Cooperative Pursuit-Evasion with Low Altitude Wireless Network: A Hierarchical Reinforcement Learning Approach

As an emerging countermeasure, cooperative interception by multiple UAVs offers an effective solution to neutralize rogue drones and safeguard low-altitude airspace operations. Effective coordination among counter-UAVs in encircling intruding drones remains challenging. This paper proposes a Hierarchical Cooperative Deep Reinforcement Learning (HCDRL) algorithm to enhance cooperation and efficiency among UAVs pursuing agile targets. The proposed approach decomposes the multi-agent pursuit-evasion scenario into multiple subtasks using a two-layer hierarchical decision-making framework. Specifically, the upper-layer network acts as a meta-strategy, dynamically assessing pursuit scenarios and assigning optimal subtasks. Meanwhile, the lower-layer policy networks of individual agents determine maneuver actions based on local observations and assigned subtasks. Simulation results demonstrate that the proposed algorithm significantly improves multi-agent cooperative encirclement performance, achieving an 11.18% higher success rate and a 9.94% reduction in completion time compared to state-of-the-art methods.