Abstract:

A lane-level skeleton guided hierarchical path planning method for lane considered, skeleton assistance in conjunction with reeds-shepp (RS) curve (LCSA-RS) was proposed to address the real-time and safety challenges of path planning for automated valet parking in complex parking scenarios. The method employed a five-layer architecture: the parking spot decision layer determined optimal park-in/park-out points based on parking map; the map abstraction layer integrated skeletonization algorithm with lane constraints to construct sparse topological map; the global guidance layer generated key waypoint sequences using the A* algorithm; the path optimization layer produced smooth paths satisfying kinematic constraints within circle regions around key points; and the collision detection layer performed real-time risk assessment and triggered path replanning when necessary. The results show that, compared with the hybrid A* algorithm, the proposed LCSA-RS reduces the number of nodes searched in the global planning phase to one-thousandth of the former and shortens the total planning time by 95.5%, while confining planned paths within their respective lanes and preventing multi-vehicle path conflicts, thus providing a novel solution for real-time path planning in complex parking environments.

Key words: autonomous valet parking, path planning, A* algorithm, Reeds-Shepp (RS) curves, skeletonization algorithm