Path Planning Based on an Improved Artificial Potential Field with Deep Deterministic Policy Gradient
Zheng Xin Guo, Jianglong Yu, Yiming Chen · IFAC-PapersOnLine · 2025
This paper presents a novel path planning algorithm for unmanned aerial vehicles (UAVs) in dynamic obstacle environments by integrating the Deep Deterministic Policy Gradient (DDPG) with the Artificial Potential Field (APF) method. The proposed DDPG-APF algorithm addresses the limitations of traditional APF by enabling adaptive parameter tuning through reinforcement learning. A comprehensive state representation framework is designed to capture geometric relationships and dynamic interactions, while a multi-component reward mechanism balances obstacle avoidance, path efficiency, and smoothness. The algorithm’s performance is validated through simulations. Results demonstrate significant improvements over baseline APF: an average 18.2% reduction in path length. The cumulative reward curves confirm stable convergence, indicating robust decision-making under diverse conditions. This research contributes a scalable solution for UAV navigation in complex obstacle environments, combining the interpretability of APF with the adaptability of DDPG. Future work will focus on optimizing computational efficiency for real-time applications and extending the framework to complex 3D scenarios such as urban airspaces.