Adaptive Force and Position Control for Robots

NASA 's Jet Propulsion Laboratory, Pasadena, California

A control system causes the end effector of a robot manipulator to follow a prescribed trajectory and apply the desired force or torque to an object that it is manipulating or with which it is in contact. The system is characterized by a hybrid control architecture, in which the positions and orientations along unconstrained coordinate axes are controlled by a position control subsystem, while the forces and torques along the constrained coordinate axes are controlled by a force-control subsystem (see figure). The system compensates for the dynamic cross-coupling between the force- and position- control loops and does not require knowledge of the complicated model of the dynamics of the manipulator and its environment. In the force-control subsystem, the feedforward force vector Pr, (representing th e desired effector contact forces) is processed along with the measured (feedback)force vector P, an auxiliary signal d, and the velocity feedback
to produce the control vector Fz for the forces and torques applied to the end effector along th e constrained axes by the manipulator-joint actuators. The control law includes proportional, integral, and derivative gain matrices Kp, Kl, and Kv, respectively , that vary with time according to the equations of the adaptation scheme, which strives to reduce the error vector, ez, between the feedforward and feedback force vectors. The position-control subsystem enforces a linear adaptive control law that produces the manipulator force-and-torque control vector Fy for the unconstrained axes from the feedforward position vector R and from the feedback position and velocity Y and
, respectively. An auxiliary vector f and t he control gain matrices K, C, A, B, and Kd are calculated according to adaptation equations, the forms of which resemble those of the force controller. and which reduce the error vector ey. The force- and position-control signals are in Cartesian coordinates of the workspace and are transformed to equivalent joint-torque-control signals in manipulator joint space via the Jacobian matrix of the manipulator-joint coordinate system. The dynamic cross-coupling effects are mathematically modeled as disturbance terms in the force- and position-control loops. The adaptive force and position controllers have, in effect, learning capabilitie s to cope with unpredictable changes in the manipulator or the environment; this is because the controller gains are adapted rapidly on the basis of manipulator performance by use of simple arithmetic operations. The control scheme is computationally fast enough for use at sampling rates as high as 1 kHz.

Point of Contact:
Homayoun Seraji,
Mail Stop 198-219
Jet Propulsion Laboratory
4800 Oak Grove Drive
Pasadena, CA 91109
seraji@telerobotics.jpl.nasa.gov

Telerobotics Program page

Please email the site webmaster with any comments, criticisms or corrections for this page.
Maintained by: Dave Lavery
Last updated: May 10, 1996