Abstract: Path planning is a key technology that urgently need to be solved in application of coal mine robots in unstructured narrow confined spaces underground. The traditional probabilistic road map (PRM) algorithms are difficult to ensure uniform distribution of sampled nodes in free space in narrow and enclosed roadway environments, resulting in path planning failure. Nodes may be close to obstacles, resulting in poor passability of the planned path. In order to solve the above problems, a path-planning method for coal mine robot based on improved PRM algorithm is proposed. In the constructive phase, the artificial potential field method is introduced to push the node falling in the obstacle to the free space along the direction of the connection line of the node in the free space nearest to it. The repulsive force field is established at the edge of the obstacle to realize uniform distribution of nodes and make them a certain distance from the obstacle. In the query phase, the D^* Lite algorithm is integrated to achieve path re-planning when encountering dynamic obstacles or when the front is impassable. The simulation results show that the nodes of the improved PRM algorithm are uniformly distributed in free space and are at a certain distance from obstacles. It improves the safety of path planning. When the number of nodes is 100, the success rate of the improved PRM algorithm is 25% higher than that of the traditional PRM algorithm. As the number of nodes increases, the number of successful path-planning attempts for both traditional and improved PRM algorithms shows an increasing trend. But the improved PRM algorithm has a more significant advantage in efficiency. When the number of nodes is 400, the operational efficiency of the improved PRM algorithm is 35.13% higher than that of the traditional PRM algorithms. The planned path is smoother and the path length is shorter. When obstacles suddenly appear, the improved PRM algorithm can achieve path re-planning.