Local Path Planning of Mobile Robots Based on the Improved SAC Algorithm

This paper proposes a new EP-PER-SAC algorithm to solve the problems of slow training speed and low learning efficiency of the SAC (Soft Actor Critic) algorithm in the local path planning of mobile robots by introducing the Priority Experience Replay (PER) strategy and Experience Pool (EP) adjustment technique. This algorithm replaces equal probability random sampling with sampling based on the priority experience to increase the frequency of extracting important samples, thereby improves the stability and convergence speed of model training. On this basis, it requires to continuously monitor the learning progress and exploration rate changes of the robot to dynamically adjust the experience pool, so the robot can adapt effectively to the environment changes and the storage requirements and learning efficiency of the algorithm are balanced. Then, the algorithm's reward and punishment function is improved to reduce the blindness of algorithm training. Finally, experiments are conducted under different obstacle environments to verify the feasibility of the algorithm based on ROS (Robot Operating System) simulation platform and real environment. The results show that the improved EP-PER-SAC algorithm has a shorter path length and faster model convergence speed than the original SAC algorithm and PER-SAC algorithm.