Innovative Cooperative Guidance Algorithm for Multiple Aerial Vehicles Based on Hierarchical Multi-Agent Deep Reinforcement Learning

This paper addresses the cooperative control problem with desired terminal line-of-sight (LOS) angles in a dynamic vehicle-target coordination system. The group of aerial vehicles is modeled as a leader-follower system aiming to coordinate actions to hit the target while avoiding engagement with defender. Unlike existing studies, we propose a novel cooperative guidance control strategy based on hierarchical multiagent reinforcement learning (H-MARL), treating the target, aerial vehicles group, and defender as participants in a collaborative decision-making game. Each vehicle is modeled as an independent agent, trained from scratch without prior knowledge. A key innovation is the introduction of a noise agent, which generates exploration noise to support each aerial vehicle's actions toward effective exploration. This forms a hierarchical structure where low-level tasks (noise generation) support high-level tasks (guidance control), enabling the system to address complex cooperative control problems more efficiently. Simulation results demonstrate the ability of the vehicles group to intercept the target's trajectory at the desired LOS angles while the leader vehicle successfully avoids engagement, achieving a high control success rate.