Learning Control for Robot Manipulators under Geometric Endpoint Constraint. Convergence of Force Trajectory and Computer Simulation.

摘要

A learning control scheme for a class of robot manipulators whose endpoint is moving under geometrical constraints on a surface is proposed. In this scheme, the input torque is composed of two different input signals updated at every trial by different laws. One is updated by the angular velocity error vector which is projected to the tangent plane of the constraint surface. The other is updated by the force error at the manipulator endpoint. A theoretical proof of the convergence of force errors when velocity and position trajectories are in a neighborhood of the desired trajectories is presented. Computer simulation results by using a 3 DOF manipulator are presented to demonstrate the efficiency of the proposed method and the convergence of force trajectories besides position/velocity trajectories. A new type of learning that uses a feedforward torque input calculated by an approximate dynamics model of the manipulator is proposed to accelerate the speed of learning. The efficiency of the proposed method is illustrated by computer simulation.