Vehicle–Infrastructure Cooperation Framework for Vehicle Navigation and Traffic Signal Control using Deep Reinforcement Learning

作者：Long Wang, Weibin Zhang, Ziqi Yan · 发表于：Transportation Research Record Journal of the Transportation Research Board · 年份：2025 · DOI：10.1177/03611981251359285 · 被引用次数：1 · 研究领域：Traffic control and management、Traffic Prediction and Management Techniques、Autonomous Vehicle Technology and Safety

As the number of vehicles continues to rise, traffic congestion has become a significant factor affecting travel efficiency. The development of intelligent connected technology offers new approaches for achieving coordinated optimization in road traffic. This study proposes a deep reinforcement learning-based vehicle–infrastructure cooperation framework to achieve cooperative control of vehicle navigation and traffic signals, aiming to minimize the impact of traffic congestion on road traffic efficiency. The tasks of vehicle navigation and signal control are modeled as a partially observable Markov decision process. Based on real-time road condition information, roadside agents can flexibly adjust signal phases, while vehicle agents determine the next routing step as vehicles approach intersections. Through communication cooperation, on-board units and roadside units can acquire more comprehensive and accurate traffic status information. Rewards for the vehicle navigation task, based on spatiotemporal dimensions, and for the pressure-based signal control task, ensure that reinforcement learning agents continuously improve their traffic efficiency optimization capabilities during training. The proposed method was implemented and evaluated in the SUMO (Simulation of Urban MObility) simulator under low, medium, and high traffic demand conditions and compared with other baseline methods. The study also analyzed the impact of reward design, the training sequence, and the intellige...