[3O1-GS-10v] AI応用：実世界システム

9:15 〜 9:30

[3O1-GS-10v-02] Cooperative Vehicle-Road Scheduling with Optimization of Reinforcement Learning Signal Control and Dynamic Route Planning

〇Yue Wang¹, Hideki Fujii¹ (1. The University of Tokyo)

キーワード：

V2X、Reinforcement Learning、Route Planning、Traffic Signal Control、Vehicle-Road Cooperation

Driven by the rapid advancement of transportation technologies, Connected Vehicles (CV) and Vehicle-to-Everything (V2X) communication have emerged as promising solutions to urban traffic congestion. While prior research has predominantly focused on either optimizing individual vehicle behavior or enhancing infrastructure-based signal control, this study proposes a collaborative optimization framework for the integrated vehicle-infrastructure system. Leveraging Vehicle-to-Infrastructure (V2I) communication, the system operates on two levels. On the infrastructure side, a Reinforcement Learning (RL) agent processes intersection states and vehicle movement intentions to provide adaptive signal control. Meanwhile, CVs utilize the Artificial Potential Field (APF) method for dynamic route planning, informed by predictive data from the signal control agents. The proposed framework was implemented and validated using the SUMO microscopic traffic simulator. Results demonstrate that the system achieves state-of-the-art performance compared to existing baselines.