Realization of High-Speed Imitation Learning for Cleaning Tools Operation

Abstract

The functions of intelligent machines and robot manipulators can be extended by attaching special tools or grippers to the end-effector, and they accurately execute various tasks based on operation commands. Here, the robot’s operation is instructed by a skilled robot operator at each step, mainly through controller operation. In order for the robot to learn to motion autonomously, the authors have proposed an algorithm in which the robot learns to imitate the use of unused tools from the information of human work movement [1]. Furthermore, it was shown that tools with relatively high similarity can be learned in a short time based on the use already learned, and executed with high accuracy. In this paper, sweeping movements as a realistic task in human daily life is aimed, and by verifying the combination of a wide variety of broom-shaped tools and objects to be collected, the adaptability of high-speed imitation learning ability by the algorithm was evaluated. The results show that the changes in the cleaning ability of the tools and objects to be handled and their tendencies are clarified. The effect of similarity on the speed of imitation learning are also confirmed. Regarding to the similarity, the validity of the proposed calculation method using multiple tools are verified.