APPLICATION OF LLM IN UAV ROUTE PLANNING TASKS TO PREVENT DATA EXCHANGE AVAILABILITY VIOLATIONS

Abstract

This paper presents a novel approach to unmanned aerial vehicle (UAV) route planning under conditions of cyber-physical threats that affect system availability. In hostile and contested environments, UAVs are increasingly exposed not only to natural obstacles and electronic interference, but also to targeted cyber threats. The proposed method addresses both intentional and unintentional disruptions to communication and control, including interference from electronic warfare (EW) systems, jamming, and the propagation of malicious signals. Importantly, EW devices are considered not only as sources of electromagnetic noise, but also as potential vectors for malware distribution or attacks on wireless protocols, making them critical components of the cybersecurity threat landscape.To ensure mission success and maintain reliable data exchange, the method integrates terrain elevation maps, open for use digital elevation models (DEMs), and geospatial data services) with reasoning powered by large language models (LLMs). The system constructs UAV routes that preserve line-of-sight communication, avoid high-risk zones, and adapt to topological and adversarial constraints. A modular architecture is introduced, incorporating data preprocessing, mission-context prompt generation, LLM-based inference, and post-processing validation. Custom prompt templates are developed to inject mission-specific cyber and physical context, guiding the LLM to avoid hallucinations and enhance security-aware planning. A computational experiment using real elevation data and the Claude Sonnet 4 model confirms the applicability of the proposed solution. Results demonstrate that LLMs, when integrated with cybersecurity-aware geospatial data, can support dynamic UAV route planning and reduce the risk of availability violations caused by both physical interference and cyber intrusion.