Experimental verification of human skill transfer to deburring robots

Abstract

A new adaptive controller for deburring robots is presented in this paper. The design of the control system is entirely based on a human skill model. With the use of the human skill model, the performance skills of a human expert can be acquired and transferred to robot control systems that allow robots to mimic human skills in performing deburring tasks. The human skill model comprises a tool manipulation strategy, a process perception model and associative memories. The tool manipulation strategy describes the dynamic interaction between the tool, held by the human expert, and the environment. Teaching data taken from human demonstration motions shows that the human expert changes his tool manipulation strategy with respect to the varying process condition. The process model, which characterizes the process conditions, provides a way to detect variations in process conditions. The associative memories, represented by mapping functions, reflect how a human expert modifies his tool control strategy with respect to changing process conditions. These mapping functions can be identified by using human teaching data. The consistency of the mapping and the transferrability of human skills are analyzed by using Lipschitz's condition. A robot controller is constructed based on the human skill model that involves associative memories derived from human teaching data. The control system is implemented on a direct-drive robot. The experimental results show that the robot can adapt itself to the deburring process in a manner very similar to a human expert.