Unstructured Road Obstacle Avoidance Path Planning Algorithm Based on BEV Grid Map

Abstract

 In view of the narrow, unstructured road surfaces, irregular distribution of obstacles, and the large body and steering lag characteristics of mining trucks, traditional path planning algorithms tend to generate edge-hugging paths or infeasible broken paths, posing collision risks. Based on high-precision BEV grid maps generated by LiDAR-camera multi-sensor fusion, this paper proposes an improved hybrid A* obstacle avoidance path planning algorithm that combines safety potential fields with kinematic constraints. The algorithm constructs an environmental safety potential field through Euclidean distance transformation to quantify collision risks in the grid space; it expands the A* algorithm's state space and improves the cost function by introducing steering penalties and maximum steering angle limits to ensure kinematic feasibility of the path; finally, cubic B-spline curves are used to smooth the discrete path, producing a curvature-continuous trajectory. Simulation results show that, compared with the traditional A* algorithm, the proposed algorithm increases the minimum safety distance between the path and obstacles by more than 17 times, effectively eliminates broken paths, and its planning time meets the real-time control requirements for low-speed operation of mining trucks, providing safe and smooth local navigation trajectories for autonomous mining vehicles on unstructured roads.