Abstract: The coal gangue sorting environment is complex. In order to avoid the collision between robot and obstacles and improve sorting efficiency, it is necessary to carry out path planning for robot. The principle of coal gangue sorting system is analyzed. The path planning problem of coal gangue sorting robot is summed up as planning a collision-free path from a given starting point to a target point in the environment of obstacles, and the two constraints of high speed and avoiding collision with obstacles must be met at the same time. Combining the advantages of Cartesian space and joint space, a path planning scheme for coal gangue sorting robot with path planning in joint space and collision detection in Cartesian space is proposed. The scheme does not need to carry out kinematic inversion of the robot, and can avoid describing obstacles in joint space. In order to solve the problem of blindness in the improved rapidly-exploring random trees (RRT*) path planning algorithm, a variable probability target bias strategy is proposed and introduced into RRT* algorithm to obtain the G-RRT* algorithm. The target bias strategy with variable probability increases the target bias probability in the obstacle-free area so as to enhance the target orientation of the algorithm. In the obstacle area, the target bias probability value is reduced to ensure the obstacle avoidance capability of the algorithm. The G-RRT* algorithm combines the variable probability target bias strategy with RRT* algorithm. The G-RRT* algorithm not only retains the asymptotic optimal path length characteristic of RRT* algorithm, but also improves the target orientation of the algorithm, and can improve the path planning efficiency greatly. The experimental results show that compared with RRT-Connect algorithm and RRT algorithm with fixed probability target bias strategy, the G-RRT* algorithm can get the shortest average path length, and is more suitable for path planning of coal gangue sorting robot.