Joint Soft Actor-Critic and Adaptive Potential Field-Based Path Planning for Vehicle Obstacle Avoidance

Vehicle obstacle avoidance path planning (VOAPP) plays a critical role in improving the safety and efficiency of autonomous driving systems. However, existing VOAPP methods frequently suffer from limited adaptability and suboptimal path quality in complex dynamic environments, resulting in reduced performance and increased safety risks. To solve these problems, this paper proposes a joint soft actor-critic and adaptive potential field (SAC-APF) VOAPP method. First, we propose the APF to represent the potential field distribution around the vehicle. Specifically, we establish an attractive potential field for the target position to represent its guiding effect on the vehicle, and develop the repulsive potential field based on the shape and size of obstacles to accurately reflect their repulsive impact on the vehicle. Second, we propose an improved SAC with a comprehensive state space, an action space, and a novel reward function to process the APF output and surrounding environmental information to generate an initial obstacle avoidance path. Subsequently, we employ cubic spline curves to optimize the initial generated path. Finally, the results demonstrate that the proposed method significantly outperforms other VOAPP methods in terms of obstacle avoidance, path smoothness and computational efficiency.