Robot Programming by Demonstration with Oral Instructions for Assembly

Abstract

Programming by demonstration has been seen as a feasible solution for transferring human's skills to robots without too much time and labor cost. So far, applications of programming by demonstration in industrial assembly have attracted many researchers' attention. In practice, the robot control policy must give consideration to both efficiency and precision. Furthermore, it is difficult for one policy to handle the whole assembly process. To deal with these issues, a programming by demonstration with oral instructions method is developed in this article. With oral instructions, the demonstration data are segmented into pre-assembly phase and precise assembly phase. Moreover two related assembly policies are learned independently. Task-parametrized Gaussian mixture model and dynamic movement primitive are selected to prestructure the assembly policies for the two phases respectively on accounting of their properties. Effectiveness of the proposed method has been demonstrated by an assembly experiment.