深層強化学習を用いた移動ロボットの行動獲得法の改良

カテゴリ: 部門大会

論文No: PS6-12

グループ名: 【C】2022年電気学会電子・情報・システム部門大会

発行日: 2022/08/24

タイトル(英語): A Study on Improved Method for Behavior Acquistion of Mobile Robot by Deep Reinforcement Learning

著者名: 曽田 涼介（松江工業高等専門学校）,福島 英（出雲村田製作所）,堀内 匡（松江工業高等専門学校）

著者名(英語): Ryosuke Sota (National Institute of Technology, Matsue College),Akira Fukushima (Izumo Murata Manufacturing Co., Ltd.),Tadashi Horiuchi (National Institute of Technology, Matsue College)

キーワード: 深層強化学習|行動獲得|移動ロボット|視覚ベース|Deep Reinforcement Learning|Behavior Acquisition|Mobile Robot|Vision-based

要約(日本語): DQN (Deep Q-network) is one of the typical methods of deep reinforcement learning. DQN uses CNN (Convolutional Neural Network), which can extract features automatically from the input images. Rainbow is one of the advanced methods of DQN, which combines six improvement methods to DQN. However, we revealed that the learning performance of Rainbow method deteriorates in the latter half of learning. In this study, we proposed the improved method for the learning system to suppress the performance degradation during the learning. This improved method can be applied not only to Rainbow but also other DQN-based methods such as Dueling DDQN. Through the simulation experiments, we showed the stability of learning by the proposed method.