Research on Global Path Planning Based on the Integration of the APF and Soft-Actor Critic Algorithms

Abstract

Global path planning module plays an important role in autonomous navigation of robots. However, traditional algorithms have slow path solving speed, long path length and poor path smoothness. To address these issues, this paper combines artificial potential field(APF) and soft-actor critic(SAC) algorithm and proposes the hybrid APF-SAC (HAS) algorithm. The method constructs a motion risk potential field model, which is used to supervise the learning of the two newly incorporated critical networks. Then, the network prediction values are added to the objective function of the actor network, guiding the agent to choose a state with a smaller potential value. Finally, we compare our proposed algorithm with other popular methods in simulated environments. The experimental results show that the method has lower time consumption, shorter path length and higher path quality compared to traditional methods and deep reinforcement learning methods.