Abstract:

A trajectory planning method for automated valet parking (AVP) was proposed in dynamic and complex parking environments, and its performance was evaluated. A spatiotemporal environment model was constructed through lightweight preprocessing of static obstacle maps and state indexing of dynamic obstacles. The temporal dimension was incorporated into the search process, and a signed distance field heuristic was introduced to improve search efficiency and dynamic obstacle avoidance. In addition, a Reeds-Shepp (RS) curve generation strategy based on configuration-space reverse sampling and cascaded pruning was developed to guide the vehicle into the target parking space. The results show that the proposed method achieves a planning success rate of 100% in conventional parking scenarios, which is 49.92% and 1.59% higher than those of conventional Hybrid A* and Spatiotemporal Hybrid A*, respectively. The planning time is reduced by 53.8% and 56.9% in perpendicular and inclined parking scenarios, respectively, and decreases to 0.61 s in the parallel parking scenario. In dead-end parking scenarios, the planning time for perpendicular and parallel parking is reduced by 71.0% and 77.6%, respectively. Moreover, the path length for parallel parking is shortened by 9.9%, and the gear shift number is reduced by 66.7%. Therefore, it demonstrates the effectiveness of the proposed method in dynamic and complex parking environments.

Key words: intelligent transportation, automated valet parking (AVP), trajectory planning, dynamic obstacle avoidance, Reeds-Shepp (RS) curve