1987 Volume 5 Issue 2 Pages 109-120
It has been pointed out in much literature that when a robot performs some tasks like polishing or grinding, not only the position but also the contact force of the tools connected to the robot hand must be controlled. However, it is generally difficult for the robot to accomplish the desired position and force patterns. The main reason of this difficulty is that the reaction force and the friction force between the tools and the object which crucially affect the robot motion can not be estimated exactly, so that the input torque to generate the desired patterns can not be obtained.
To realize the desired position and force patterns without estimating the reaction and friction forces, a new control scheme based on the iteration of trials (test motions) is proposed. In this control scheme, at the first trial the robot is excited by an input torque pattern without knowledge of the robot dynamics, reaction force, and friction force. Therefore, the robot motion does not generally coincide with the desired motion patterns. Then, at the second trial the input is simply modified by the errors which are, the differences between the real force and position patterns and the desired ones. By repeating such operations under reasonable conditions, the force and position patterns approach the desired ones. In this paper, it is theoretically shown that the real force and position patterns of robots converge to the desired ones as the trial number tends to infinity. Moreover, the effectiveness of the proposed scheme is verified by two such experiments that a serial link robot manipulator with three degrees of freedom carries out two tasks (polishing and grinding) .
