Learning Control of Position and Force of a Direct-Drive Robot with a Variable Table Using Neural Networks.

Abstract

In this paper, a learning control algorithm using neural networks (NN's) for controlling the motion and force of a direct-drive (DD) robot with a variable table, which has six degrees of freedom, is proposed. The motion of an end effector attached on the robot arm and the internal force between the end effector and the table through an object can be obtained by the DD robot arm in cooperation with the variable table. The control of the motion and the internal force is achieved by hybrid controllers and a feedforward controller using NN's. After the completion of learning, the NN acquires a model of inverse dynamics for the DD robot arm and the DD table. Thus the outputs of the hybrid controllers are nearly equal to zero. The effectiveness of the proposed control algorithm is demonstrated by an experiment on the control of both the position of an end effector and the contact force on the constraint surface of the object.