Agentic AI-Driven UAV Network Deployment: A LLM-Enhanced Exact Potential Game Approach

TL;DR

A dual spatial-scale UAVN topology optimization framework based on exact potential games (EPGs), enhanced by Agentic AI is proposed, demonstrating that the proposed framework consistently outperforms baseline methods in terms of energy consumption, end-to-end latency, and system throughput.

Abstract

Unmanned Aerial Vehicular Networks (UAVNs) are envisioned to provide flexible connectivity, wide-area coverage, and low-latency services in dynamic environments. From an agentic artificial intelligence (Agentic AI) perspective, UAVNs naturally operate as multi-agent systems, where autonomous UAVs act as intelligent agents that coordinate deployment and networking decisions to achieve global performance objectives. However, the strong coupling between discrete link decisions and continuous deployment parameters makes UAVN topology optimization a mixed-integer nonconvex problem, resulting in challenges in scalability, efficiency, and solution consistency under dynamic network conditions. This paper proposes a dual spatial-scale UAVN topology optimization framework based on exact potential games (EPGs), enhanced by Agentic AI. At the large spatial scale, a log-linear learning based EPG (L3-EPG) algorithm is developed to optimize inter-UAV link configurations, enabling sparse yet connected network topologies while reducing redundant links and interference. At the small spatial scale, an approximate gradient based EPG (AG-EPG) algorithm jointly optimizes UAV deployment, transmission power allocation, and ground user (GU) association to improve network throughput and latency. To further enhance adaptability across heterogeneous scenarios, a large language model (LLM) is incorporated as a knowledge-driven decision enhancer to automatically generate utility weights according to network characteristics, alleviating reliance on manual parameter tuning. Simulation results demonstrate that the proposed framework consistently outperforms baseline methods in terms of energy consumption, end-to-end latency, and system throughput.

Figures (8)

• Figure 1: Agentic AI–driven UAV network deployment model. The model presents an intent-driven UAV network deployment framework. A Agentic AI system translates high-level deployment intents into structured parameters for an EPG. Each UAV optimizes its coordinates and transmit power in a distributed manner, leading to dynamic topology evolution. The network converges to a stable Nash equilibrium aligned with the specified deployment intent.
• Figure 2: The decomposition process of the Agentic AI-driven UAVN deployment problem. Agentic AI is capable of decoupling high-level tasks, generate sub-problems, plan actions, and interact with other tools or environments with human zhang2026toward. First, the network deployment problem is divided into two subproblems. Then, these subproblems are solved using L3-EPG and AG-EPG, respectively. Finally, the RAG-LLM module is used to generate adaptive weight coefficients for the utility function through multi-source knowledge retrieval and logical verification, aiming to obtain a network deployment scheme and parameters that meet the actual needs of the network.
• Figure 3: Framework of RAG-based Large Language Model assisted UAV network optimization. The framework consists of four modules: ① Knowledge Vector Database Module, which performs text segmentation and vector embedding on the pre-constructed knowledge base and literature to establish a retrievable database; ② Knowledge Retrieval Module, which receives the network manager's intent and retrieves the most relevant knowledge fragments via cosine similarity calculation; ③ Result Generation Module, which feeds the retrieved domain knowledge into the LLM as contextual prompts; and ④ Iterative Refinement Module, where the LLM generates initial utility functions and weighting coefficients. The manager provides feedback for the LLM to refine the results by integrating knowledge base information until the generated parameters satisfy the specific network requirements.
• Figure 4: Retrieval precision under different knowledge block sizes.
• Figure 5: Verification of convergence and global consistency of algorithms.