The integration of artificial intelligence has advanced the development of intelligent microrobots. Yet, these robots often operate in complex, dynamic, and highly uncertain environments, requiring specialized guidance and control strategies. To address this challenge, this paper proposes a large language model (LLM)–integrated framework that uses speech recognition to capture user instructions in natural language and an LLM to interpret them for precise microrobot navigation. The system adopts a modular architecture consisting of voice input, speech‐to‐text conversion, LLM pseudocode generation, closed‐loop execution, and interruption handling. Key challenges addressed include handling linguistic uncertainties, correcting speech recognition errors, preventing LLM hallucinations, and enabling real‐time interruption handling for safe operation. The results demonstrate the feasibility and reliability of this architecture in achieving precise microrobot navigation via natural language. By evaluating several state‐of‐the‐art LLMs using the same set of prompts, the ChatGPT‐5 model achieved the best performance (with an 88.89% success rate and a confidence score of 1.2, the highest among all tested models), highlighting the potential of LLMs as high‐level interpreters for natural language–driven microrobot control.
Zhu et al. (Mon,) studied this question.