2023 Fiscal Year Final Research Report
Realization of deep reinforcement learning in real environment and its application to swarm robots
| Project/Area Number |
19K12147
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61040:Soft computing-related
|
|---|
| Research Institution | Matsue National College of Technology |
|---|
Principal Investigator |
HORIUCHI Tadashi 松江工業高等専門学校, 電子制御工学科, 教授 (50294129)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
青代 敏行 東京都立産業技術高等専門学校, ものづくり工学科, 准教授 (40571849)
|
|---|
| Project Period (FY) |
2019-04-01 – 2024-03-31
|
| Keywords | 深層強化学習 / 知能ロボティクス / 群ロボット / 行動獲得 |
|---|
| Outline of Final Research Achievements |
In this study, we realized deep reinforcement learning in real environment and applied it to acquiring cooperative behavior of swarm robots. First, we investigated the method to reduce the difference between the simulation environment and the real robot environment. Using this method, we achieved behavior acquisition based on visual information for a single mobile robot using deep reinforcement learning. Next, assuming the environment in which multiple robots exist, we achieved the acquisition of cooperative behavior for real swarm robots. More concretely, we realized to acquire following and overtaking behaviors as cooperative behaviors in swarm robot environment by deep reinforcement learning to enable robot. Finally, in order to improve the explainability of robot action selection using deep reinforcement learning, we confirmed the effectiveness of visualizing the attention area, that indicates which area in the camera image the robot focused on when selecting an action.
|
| Free Research Field |
知能システム
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究は，人工知能の中核である機械学習の技術を実機ロボットに応用したものである．深層学習と強化学習を組み合わせた深層強化学習という手法を用いて，実機ロボットが視覚情報に基づいて行動を自分自身で学習することを実現した．また，複数のロボットが存在する環境を想定し，実機の群ロボットを対象とした協調行動の獲得を実現した．具体的には，群ロボットの環境において，追従行動および追い抜き行動の獲得を実現し，深層強化学習による協調行動の学習が可能であることを明らかにした．これらの成果は，物流倉庫や福祉施設などにおける搬送ロボット群の行動制御などにつながると期待できる．
|
