Fine-Tuning Large Language Models for Autonomous Tactical Deconfliction of Small Unmanned Aerial Systems

This paper presents a simulation-to-language data generation pipeline using BlueSky ATM for fine-tuning Large Language Models on autonomous tactical deconfliction of small UAS, achieving improved decision accuracy and reduced near mid-air collisions through parameter-efficient fine-tuning (LoRA, GRPO).

This manuscript is currently under preparation for presentation at ATRD 2026.