LLM-Based Autonomous Drone and Vehicle Navigation

AUTODRONENAV proposes an LLM-guided planning pipeline for physical DJI Mini 3 Pro drones and ground vehicles. The system uses a multi-agent architecture with Planner and Executor LLM agents (Gemini and GPT-4o) to translate high-level instructions into structured JSON mission plans. It integrates a local perception pipeline for real-time object detection and depth estimation, executed via a custom Android application. After simulation validation, physical flight tests demonstrated dynamic mission planning and target tracking, with 90% of test queries successfully planned.