A Roadside End‐to‐End Model for Rapid Parking Path Planning Over the Occupancy Grid Map

Rule‐based parking path planning algorithms have been widely employed in automated parking systems (APS), relying on the prior map or data collected by sensors to perceive the environment and generate paths. However, they often perform poorly in terms of perception and rapid path generation in complex scenes. In this paper, we propose the r oad s ide p arking p ath p lanning model (RSP3), a novel end‐to‐end model that combines the advantages of the roadside unit (RSU) and end‐to‐end neural networks. Based on the vehicle‐to‐infrastructure cooperation, the RSP3 can use the global view provided by the RSU to acquire the scene information and generate discrete path points on the occupancy grid map with a classification‐based modelling method. It reduces task complexity and enhances the abilities of learning and imitating the traditional algorithm with improved flexibility and robustness. We developed an automatic data collection and evaluation platform based on the CARLA simulator and used the hybrid A* algorithm as the baseline. The results show that the RSP3 can not only quickly perform accurate segmentation of the current scene and understand the intention of the autonomous driving vehicle but also generate high‐quality paths, which shows its potential in the efficient and robust automatic parking task. To advance further research, the code will be available at: https://github.com/ppllracing/RSP3.git .