2022 Fiscal Year Final Research Report
Development of AI for autonomous ship handling to accelerate ocean-bottom exploration, and its demonstration at actual sea
| Project/Area Number |
20H00284
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (A)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Review Section |
Medium-sized Section 24:Aerospace engineering, marine and maritime engineering, and related fields
|
|---|
| Research Institution | Osaka Metropolitan University (2022)
Osaka Prefecture University (2020-2021) |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
青木 尊之 東京工業大学, 学術国際情報センター, 教授 (00184036)
松田 秋彦 国立研究開発法人水産研究・教育機構, 水産技術研究所(神栖), 主幹研究員 (10344334)
島 伸和 神戸大学, 理学研究科, 教授 (30270862)
|
|---|
| Project Period (FY) |
2020-04-01 – 2023-03-31
|
| Keywords | 自律操船AI / 深層強化学習 / 逆強化学習 / 測線トラッキング / 定点保持 / 自動避航 / 自動離着岸 / 実船実験 |
|---|
| Outline of Final Research Achievements |
In this research, we developed AI for automating a series of ship maneuvers required for ocean-bottom exploration, such as line tracking, dynamic positioning, collision avoidance, docking and undocking. Deep reinforcement learning is used to acquire the autonomous control laws, so we designed state inputs, action outputs, and reward functions according to each maneuver. Demonstration experiments were conducted for validating the ship maneuvering AI using model ships or actual ships. Then, we have confirmed not only its maneuvering ability, but also its robustness against sensor noises, modeling errors, and external disturbances. In addition, we have succeeded to visualize rewards reflecting the sensation of experts by inverse reinforcement learning, and to incorporate the transfer learning into enhancing the ship maneuvering AI. These achievements can be led to the basic technologies that will accelerate future ocean-bottom exploration.
|
| Free Research Field |
船舶海洋工学
|
| Academic Significance and Societal Importance of the Research Achievements |
船舶は慣性が大きく，操船に対する応答性が極めて遅い．また，波や風などの自然外乱を空間的・時間的に把握することは困難である．本研究では，深層強化学習を用いて海洋底探査に求められる高度な判断と緻密な操船を自律操船AIに置き換えることに挑戦した．シミュレーションのみによる研究が数多く報告される中で，実船を用いた実証実験を繰り返し，AIの操船能力に加えてモデル化誤差や外乱影響に対するロバスト性までを実証できたことは学術的に意義がある．また，若年労働者の減少により船員の安定確保が困難となりつつある日本において，無人運航船の早期実現へと繋がる本研究の成果は社会的にも大きな意味がある．
|
