A trivial and efficient learning method for motion and force control

Abstract

The feedback produced by the linear controller of a manipulator executing a trajectory corresponds approximately to the inverse dynamics necessary to drive this trajectory. When a single movement is repeated, the feedback measured in one run can therefore be used as feedforward for the next runs. The learning position-force controller introduced in this paper is based on this idea and a parallel control of force and position. Experiments on 2- and 3-degree-of-freedom parallel manipulators show the simplicity of its implementation and its efficiency for gradually improving trajectory control in repeated movements. The control is robust to high level of noise and the performance is superior than with a parametric controller based on the rigid-body dynamic model. Simulations suggest that these properties also hold when the force is controlled simultaneously to the position. © 2002 Elsevier Science Ltd. All rights reserved.