To solve the problems of inefficient expansion and time-consuming planning of the traditional hybrid A* algorithm in parking scenarios, an improved Hybrid A* path planning method with dynamic adjustment of sampling arc length is proposed. When the algorithm is executed, the number of collision-free child nodes of the parent node is used to represent the distribution of environmental obstacles, and the sampling step length of the subsequent node expansion is corrected in real-time based on this. This not only improves the node expansion efficiency of the vehicle in the open scene outside the garage but also takes into account the obstacle avoidance requirements of the complex environment near the garage. ROS simulation experiments show that the improved hybrid A* algorithm has similar parking planning path lengths, but has obvious advantages in search time, expansion nodes, and running time, which can effectively solve the path planning problem in the parking field.