BashMocha/Prompting-LLMs-for-Aerial-Navigation

Implements few-shot prompting evaluation across multiple LLM providers (ChatGPT, Gemini, Mistral, Claude) to generate navigation code executed in Microsoft's AirSim simulator, with automated trajectory recording and comparison against ground-truth paths. The architecture extracts model-generated source code directly from prompts and validates performance across three defined aerial navigation test cases—obstacle avoidance, pathfinding, and basic flight—using trajectory deviation metrics. Includes curated prompt sets and reference datasets for reproducible benchmarking of LLM capabilities in robotic control tasks.

No commits in the last 6 months.