The invention relates to a variable probability bidirectional rapidly-exploring random tree improved path planning algorithm. Firstly, when importing a state space map, state space preprocessing needsto be performed according to the setting of the vehicle volume, and the state space edge is extended and protected to prevent the node from excessively approaching the state space edge to cause a collision. Secondly, a variable probability target selection strategy based on the node environment is adopted to accelerate the convergence speed. Finally, the bent part in the generated path is removedwhile the straight part in the generated path is retained, that is, whether connection lines between the first node as the start and the subsequent nodes have an obstacle is judged, the redundant nodes are deleted, the path is optimized, and the number of turns and the total path length during the driving of the trolley are reduced. The variable probability optimization algorithm is achieved based on the original bidirectional RRT algorithm, and the target pointing method is used to improve the search speed and reduce the calculation amount. At the same time, the path length and the number ofnodes are reduced, and the passability is ensured.