深層強化学習を用いた群ロボットの行動獲得に関する実験的考察

カテゴリ: 部門大会

論文No: TC11-8

グループ名: 【C】2021年電気学会電子・情報・システム部門大会

発行日: 2021/09/08

タイトル(英語): A Study on Behavior Acquistion in Multi-robots System by Deep Reinforcement Learning

著者名: 福島 英（松江工業高等専門学校）,綿貫 零真（奈良先端科学技術大学院大学）,堀内 匡（松江工業高等専門学校）

著者名(英語): Akira Fukushima (National Institute of Technology, Matsue College),Ryoma Watanuki (Nara Institute of Science and Technology),Tadashi Horiuchi (National Institute of Technology, Matsue College)

キーワード: 深層強化学習|行動獲得|群ロボット|視覚ベース|Deep Reinforcement Learning|Behavior Acquisition|Multi-robots System|Vision-based

要約(日本語): Deep Q-network (DQN) is one of the most famous methods of deep reinforcement learning. In this research, we apply DQN method to multi-robot environment and we aim to realize that multiple mobile robots acquire behavior to avoid walls and other robots. This task is more difficult in multi-robot environment than in a single robot environment, due to the in uence by the actions of other robots. We realized that each robot acquired behavior to avoid other robots based on images of the camera and values of distance sensors in the problem where all three robots move around in the same direction. Moreover, we considered that each robot acguired behavior to avoid other robots in the problem where one robot moves around in the different direction from other two robots.

PDFファイルサイズ: 1,035 Kバイト