1A1-E10 Motion Learning Using Direct Teaching Including Improper Instructions(Evolution and Learning for Robotics(1))

Abstract

Robots are expected to be able to cooperate with and assist us in our daily life. It is necessary for robots to perform a variety of tasks that were required by ordinary people who is not skilled at the control of a robot. In such a case, a direct teaching seems effective to program robot's motions. Since ordinary people may sometimes teach improper motions, it is necessary for a robot to learn appropriate motions autonomously using instructions including improper motions. To achieve this, we propose a control method inspired by a nonparametric model. In this method, a robot selects an action out of instructions where the fitness of each data is determined by a reinforcement learning algorithm. We also analyzed the properties of the trained controller.