Path Planning for Unmanned Aerial Vehicle Swarm Based on Electromagnetic Environment Sensing

Unmanned aerial vehicle (UAV) swarm is widely used in tasks such as post-disaster rescue and battlefield monitoring. These tasks are often executed in unknown or complex environments, necessitating the programming of safe and efficient paths for UAV swarm to ensure the completion of missions. To address the path planning problem for UAV swarm in unknown electromagnetic environment, we propose a multi-agent deep deterministic policy gradient algorithm based on environment sensing where a safe learning mechanism is designed by using control barrier function. Additionally, a weakly supervised learning-based generative adversarial network algorithm is employed to construct an electromagnetic environment sensing module. By using the algrothim we propose, UAV swarm can avoid zones with strong electromagnetic interference and guarantee inter-UAV collisions avoidance during task execution. Compared to reinforcement learning algorithm without environment sensing module and safe learning mechanism, the algorithm we propose reduces convergence time by approximately 2.5 times. Simultaneously, it prevents individual trial-and-error learning process from violating safety constraints, ensuring the safety of UAV swarm in unknown environment. Finally, we verified the effectiveness of our algorithm on the experimental platform which is constructed by using universal software radio peripherals and quadcopter UAVs.