A Localized Regulatory-Aware Intelligent Assistant Framework Based on Large Language Models for Compliant UAV Operations

Consumer drones are widely used for recreational aerial photography, while cross-border operations face inconsistent regulations, fragmented official information, and changing restricted-airspace requirements. This paper proposes LUCA, a localized regulatory-aware intelligent assistance framework based on Large Language Models (LLMs). LUCA acquires regulatory and airspace information, constructs a multidimensional structured Retrieval-Augmented Generation (RAG) knowledge base, and transforms natural-language provisions into computable conditions through a constrained reasoning loop for pre-flight compliance review. On a locked benchmark of 1000 scenarios drawn from United States FAA, Chinese civil-aviation, and EASA regulatory sources, each configuration was independently run five times with the same prompts and retrieval snapshot. LUCA-Qwen3-8B achieved a five-run mean Accuracy of 95.4%; the prespecified primary run classified 954/1000 cases correctly, with 37 false positives and 9 false negatives. The framework supports traceable pre-flight review, but its output is advisory and does not constitute a legal determination or official flight authorization or official flight authorization.