| 講演抄録/キーワード |
| 講演名 |
2025-08-29 09:25
A Hybrid Reinforcement Learning Framework for Cooperative Driving: From Algorithm Design to Real-World Validation ○Yu Taoyuan・Wang Kui・Li Zongdian・Yu Tao・Kei Sakaguchi(Institute of Science Tokyo) RCS2025-114 |
| 抄録 |
(和) |
Unsignalized intersections face significant challenges in safety and efficiency due to complex traffic dynamics. This paper proposes an RSU-centric cooperative driving framework that leverages global perception and vehicle-to-infrastructure communication. The framework employs a two-stage hybrid reinforcement learning method, combining offline pre-training with online fine-tuning, and incorporates the multi-attention mechanism to address inter-agent dependencies. Extensive experiments show that the proposed method achieves a failure rate below 0.03% when controlling three connected autonomous vehicles. Moreover, it exhibits strong robustness to different numbers of agents and generalizes well to other traffic maps. |
| (英) |
Unsignalized intersections face significant challenges in safety and efficiency due to complex traffic dynamics. This paper proposes an RSU-centric cooperative driving framework that leverages global perception and vehicle-to-infrastructure communication. The framework employs a two-stage hybrid reinforcement learning method, combining offline pre-training with online fine-tuning, and incorporates the multi-attention mechanism to address inter-agent dependencies. Extensive experiments show that the proposed method achieves a failure rate below 0.03% when controlling three connected autonomous vehicles. Moreover, it exhibits strong robustness to different numbers of agents and generalizes well to other traffic maps. |
| キーワード |
(和) |
Cooperative driving / reinforcement learning / V2I / CAVs / RSUs / / / |
| (英) |
Cooperative driving / reinforcement learning / V2I / CAVs / RSUs / / / |
| 文献情報 |
信学技報, vol. 125, no. 163, RCS2025-114, pp. 81-86, 2025年8月. |
| 資料番号 |
RCS2025-114 |
| 発行日 |
2025-08-21 (RCS) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
RCS2025-114 |