Sei sulla pagina 1di 6

On Robust Impedance Force Control of Robot Manipulators

Seul Jung, T. C. Hsia and R. G. Bonitzy


Robotics Research Laboratory
Department of Electrical and Computer Engineering
University of California, Davis
Davis, CA 95616

Abstract ing force tracking errors have been proposed to esti-


In this paper, a new impedance function is proposed
mate environment sti ness or adjust controller gains
to achieve accurate force tracking under the presence
to compensate for unknown environment sti ness [5].
of uncertainties in robot dynamics and environment
A simple trajectory modi cation scheme has been pro-
models. The new impedance function is formulated
posed using the robot controller to compensate for
on the basis of PID control of the force tracking er-
robot dynamics uncertainties and environment sti -
ror which compensates for the unknown environment
ness is replaced by contact force information [6].
sti ness and position. The robot dynamics uncertain-
The purpose of this paper is to provide a simple
ties are compensated by a simple time-delayed robust
practical solution to force tracking impedance control
control algorithm. Stability and convergence of the
that is robust with respect to uncertainties in both
control scheme are analyzed. Simulation studies with
robot dynamic model and environment position and
three link rotary robot manipulator are shown. Fur-
sti ness. The main idea is to set the sti ness gain in
thermore, experimental results on a PUMA 560 arm
the impedance function to zero in force controlled di-
are carried out to con rm the proposed impedance con-
rection as soon as contact to the environment is made,
troller's performance.
and the nominal environment position is chosen in
such a way that the robot will be assured to make
contact with the environment from any position in
1 Introduction free space. The force tracking error is compensated
by an integral control. This results in an impedance
After the pioneering work of impedance control by function having a PID control structure for the force
Hogan [1] there have been two major research issues to tracking error. Environment sti ness knowledge is not
be solved on the impedance control: one is to give the required by the algorithm. Stability analysis of the
force tracking capability to impedance control and an- controller is presented.
other is to compensate for the uncertainties occurred Simulation studies of a three link rotary robot ma-
in both robot dynamics and the environment(position nipulator are carried out to demonstrate the robust-
and sti ness). ness under uncertainties in robot dynamics, environ-
Within this impedance force control framework, ment position, and environment sti ness. Experimen-
many control algorithms have been proposed. Lasky tal results using a PUMA 560 robot arm are presented
and Hsia have proposed the inner/outer loop control to con rm the performance of the control scheme.
scheme where the robot dynamics uncertainties are
compensated by a robust position control algorithm
in the inner loop [2] and estimated environment posi-
2 The Proposed Force Control
tion is modi ed using an integral control of the force Scheme
tracking error in the outer loop [3]. The generalized
impedance control of considering a general dynamic 2.1 Control law
relation between a position error and a force error to We start from the ideal impedance function as follows:
deal with the unknown environment sti ness has been
proposed [4]. The adaptive control techniques of us- Fe = M E + B E_ + KE (1)
 The author is currently at AHMCT in California. where E = Xr ? X and Xr is a 3 x 1 vector of reference
y The author is currently at JPL in California. position (Orientation control is not considered). For
simplicity, we consider that force is applied to only one the uncertainty of xe. De ne e0 = e+xe , then e0 will
direction of x that is Xr = [xe xd1 xd2 ]T where xe is replace e in both (4) and (5) yielding
an environment position and xdi are desired trajecto-
ries in position control directions. The corresponding me0 + b_e0 + ke0 = ?fd (free space) (7)
force is Fe = [fe 0 0]T :
Subtracting the desired force Fd from Fe yields the me0 + b_e0 = fe ? fd (contact space) (8)
generalized impedance function For free space motion (7), our previous analysis has
shown that when xe < 0 the robot manipulator may
Fe ? Fd = M E + B E_ + KE (2) not always make contact with the environment [7].
Let fd ; fe ; m; b; k be elements of Fd ; Fe; M; B; K re- Therefore, we should always overestimate the envi-
spectively in the force control direction xe. Then, ronment position x0e such that xe > 0 to guaranteed
equation (2) becomes that contact will always occur.
For contact space control (8), we can easily show
me + b_e + ke ? fe + fd = 0 (3) that exact force tracking can be achieved if the envi-
ronment is a at surface where xe is a constant and
where e = xe ? x. x0e is also a constant satisfying x0e ? xe > 0. In this
In free space control, the control law for the force case (8) becomes
control direction can be obtained from (3) for fe = 0 mx + bx_ = fd ? fe (9)
as follows:
me + b_e + ke = ?fd (4) Hence fe = fd in steady state. However when fd ; xe
We note that if the desired force fd is set to zero in and xe are time varying, force tracking error will oc-
(4) when the environment position is known exactly, cur. For example, there will be nite steady state force
then the robot would stop at x = xe which would just tracking error when fd and/or xe are ramp functions,
make contact with the environment. The presence of and the steady state force error would become in nite
fd in (4) serves as the driving force to enable the robot if fd and/or xe are parabolic functions. To improve
to exert a force on the environment. the performance of (8), we propose the inclusion of an
In contact space, careful investigation of the integral control as follows:
impedance function (3) suggests that setting the sti - Z
ness gain k to zero will satisfy the ideal steady state me0 + b_e0 +  (fd ? fe )dt = fe ? fd (10)
condition of fe = fd for any ke. So the proposed law
is to set the sti ness gain k = 0 in the force control where  is the integral gain.
direction at the time of contact, that is (3) becomes
me + b_e ? fe + fd = 0 (5) 2.3 Stability and Convergence of the
integral control law
Substituting fe = ke(x ? xe ) = ?ke e in (5) yields the Here, we like to reformulate (10) on the basis of force
contact space impedance law : tracking error. From the relationship fe = ke(x ? xe ),
me + b_e + kee = ?fd (6) we can express x = xe + kfee ; x_ = x_ e + kf_ee , and x =
xe + kfee .
We see that (6) is asymptotically stable. Even though Substituting x; x;_ x into (10) yields the second order
ke is not known exactly in practice, the proper gains m force error equation as
and b based on the approximation of ke can be chosen
to achieve good transient response of (6). Therefore,
Z m(fd ? fe ) + b(f_d ? f_e ) + ke (fd ? fe )
our proposed impedance function (5) is simple, stable
and robust in force tracking under unknown environ- + (fd ? fe )dt = mfd + bf_d ? mke xe ? bke x_ e (11)
ment sti ness condition.
De ne force tracking error  = fd ? fe and rewrite (11)
2.2 Inaccurate environment position as
Z
Consider for the practical case that we do not have m + b_ + ke +  dt
the exact knowledge of the environment position xe ,
but we do have an estimate x0e and xe = x0e ? xe is = mfd + bf_d ? mke xe ? bke x_ e (12)

2
De ne the F as the force due to uncertainty of the where V = [vf vp1 vp2]T and
environment position xe as fe = kexe . Then (12)  0 1
becomes vf = xe + m (fd + b_e0 + ke 0
R ) free (21)
Z 0 0
xe + m ( + b_e +  dt) contact
1

m + b_ + ke +  dt = m" + b"_ (13) vpi = xdi + m1 (b(x _ di ? xpi ) + k(xdi ? xpi ) (22)
where " = fd ? fe . In (20), w is function of the robot model which is
Laplace transform of (13) and multiply by s yields not only complex to compute but is also subjected to
various model uncertainties. In this paper, we apply
(s) = ms3 + bs2 (14) a previously well established technique to estimate w
E (s) ms3 + bs2 + kes +  using the time delayed measurements of  and q [2, 8].
The stability of the system (14) by examining the We rst express w as
characteristic equation of (14).  = (t) ? D
w(t)  q(t) (23)
ms3 + bs2 + kes +  = 0 (15) Then we estimate w(t) by
For stability, m; b;  should be  q(t ? )
^ = (t ? ) ? D
w(t) (24)
0 < m; b and 0 <  < bkme (16) where  is the sampling period of the controller. We
see that w(t)
^ is independent of the robot model. This
so when the input E is a step or a ramp the output yields the following complete robust force/position
(t) goes to zero in steady state so that robot controller
limt!1 (t) = 0 (17) (t) = J ?1[V ? J_q]_ + (t ? ) ? D
^ q(t ? ) (25)

Therefore, when t ! 1, fe ! fd . The stability of controller (25) has been analyzed ex-
tensively. It has been shown that the necessary con-
dition for stability is that should be chosen within
3 Compensation for Robot the bounds 0 < < 2. Experimental results have
veri ed the validity of the design technique [6, 8].
Model Uncertainty
The dynamic equation of an n degrees-of-freedom ma- 4 Simulations
nipulator is given by :
The nominal system parameters are used as the basis
Dq + h + f =  ? e (18) in forming the robot model D(q)^ and ^h(q; q).
_ Model
uncertainties included a 10 Kg mechanical tool at-
where D is an nxn inertia matrix, h is an n x 1 Cori- tached to the third link, Coulomb friction and vis-
olis ,centrifugal torque and gravity, f is an n x 1 cous friction forces f (q)_ added to each joint where
joint friction torque,  is the joint torque, and e is f (q)_ = 5:0sgn(q)_ + 8:0q._
the external applied joint torque. For the purpose of In order to show the robustness to unknown envi-
the subsequent presentation, we will write the robot ronment sti ness of the proposed scheme, we tested
dynamic equation (20) as the system performance for abruptly changing envi-
 q(t) + w(t) = (t) ronment sti nesses with the sti ness pro le as
D (19) 
40000(N=m) 0  t < 3
ke = 80000(N=m) (26)
where D = I and w = h(t) + f (t) + e (t) + (D ? 3t<6
 q(t). For stability, it is required that the scalar
D)
constant be bounded by 0 < < 2 where  is the which is considered as a sti environment.
minimum eigenvalue of D [2]. Sample task is tested for the robot to follow the
To implement the position and force tracking con- at wall with a large triangle shaped crack in the mid-
trollers presented above, the control law  is dle as shown in Figure 4. The initial controller gains
are selected as M = I; B = diag[440; 40; 440] and
 ?1(V ? J_q)_ + w
(t) = DJ (20) K = diag[300; 100; 300]. At contact, K at the force

3
controllable direction is set to zero. The environment Two experiments were conducted each consisting of
position is not correctly estimated but speci ed inside moving in the negative z direction onto a surface, and
the environment to guarantee the contact. Since the then moving 25cm along the surface in the x direction
deepest point of the crack is 0:02m from the surface in 15 seconds with fd = [0 0 ?20N 0 0 0]. The rst
the user speci ed xe is selected at 0:03m such that surface is a at hard surface and the second surface
the estimated environment position x0e = xe + 0:03m. is sti curved sheet of galvanized steel. The exact lo-
Since we assume that the environment pro le is not cation and sti ness of each surface is unknown. For
available, xe is constant and  x_ e ; xe are assumed the at hard surface,  was chosen as 0.001 and for
to be zero. In this case, the control behavior is con- the curved sti surface  was increased to 0.02 since
trolled by (9) and the plots are shown in Figure 2. the surface is more compliant. The results of the ex-
Even though the force tracking is excellent when the periments are shown in Figures 8 and 9. Initially, the
environment is at, the force tracking is not e ective end e ector is commanded to move to a point beyond
with notable errors when the robot follows the crack the surface. Upon contact with the surface, the initial
position. The deviation in force tracking in Figure 2 overshoot is approximately 60% and the force settles
results from not specifying x_ e and xe in the control to -20N in approximately two seconds. During the
law. In most practical case, the environment is not motion the force is controlled to within 20%. The
assumed to be known so that we adopt the integral experiments demonstrate that the algorithm exhibits
control idea (10) to compensate for x_ e which is a ma- good force tracking in the presence of environmental
jor contribution term to the force deviation. The force uncertainty.
tracking plot using the control law is shown in Figure
3. The update rate is selected as  = 0:3 The corre-
sponding position tracking plot is shown in Figure 4. 6 Conclusions
is excellent.
Another simulation is carried out for the sinu- A simple force control scheme of robot manipulator
soidally shaped environment. The force and posi- is presented in this paper. The new impedance con-
tion tracking are shown in Figure 5 and 6, respec- troller realized by setting the sti ness gain to zero
tively. In this case, the update rate  is set to 0:2 performs very well under the condition of inaccurate
and the controller gains B = diag[600; 20; 20];K = estimation of the environment position and sti ness.
diag[500; 100; 100] are used in free space and K = In order to deal with unknown environment position
diag[0; 100; 100] is used in contact space. Again, the we employed an integral control that minimizes a force
tracking performance is excellent. tracking error. The conventional impedance equation
are newly formed as a second order force error equa-
tion. It allows us to analyze the stability of the pro-
5 Experimental Results posed control that show the bounds for the controller
parameters. Simulation / experimental results prove
To demonstrate the performance of the proposed al- that the proposed controller is excellent to achieve the
gorithm, experiments were conducted using the UCD best force/position tracking under the situation that
Robotics Research Laboratory (RRL) experimental the environment is not known.
testbed consisting of a PUMA 560 industrial manipu-
lator tted with a JR3 wrist force sensor.
Single arm experiments were rst conducted on
the PUMA manipulator operating in both free space
References
and in contact with a rigid surface using conventional [1] N. Hogan, \Impedance control : An approach to
impedance control which resulted in the impedance manipulator, part i, ii, iii", ASME Journal of Dy-
chosen as follows: M = diag f50 50 50 1 1 1g, namic Systems, Measurement, and Control, vol. 3,
B = diag f3142 3142 1571 31:4 31:4 31:4g, and pp. 1{24, 1985.
K = diag f31583 31583 11844 237 237 237g. The
inertia matrix was chosen to satisfy a lower bound [2] T.C. Hsia, \Simple robust schemes for cartesian
requirement which results from digital implementa- space control of robot manipulators", Interna-
tion during constrained operation. The preceding tional Journal of Robotics and Automation, pp.
impedance yields good trajectory tracking and force 167{174, 1994.
step response performance. Damping in the z direc-
tion was reduced to achieve faster response in the force [3] T. Lasky and T.C. Hsia, \On force-tracking
tracking direction. impedance control of robot manipulators", Proc.

4
X q
Forward
Kinematics

. .
X q
J
..
q
^
D s s
Desired force() and Actual force(___)
+

25
.. .
Xe(t) J Delay
_
E + V U h + q
Xe(t) M K
-1
J -1 ^
D Robot
+ + + + +
+ Fe 20
+
.
.
E -1 -1
Adaptive
JT
Xe(t) M B M e
+ law

Force (N)
15
+
Fd
Fe

Figure 1: Robust Force Control Structure 10

of the IEEE International Conference on Robotics 5

and Automation, pp. 274{280, 1991.


[4] Sukhan Lee and Hahk Sung Lee, \Intelligent con- 0
0 1 2 3 4 5 6

trol of manipulators interfacing with an uncertain Time (sec)

environment based on generalized impedance", Figure 2: Force Tracking for a Triangle Crack when
Proc. of IEEE Symposium on Intelligent Control, x_ e = 0
pp. 61{66, 1991.
[5] H. Seraji and R. Colbaugh, \Force tracking in
impedance control", Proc. of the IEEE Interna-
tional Conference on Robotics and Automation,
pp. 499{506, 1993.
[6] S. Jung, T. C. Hsia, and R. G. Bonitz, \On
force tracking impedance control with unknown
environment sti ness", Proc. of IASTED Interna- Desired Force() and Actual Force(___)

tional Conference on Robotics and Manufacturing, 25

pp. 181{184, Cancun, June, 1995.


[7] S. Jung and T. C. Hsia, \Neural network tech- 20

niques for robust force control of robot manipu-


lators", Proc. of IEEE Symposium on Intelligent
Control, pp. 111{116, Monterey, August, 1995.
15
Force (N)

[8] R.G. Bonitz and T.C. Hsia, \Robust internal-force contact

based impedance control for coordinating manip- 10

ulators - theory and experiments", IEEE Inter-


national Conference on Robotics and Automation, 5
1996.
0
0 1 2 3 4 5 6
Time (sec)

Figure 3: Force Tracking for a Triangle Crack with


the integral control

5
Real() and estimated (..) environment, Actual position(___)
0.65

Real () and estimated (..) environment, Actual position (___)


0.9
0.6

x axis (m)
0.85

0.55

0.8
z axis (m)

0.75 0.5
0 1 2 3 4 5 6
Time (sec)

0.7
Figure 6: Position Tracking for a Sinusoidal Surface
Position z
100

0.65 80
0.45 0.5 0.55 0.6 0.65 0.7
x axis (m) 60
contact
mm

Figure 4: Position Tracking for a Flat Wall


40

20
estimated actual
0
0 5 10 15 20 25 30
Time (sec)
Force z
0

10 actual
Newtons

desired
20
motion
30
contact
40
0 5 10 15 20 25 30
Desired force () and Actual force (___) Time (sec)

20

18
Figure 7: Experiment - Flat Environment Surface
16
Position z
14
80
12 70
Force (N)

contact
60 actual
mm

10
50

8 40
30 estimated
6 20
0 5 10 15 20 25 30
Time (sec)
4
Force z
0
2

10 actual
0
0 1 2 3 4 5 6
Newtons

Time (sec) desired


20

Figure 5: Force Tracking for a Sinusoidal Surface


motion
30
contact
40
0 5 10 15 20 25 30
Time (sec)

Figure 8: Experiment - Curved Environment Surface

Potrebbero piacerti anche