確率的なゆらぎを有する強化学習を用いた大車輪ロボットの行動獲得と報酬の関係について

概要

論文の詳細を見る
We have succeeded in acquiring forward actions to various robot systems using Reinforcement Learning. We have also succeeded in acquiring a giant swing motion as dynamic task by devising its rewards. Although the giant swing robot has a continuous dynamic motion such as its angle and angler velocity, its state of the motion must be divided into discrete states in order to apply the reinforcement learning. Moreover, this giant swing robot system is not under Markov decision process by both control and defective sensation problem. For these reason, this robot shows probabilistic behavior. Then, this paper attempts to clarify the effect of probabilistic behavior of giant swing on the view point of various rewards, whose results are visualized using rotation rate. The results also show that features of the effect of probabilistic behavior are different for each reward.