Agentic AI-Empowered Reliable Routing in Interference-Aware UAV Networks
Abstract
The burgeoning low-altitude economies place stringent demands on reliable communication within uncrewed aerial vehicle (UAV) networks. However, in complex electromagnetic environments, the topology and communication link quality of highly maneuverable UAVs exhibit strong random fluctuations, leading to frequent routing path failures. Agentic AI integrated with embodied intelligence is leveraged to empower UAV networks, thereby significantly enhancing communication reliability and autonomous adaptability. Specifically, we propose an interference-aware multi-agent cooperative routing optimization framework HRC-QMIX for embodied-enhanced communication, integrating the hypergraph module to characterize the cooperative dependencies among multiple embodied nodes, thereby supporting more stable forwarding decisions. Within a value decomposition learning framework, we further introduce a dynamic mixing mechanism based on recursive atrous self-attention (RASA) to enhance the expressive ability of the joint value function for complex cooperative relationships. Furthermore, causal-inspired regularization term is designed to alleviate the instability of credit allocation in strongly non-stationary scenarios and improve training stability. Simulation results demonstrate that the proposed method exhibits superior communication performance and stronger anti-interference robustness in scenarios with multiple interference sources and varying maneuverability levels.