Path planning for autonomous vehicles using an improved artificial potential field method

Abstract

To address the problems of unreachable target, collisions near the start point, local minima solutions, and weak dynamic programming capabilities of the traditional artificial potential field (APF) in path planning, this paper proposes the B-APF algorithm, which combines the improved APF method with the fusion RRT algorithm. Firstly, introduce a target distance adjustment factor in the repulsive potential field to solve the problem of an unreachable target. Secondly, a dynamic gain coefficient is introduced into the gravitational field to alleviate the collision problem near the start point. Thirdly, the virtual target point mechanism is adopted to overcome the local minima problem. Finally, integrate the RRT algorithm to enhance the dynamic programming capability. MATLAB simulation shows that, compared with the F-APF algorithm, B-APF can reduce the average path length by 0.54%, the computation time by 2.93%, the number of nodes by 2.62%, and significantly improve the path smoothness at the same time. Compared with the RRT algorithm, B-APF achieves a 7.43% reduction in the average path length, a 94.00% decrease in the average curvature, an 80.99% reduction in the maximum curvature, a 91.30% decrease in the standard deviation of curvature, and significantly improves the path smoothness at the same time. Robustness tests under sensor noise, parameter uncertainty, and external interference conditions show that the average position error is only 0.025 m and the average lateral acceleration is only 2.851 m/s², confirming the real-time feasibility. The MPC-based tracking validation achieved a lateral error of <0.053 m and an average center-of-mass lateral deflection angle of <0.07°, demonstrating dynamic feasibility within friction limits. These results establish B-APF as a practical, robust, and dynamically feasible solution for self-driving vehicle path planning in complex environments.