Preliminary Study on Cooperative Route Planning Reinforcement Learning with a Focus on Avoiding Intersection Congestion

Abstract

Intersection control systems have been actively studied in recent years as they could potentially replace traffic signals via the utilization of the communication and automatic driving capabilities of connected and autonomous vehicles (CAVs). In these studies, conflicting travel trajectories at intersections that could cause accidents and delays were safely and efficiently avoided by controlling the vehicle’s speed. However, routing approaches for avoiding conflicts at intersections have only been discussed in a few studies. To investigate the feasibility of avoiding intersection conflicts through network-level route allocation, we propose a cooperative route allocation model using reinforcement learning than can model the relationship between the complex traffic environment and optimal route solutions. Models aimed at decreasing the total travel time and those with high delay importance owing to conflicts in travel times were trained and verified under multiple traffic conditions. The results indicate that our model effectively allocates vehicles to their optimal routes, reducing the number of intersection conflicts and decreasing the average travel time by up to approximately 40 s compared to random allocation, demonstrating the potential of reinforcement learning for cooperative route allocation in the management of multiple vehicles.