In recent decades, RRT* algorithm has received wide attention because of its asymptotic optimization problems. However, RRT* algorithm still has some problems such as slow convergence, low efficiency of nearest neighbor search and large randomness of search range. In view of the above shortcomings, a hybrid APF-RRT* algorithm based on KD-Tree nearest neighbor search is proposed. By introducing K-dimension tree and Artificial Potential Field (APF), the disordered sampling points of RRT tree are divided into ordered sampling points by KD-tree. In addition, the sampling points of RRT tree are sampled to target points according to the way of potential field decline, so as to improve the speed, purpose and direction of RRT tree expansion. Simulation experiments show that the improved APF-RRT* algorithm reduces the number of iterations and convergence time. APF-RRT* algorithm has strong adaptability to the complexity of maps and the change of area.