Cusp poses prediction algorithm for parking paths inspired by the curl field

Efficiently completing parking path planning in busy, crowded environments is critical for improving parking efficiency and passenger experience. Parking path planning must satisfy kinematic constraints and obstacle avoidance in a continuous state space. This problem is PSPACE-hard and generally requires exponential time. To address this challenge, we propose an engineering approximation method with polynomial complexity. The method integrates start pose, end pose, and scene boundaries. It employs a curl-field-inspired heuristic function to predict the cusp pose of the parking path (CFCP). The path is then decomposed into sub-paths based on the predicted cusp pose. This reduces planning time to milliseconds while maintaining a high success rate. Simulations validate the algorithm’s performance across various parking environments. Results show that CFCP achieves successful obstacle avoidance in all five typical parking scenarios evaluated in this study, yielding a 100% success rate within the tested scenario set. Its planning time scales with scene complexity but remains an order of magnitude faster than hybrid A*. Notably, in a right-angle turn parking scenario at a road end, CFCP requires only 3.8% of the time needed by hybrid A*.