UAVGENT：一种语言引导的分布式控制框架

This research proposes a language-guided control framework for multi-drone systems, aimed at executing evolving high-level missions while maintaining formal robustness guarantees at the physical layer. The framework features a three-layer architecture: the first layer involves a human operator issuing natural language instructions; the second layer consists of an LLM-based supervisor that periodically interprets, verifies, and corrects the commanded tasks in the context of the latest state and target estimates; the third layer is a distributed inner-loop controller that tracks the resulting reference using only local relative information. The study also derives a theoretical guarantee characterizing tracking performance under bounded disturbances and piecewise-smooth references with discrete jumps induced by LLM updates.