3601 Q学習を用いたヤモリ型ロボットの歩行パターンの獲得(G15-1 ロボティクス・メカトロニクス(1) 歩行ロボット・アクロボット,21世紀地球環境革命の機械工学:人・マイクロナノ・エネルギー・環境)

概要

論文の詳細を見る
This paper attempts to make gecko-shaped robot voluntarily acquire the advance motion only by the interaction with environment; the desired trajectory, such as action patterns, is not preliminary given to the robot. In this paper, Q-learning algorithm, one of reinforcement learning algorithms, is applied to the gecko-shaped robot in order to learn the robotic gait patterns. This study, in particular, attempts to reveal the evolutionary process by observing the transition of motion forms appeared in the learning. The results show how a gecko-shaped robot, in which Q-Learning algorithm is implemented, acquires the advance motion. Furthermore, this paper deliberates the effect of robotic functions, such as a kinematic constraint and actuating performances, on the learned gait patterns.
2008-08-02