Multi-UAV Cooperative Path Planning via Graph Neural Network and Proximal Policy Optimization

Cooperative path planning for multiple unmanned aerial vehicles (UAVs) is a challenging problem in the field of multi-agent reinforcement learning, characterized by high-dimensional state spaces, dynamic environments, and communication constraints. To address these challenges, this paper proposes a novel framework integrating Graph Neural Networks (GNNs) with Proximal Policy Optimization (PPO). In the proposed approach, the multi-UAV system is modeled as a dynamic graph where GNNs aggregate interaction features among agents. A dynamic neighbor selection mechanism is introduced to enhance scalability and robustness in large-scale scenarios. Furthermore, a multi-objective reward function is designed to balance path efficiency, collision avoidance, energy consumption, and cooperative behavior. In challenging high-density scenarios, extensive simulations demonstrate that the proposed method improves the success rate by over 78% and reduces the collision rate by 73% compared to representative MARL baselines. These results confirm that the presented GNN-PPO approach is an effective and scalable solution for complex multi-UAV path planning tasks.