3D UAV Path Planning via Potential Filed-Imitation Reinforcement Learning

UAV applications have surged in recent years, creating increasingly complex task environments. Path planning algorithm quality directly impacts UAV safety and task efficiency. While the artificial potential field method (APF) excels in multi-UAV path planning, it is susceptible to local optima and unattainable goals. To overcome these difficulties, we introduce a dynamic APF method based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. Additionally, we propose a multi-agent TD3 (MATD3) algorithm based on the APF method. Lastly, we leverage the behavioral cloning method to validate the network performance. Experimental results show the effectiveness of the proposed algorithms.