AdaTP: An Adaptive Trajectory Planning Method for Intelligent Vehicles in Cooperative Driving Based on Reinforcement Learning

Cooperative driving represents a promising avenue for the implementation of Intelligent and Connected Vehicles (ICVs) and holds great potential for overcoming the limitations of individual vehicle perception, thereby enhancing the overall efficiency of the transportation system. Current studies dominantly focus on cooperative driving strategies in a multi-vehicle context, neglecting the specific actions undertaken by each vehicle. In this paper, we introduce Adaptive Trajectory Planning (AdaTP), a trajectory planning method for intelligent vehicles in cooperative driving. It was devised to make the trajectory granularity adjustable for the adaptation of the requirements in different traffic scenarios. Vehicle’s Markov Decision Process (MDP) quaternion and trajectory generation policy were exquisitely designed, adopting the framework of discrete reinforcement learning. Subsequently, we instantiated AdaTP in highway scenario and intersection scenario, where an intelligent vehicle was required to automatically plan its trajectory. Results show that AdaTP could perform a higher quality in both highway and intersection scenarios compared with traditional methods. Our proposed AdaTP method is expected to promote the research on fine-grained trajectory planning of ICVs.