ActorRL: A Novel Multi-level Receptive Fields Adversary Reinforcement Learning for Automoumous Intersection Management

As an emerging technology, Connected Autonomous Vehicles are believed to be able to pass intersections with greater efficiency, and related researches have been conducted for decades, however, compared to pre-designed model-based or optimization-based scheduling passing plan, the application of distributed reinforcement learning in the field of autonomous intersection management (AIM) has only begun to emerge in the past two years and confronts many challenges. Our study design a multi-level learning framework with various observation scope, action steps and reward period to make full use of information around vehicle and help to figure out the best interactive strategy for all vehicles. Our experiment has proven this framework can significantly enhance safety compared to RL without it, and improve efficiency compared to baselines.