1A1-P05 Body structure Effects on Locomotion-pattern Acquisition of Six-legged Robot using Reinforcement Learning(Evolution and Learning for Robotics)

Abstract

This paper attempts to acquire locomotion patterns of locomotion robots by using Q-Learning method. Generally, primitive motion forms are achieved using basic motion rhythm by designers. However, in this study, locomotion robots voluntarily acquire locomotion patterns only by using information of interaction with environment. An advantage of this method is able to ensure that these motions are best for these robot mechanisms and environment they exist. In this paper, we apply reinforcement learning to some six-legged robots which have different shapes respectively in simulation. Then we compare each locomotion patterns and find an actual insect locomotion pattern by clarifying mechanical factors.