Research on Cooperative Path Planning for UAVs Based on Multi-agent Deep Reinforcement Learning

This paper proposes a cooperative path planning method for multi-UAV systems based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. A composite reward function is designed, which includes arrival rewards, obstacle avoidance penalties, and formation maintenance rewards. The method combines centralized training and decentralized execution frameworks to achieve efficient cooperation among multiple UAVs. Simulation results show that this approach can effectively plan smooth paths in complex environments with obstacles and significantly improve the average reward. This study provides new theoretical support and technical solutions for the autonomous decision-making of multi-UAV systems.