深層強化学習を用いた節間協調の平面則に基づく2足ロボットの歩行運動生成

抄録

Recently, a deep reinforcement learning approach has been proposed as one of the control methods for legged robots. This paper describes walking motion of a bipedal robot using deep reinforcement learning. In this research, we focused on kinematic synergy in the legs and proposed reward of planar covariation of elevation angles at the thigh, shank and foot during human locomotion. By introducing the reward, the agent was trained, and as a result, the bipedal robot realized forward walking on the heels and off the toes. In addition, the proposed reward is considered to be one of the factors that characterize human walking motion by comparing it with conventional rewards.