force fd different from
zero can be specified.
Theorem 1. Consider the system (l)&with the environment Another simple choice for the force action is k,(s) =
model (2), (3), the control law (5) (8=@) and (6), (7) and Az + A,/s, which determines an additional zero of h(s) at the
(14). Then, given the assumptions origin and thus ensures a null steady-state force error along
the normal in case of a constant .&. Nevertheless, the
(Al) xd, %d,%, o 2% presence of a double integrator on the force error may lead
to oscillatory behaviour in the force response. More complex
(A2) fd is a constant with bp = 0,
choices for k,(s) and k,(s) are feasible if suitable filtering
(A3) k,(s) and kf(s) have relative degree greater than or actions on the position and force errors are sought. In such
eaual to 0 and are chosen so that g(s) and l/js + k,(s)1 cases, more accurate estimates of the stiffness coefficient are
are both exponentially stable, - _ required to tune the coefficients of the compensators.
(A4) the poles of k,(s) and Sk,(s) have all negative real In summary, in both the known-parameter and unknown-
parts, parameter cases it is possible to design a passivity-based
control scheme (with adaptive law) that guarantees tracking
the following statements hold: of the end-effector position along the unconstrained
directions with regulation of the contact force along the
(Sl) Axp, A$, E 2: ~72% and AxP, A$ + 0 as t + “; constrained direction.
(S2) Afn E Z; moreover, if xdn is a constant then
5. Case studies
Af,E~~nn6P,andAf”jOast~m;
The proposed control scheme was tested in simulation on
(S3) all signals in the system remain bounded. the industrial robot COMAU SMART-36.12R. Only the first
three joints were considered, constituting an anthropomor-
Proof. In the known-parameter case, the right-hand side of phic manipulator geometry with zero shoulder offsets. The
(8) vanishes and the mapping -e-O is passive. Then, in complete dynamic model in the joint space and numerical
view of Lemma 1, e,, E Z’s n 22, e,, E Z$ n %, and e,, -+ 0 data of robot parameters were taken from Caccavale and
ast--+m. Chiacchio (1994). Both static and viscous joint friction was
For (Sl), the assumption (A3) and the result e, E Z$fl considered. A load was added at the end effector constituted
25, e,, -+O as t + m imply that AxP, A$ E 6p$ n 25 and by a cylinder of 12 kg mass and moments of inertia about
Axp, A$+0 as t+m. central axes of 0.0475. 0.0475. 0.015 kg m*. Simulations were
For (S2), the assumption (A3) and the result e, E Zz tl% run in MATLAB at a’sampling time of 2 ms.
imply that g(s)e, E YZ II Z and g(s)e,, + 0 as t+ m; also The geometry of the contact plane is characterized by
/I(s)(x~ - k-If& - x0,,) E 2’_ and tends exponentially to n = I1 0 OIT and x,, = I1 0 Ol* m. The control law (5) with
zero if xdn is a constant. It follows that Afn E E; moreover, if (23)’ and ‘(24) was &ed w-ith A, =2OOs-‘, A2=d.2 kg-‘,
xdn is a constant then Afn E L& n LK and Afn + 0 as t + cc. Ko = ko1 and k,, =400 kgs-‘. This choice guarantees
The proof of (S3) goes as follows. (A3), (17) and (18) and well-damped behaviour with an estimate of the stiffness
the result e,, E L& fl Z’= imply that x,, f E LL; then, in view coefficient of the plane k = lo5 N rn--‘.
of (Al) and (Sl), x, i E 2% and, from (6), r E 5% (A2) and In the following, simulation results relative to two different
(S2) imply ‘that f E 2% (Al) and (A4) imply that case studies are presented. For simplicity, a bidimensional
k.(s)
^\ Ax E 2% then. from (14). s-‘kJs)Af
I --,
E 2% From task geometry in the (x, z) plane was considered: thus just
(A3) and (18), s -I Aic 2%; then: in view‘df (Al) and (A4), the x and z components of the position vector and the x
Sk,(s) Ax, kf(s) Af E U?, and then i E U-l. From (8), 1 E 2% component of the force vector are of concern.
and then, via (7), f E 2%. 0 In the first caSe studv, the end effector is in contact with
the plane at x = xe. A straight-line path was commanded
In the unknown-parameter case (6#0), the following from si = [l 0 OITm to or = [I 0 0.151Tm; a trapezoidal
result holds. velocity profile was assigned with maximum velocity of

AUTO
32:3-J
446 Brief Papers

150.

Z
0 100.
2
8
50.

n
“0 0.5 1 1.5 -0 0.5 1 1.5
time [s] time [s]
Fig. 1. Time history of the contact force for the tint case Fig. 3. I’ime history of the contact force for the second case
study. study.

0.3 m s I and time duration of 1 s. A force set-point of Nevertheless. it is believed that it is worth testing the
fd = [50 0 O]'N was assigned. robustness of the proposed control scheme in simulation
Three sets of simulations were carried out using the against uncertainty on the environment.
following controllers: Figure 3 shows the time history of the force. The contact
occurs at I = 0.66s, with an end-effector velocity component
(a) control law (5). (23) and (24) with compensated load
of 0.2 ms-’ along the normal. It can be seen that the force
mass:
remains bounded during the contact, despite the imposed
(b) control law (5). (23) and (24) with non-compensated load time-varying position. Then the force goes to zero at steady
mass; state (.& = 0), thanks to the presence of the integral action.
Figure 4 shows the desired path (dashed line) and the actual
(c) control law (5), (23) and (24) with non-compensated load
path (solid line) of the end effector in the (x, Z) plane; note
mass and adaptive law (22).
that only the unconstrained component of the desired
The joint-space dynamic model was parameterized with position vector. i.e. the z component. is followed, whereas
respect to the parameters related to the load: namely, the the I component at steady state reaches that of the
mass, the three first-order moments and the six elements of environment at rest (fdn = 0).
the inertia tensor. The numerical results are illustrated in
terms of the time histories of the contact force and the ; 6. Cbnclusions
component of the position tracking error. A passivity-based force/position control scheme for robot
Figure I shows that the contact force reaches the desired manipulators in contact with an elastically compliant surface
value at steady state in the three cases thanks to the integral has been proposed. Inspired by the parallel force/position
action on the force error in the control law: however, a delay control framework, the reference vector used in the control
occurs for cases (b) and (c) in view of the initial parameter law has been related to both the end-effector position and
mismatch. Figure 2 shows how the adaptation mechanism (c) velocity errors and the integral of the contact force error.
allows recovery of the tracking position error. which affects Both theory and simulation results have demonstrated that
the behaviour of the control law (b). and the performance the scheme ensures force regulation along the constrained
compares favourably with that of the controller with perfect direction and motion control along the unconstrained
compensation (a). directions.
In the second case study, the end effector is at position By virtue of the passivity formulation, the scheme has been
x = xd, = [0.9 0 01“ m. A straight-line path was commanded naturally extended to handle adaptation with respect to a set
from xd, to xdf = Il.03 0 0.05] m; a trapezoidal velocity of manipulator and load parameters. In particular, the
profile was assigned with maximum velocity of 0.28 m s ’ and Gmulation results have demonstrated the effectiveness of the
time duration of 1 s. It is anticipated that a contact occurs at control scheme with adaptive law in the case of an unknown
a distance of 0.03 m from the target point xd along the .x-axis load mass. It is worth remarking that the advantageous
direction. A set-point fd = 0 was assigned to the contact feature of the proposed schemes consists in the simplicity of
force, because the orientation of the plane was assumed to be the control laws, which do not contain explicit information
unknown. on the environment; these control laws have been derived
It is appropriate to point out that the transition from under the assumption that contact is not lost. It should be
non-contact to contact deserves a rigorous theoretical clear, however, that the geometry and the mechanical
analysis that goes beyond the purpose of the present work. characteristics of the environment influence the performance

x 1o-3
0.06
II

I
0
0.5 1 1.5 0.9 0.95 1 1.C15
time [s] x-position [m]
Fig. 2. Time history of the ; component of the position error Fig. 4. Desired and actual paths of the end-effector position
for the first case study. in the (x, z) plane for the second case study.
Brief Papers 447

of the manipulator during the contact under such kind of Robotics Automation, RA-2, 83-105.
controllers. Kelly, R., R. Carelli, M. Amestegui and R. Ortega (1989).
It is understood that if a complete description of the Adaptive impedance control of robot manipulators.
environment is available then better performance may be IASTED Int. J. Robotics Automation, 4, 134-141.
obtained with control strategies that make use of selection Khatib, 0. (1987). A unified approach for motion and force
mechanisms to separate the time-varying directions to control of robot manipulators. IEEE J. Robotics
control in force -from those to control in position Automation, RA-3, 43-53.
(McClamroch and Wane. 1988: Mills and Goldenbere. 1989: Lozano, R. and B. Brogliato (1992). Adaptive hybrid
De Luca and Manes, &l; Arimoto et al., 1993; Yio and force-position control for redundant manipulators. IEEE
Tomizuka, 1993). With such strategies, it is also possible to Trans. Autom. Control, AC-37,1501-1505.
achieve force tracking in lieu of force regulation. McClamroch, N. H. and D. Wang (1988). Feedback
stabilization and tracking of constrained robots. ZEEE
Acknowledgement-This work was supported by Minister0 Trans. Autom. Control, AC-33,419-426.
dell’Universit8 e della Ricerca Scientitica e Tecnologica. Mills, J. K. and A. A. Goldenberg (1989). Force and position
control of manipulators during constrained motion tasks.
References IEEE Trans. Robotics Automation, RA-5, 30-46.
Anderson, R. J. (1989) Passive computed torque algorithms Niemeyer, G. and J.-J. E. Slotine (1991). Performance in
for robots. In Proc. 28th IEEE Co& on Decision and adaptive manipulator control. Znt. J. Robotics Res., 10,
Control, Tampa, FL, pp. 1638-1644. ’ 149-161.
Arimoto, S., Y. H. Liu, T. Naniwa, V. Parra-Vega and T. Ortega, R. and M. W. Spong (1989). Adaptive motion
Tsubouchi (1992). Model-based adaptive control for robot control of rigid robots: a tutorial. Automatica, 25, 877-888.
manipulators under geometric constraints. In Proc. 2nd Raibert, M. H. and J. J. Craig (1981). Hybrid position/force
International Symp. on Measurement and Control in control of manipulators. ASME J. Dyn. Syst.,
Robotics, Tsukuba, Japan, pp. 589-596. Measurement, Control, 103, 126-133.
Arimoto, S., Y. H. Liu and T. Naniwa (1993). Model-based Siciliano, B. and J.-J. E. Slotine (1991). A general framework
adaptive hybrid control for geometrically constrained for managing multiple tasks in highly redundant robotic
robots. In Proc. IEEE International Conf on Robotics and systems. In Proc. 5th International Conf on Advanced
Automation, Atlanta, GA, pp. 618-623. Robotics, Pisa, pp. 1211-1216.
Caccavale, F. and P. Chiacchio (1994). Identification of Siciliano, B. and L. Villani (1993a). Force/position regulation
dynamic parameters and feedforward control for a of robot manipulators with gravity parameter adaptation.
conventional industrial manipulator. Control Engng In Preprints 12rh IFAC World Conaress. - Vol. 7. DD.
LI
Practice, 2, 1039-1050. 403-4&, Sydney, AUS.
Chiaverini, S. and L. Sciavicco (1988). Force/position control Siciliano, B. and L. Villani (1993b). An adaptive
of manipulators in task space with dominance in force. In force/position regulator for robot manipulators. ht. J.
Proc. 2nd IFAC Symp. on Robot Control, Karlsruhe, pp. Adaptive Sig. Process., 7, 389-403.
137-143. Siciliano, B. and L. Villani (1994). A passivity-based
Chiaverini, S. and L. Sciavicco (1993). The parallel approach force/position control scheme for robot manipulators. In
to force/position control of robotic manipulators. IEEE Proc. IEEE International Conf on Robotics and
Trans. Robotics Automation, RA-9, 361-373. Automation, San Diego, CA, pp. 3265-3270.
Chiaverini, S. and B. Sicilian0 (1991). On the stability of a Slotine, J.-J. E. and W. Li (1987). Adaptive strategies in
force/position control scheme for robot manipulators. In constrained manipulation. In Proc. IEEE International
Proc. 3rd IFAC Symp. on Robot Control, Vienna, pp. Conf on Robotics and Automation, Raleigh, NC, pp.
183-188. 595-601.
Chiaverini, S., B. Sicilian0 and L. Villani (1994). Su, C.-Y., T.-P. Leung and Q.-J. Zhou (1992). Force/motion
Force/position regulation of compliant robot manipulators. control of constrained robots using sliding mode. IEEE
IEEE Trans. Autom. Control, AC-39,647-652. Trans. Autom. Control, AC-37,6681672. -
De Luca, A. and C. Manes (1991). Hybrid force-position Voloe. R. and P. Khosla (1993). A theoretical and
I \ ,

control for robots in contact with dynamic environments. experimental investigation of explicit force control
In Proc. 3rd IFAC Symp. on Robot Control, Vienna, pp. strategies for manipulators. IEEE Trans. Autom. Control,
377-382. AC-38,1634-1650.
De Schutter, J. and H. Van Brussel (1988). Compliant robot Wen, J. T. and S. Murphy (1991). Stability analysis of
motion II. A control approach based on external control position and force control for robot arms. IEEE Trans.
loops. Int. J. Robotics Res., 7(4), 18-33. Autom. Control, AC-Xi, 365-371.
Desoer, C. A. and M. Vidyasagar (1975). Feedback Systems: Wilfinger, L. S., J. T. Wen and S. H. Murphy (1994). Integral
Input-Output Properties, Academic Press, New York. force control with robustness enhancement. IEEE Control
Fossen, T. I. (1993). Comments on ‘Hamiltonian adaptive Syst. Mag., 14(l), 31-40.
control of spacecraft’. IEEE Trans. Autom. Control, Yao, B. and M. Tomizuka (1993). Adaptive control of robot
AC-Xl, 671-672. manipulators in constrained motion. In Proc. American
Gravdahl, J. T., 0. Egeland, B. Sicilian0 and S. Chiaverini Control Conf, San Francisco, CA, pp. 1128-1132.
(1994). Stability analysis of 6-dof force/position control for Yao. B. and M. Tomizuka (1994). Smooth robust adantive
robot manipulators. In Proc. 33rd IEEE Conf on Decision sliding mode control of robot manipulators ‘with
and Control, Lake Buena Vista, FL, pp. 2408-2409. guaranteed transient performance. In Proc. American
Hogan, N. (1985). Impedance control: an approach to Control Conf, Baltimore, MD, pp. 1176-1180.
manipulation, Parts I-III. ASME J. Dyn. Syst., Yoshikawa, T. (1987). Dynamic hybrid position/force control
Measurement, Control, 107, l-24. of robot manipulators-description of hand constraints and
Kazerooni, H., P. K. Houpt and T. B. Sheridan (1986). calculation of joint driving force. IEEE J. Robotics
Robust compliant motion for manipulators. IEEE J. Automation, RA-3, 386-392.

Das könnte Ihnen auch gefallen