Autonomous parameter identification by optimal learning control

Abstract

An autonomous control technique that allows a robot to generate its own sequence of optimal configurations during calibration is described. The algorithm attempts to tell the controller where to go in its configuration space so that the estimates of kinematic link parameters converge faster than for any other sequence of positions. It assumes that there is an external measuring device that can sense the position (but not necessarily the orientation) of a fixed point on the end effector. As the parameters are identified with a recursive estimation routine, a cost function that embodies the covariance matrix of the parameter error estimates is computed. The rate of decrease of this function is then maximized over all possible directions in the joint tangent space so that the next position chosen is automatically in the most exciting direction for the estimator. The technique was tested in a simple simulation and formulated for application to practical 6R robots.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>