PG-ITD3: A Potential Field-Guided Deep Reinforcement Learning Approach for UAV Path Planning After Disaster

To address the target inaccessibility issue of the Artificial Potential Field (APF) method and the slow convergence of deep reinforcement learning algorithms, we propose a method combining the artificial potential field and an improved twin delayed deep deterministic policy gradient algorithm (PG-ITD3) for UAV path planning in post-disaster scenarios. This approach considers Unmanned Aerial Vehicles (UAV) kinematic constraints and dynamically adjusts the repulsive gain coefficient through deep reinforcement learning to enhance planning efficiency. The introduction of a virtual obstacle strategy, combining the aftershock probability model with prioritized experience replay (PER), and a novel reward function facilitates real-time reward acquisition and accelerates convergence. Simulation results demonstrate that our proposed algorithm outperforms traditional APF method and other advanced deep reinforcement learning methods in success rate, path length, path reward, and global smoothness. This research offers an effective solution for post-disaster UAV rescue path planning, contributing significantly to enhancing rescue efficiency and safety.