Autonomous navigation at unsignalized intersections: A coupled reinforcement learning and model predictive control approach

This paper develops an integrated safety-enhanced reinforcement learning (RL) and model predictive control (MPC) framework for autonomous vehicles (AVs) to navigate unsignalized intersections. Researchers have extensively studied how AVs drive along highways. Nonetheless, intersections in urban environments remains a challenging task due the constant presence of moving road users, including turning vehicles, crossing or jaywalking pedestrians, cyclists. are thus required learn adapt dynamically evolving traffic environment. proposes design benchmark that allows sense real-time environment perform path planning. The agent generates curves feasible paths. ego vehicle attempts follow these paths under specific constraints. RL MPC navigation algorithms run parallel suitably selected enhance safety. AV is modeled with lateral longitudinal dynamics trained T-intersection using Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm various scenarios. It then tested on straight single multi-lane All experiments achieve desirable outcomes terms crash avoidance, driving efficiency, comfort, tracking accuracy. developed system provides adaptive can