2A2-C21 コーチングに基づくロボットのオンライン行動学習

概要

論文の詳細を見る
In this study, we propose a novel method of action learning of robot based on Coaching scheme. Coaching is a framework of human-robot interaction in order to utilize human subjective evaluation for training a robot. The proposed method integrates reinforcement learning and coaching scheme which enables the robot to achieve a given task based on both of human's evaluation and machine learning. We conducted several experiments and verified that the proposed method could improve the performance of conventional reinforcement learning. So that, robot could acquire the motion in a shorter time. This coaching scheme is suitable for the use in the real environment in which human and robot interact in a daily life.
一般社団法人日本機械学会の論文
2009-05-25