Agentic AI Networking for Heterogeneous Drone Systems in Low-Altitude Wireless Networks

Summary

Low-altitude wireless networks are being developed to support heterogeneous unmanned aerial systems that provide multiple services in the same three-dimensional airspace. Because mobility, connectivity, and shared resources interact, while service requirements change over time, the paper models coordination as a dynamic non-cooperative game. It argues that conventional optimization and learning controllers are constrained by predefined objectives and therefore have limited ability to adapt autonomously when priorities change. The proposed system is a hierarchical hybrid architecture with two loops. In the outer loop, an LLM interprets service requirements and operator intent, orchestrates the game, and reconfigures objectives and resource priorities. In the inner loop, decentralized, parameter-conditioned multi-agent reinforcement learning policies operate under the configured game. A logistics-monitoring case study shows coordinated coexistence among heterogeneous services as operating conditions evolve, without retraining the underlying MARL policies. The paper closes by identifying scalability, trustworthiness, and adaptive coordination as open challenges for agentic low-altitude wireless networks.