Research on Path Planning based on improved SAC Algorithm fused with RRT algorithm

This paper is concerned with a novel path planning algorithm that integrates the improved Soft Actor-Critic (SAC) algorithm with Rapidly-exploring Random Trees (RRT) to address the path planning problem for complex environments. Al-though SAC performs well in continuous action spaces, it typically provides rewards only when the agent reaches the target or encounters a collision. This can result in low exploration efficiency and slow convergence, particularly in complex environments. To address these issues, this study incorporates Potential-Based Reward Shaping (PBRS) into the SAC algorithm, enhancing the guidance provided to the agent by the target point and utilizing it as a control mechanism. The RRT algorithm is then used to generate a collision-free path from the start to the target, producing a sequence of waypoints. Finally, the controller directs the agent to traverse these waypoints to reach the target. Simulations conducted in Safety-Gym show that incorporating Potential-Based Reward Shaping (PBRS) into SAC facilitates faster learning, while combining it with the RRT algorithm improves the success rate and reduces collisions.