Abstract: In response to the problems of blindness, low efficiency, and uneven path in 3D obstacle avoidance path planning of the rapid-exploration random tree^* (RRT^*) algorithm. This article proposes an improved RRT^* algorithm to enhance the performance of welding robot path planning. This algorithm uses a bidirectional search strategy to shorten search time. Combining artificial potential field (APF) and RRT^* algorithms to improve path smoothness and balance local optimization and global optimization; Propose an improved APF strategy based on angle and density to enhance the efficiency of obstacle avoidance and path guidance. Propose a dynamic target bias strategy and a dynamic step size strategy to enhance the algorithm's adaptability and search efficiency in areas with dense and sparse obstacles; Adopt a path pruning strategy to shorten and smooth the path. Finally, the efficiency and practicality of the improved algorithm were verified through comparative simulation experiments and real machine experiments with three algorithms: RRT^*, APF-RRT^*, and bidirectional-APF-RRT^* (Bi-APF-RRT^*).