Abstract: To address the issues of insufficient goal orientation, high randomness, low planning efficiency, and path redundancy leading to non-smooth motion in the traditional Informed-RRT^* algorithm for obstacle avoidance path planning of robotic arms, an improved APF-Informed RRT^* algorithm is proposed. In the sampling stage, the artificial potential field method is introduced to guide the sampling points, enhancing the goal orientation of the random tree expansion. In the tree growth stage, a collision detection cache key and an adaptive variable step size strategy are proposed. Collision detection results are cached and a unique identifier, the cache key, is generated. The step size is dynamically adjusted based on the distance to obstacles to reduce the number of collision detections and improve search efficiency. In the optimization stage, a piecewise greedy algorithm and cubic B-spline curve are used to optimize the path. Through simulation experiments in a three-dimensional environment comparing with Informed-RRT^*, GoalBias-RRT^*, and RRT^* algorithms, the results show that the path length is reduced by 14.08% compared to Informed-RRT^*, and the time is saved by 96.17%. Compared with RRT^* and GoalBias-RRT^*, the path lengths are reduced by 19.89% and 12.24% respectively, and the time is saved by 88.52% and 12.5% respectively. The search efficiency of the algorithm is significantly improved. When the algorithm is applied to the AUBO i5 robotic arm platform, the robotic arm can successfully avoid obstacles and accurately reach the target point, and the motion curves of each joint are smooth, further demonstrating the effectiveness of the improved algorithm.