Innovative multi-dimensional learning algorithm and experiment design for human-robot cooperation

Abstract

This paper proposed a learning algorithm for human-robot skill transfer based on dynamic movement primitive. Through the establishment of Transmission Control Protocol communication, the position trajectory and force characteristic of the human demonstrator while demonstrating the robot were recorded as the training data of the dynamic movement primitive model. The demonstrated skill/motion reproduction and generalization were successfully achieved through adjusting the parameters of the dynamic movement primitive model. The impedance control system investigated in this paper improved the compliance of the robot trajectory reproduction and skill generalization, and improved the robustness of robot control system in different environments. In addition, for the problem of inverse solution of redundant robots, this paper improved the dimension of constraint equations in the inverse solution of robots by training multiple dynamic movement primitive models, which guaranteed the constraints of inverse solutions of redundant robots. The proposed algorithm had been validated on a simulated robot platform using MATLAB and a ROKAE xMate robot platform, respectively.