This research introduces a novel framework that allows robots to leverage LLMs for decision-making while maintaining provable safety guarantees through reachability analysis.

• Creates a safety filter layer that validates LLM outputs before execution
• Implements formal verification techniques to constrain robot actions within safe operating boundaries
• Demonstrates real-world validation in autonomous navigation scenarios
• Achieves both flexibility of LLM intelligence and rigorous safety assurance

This breakthrough addresses a critical gap in deploying AI-powered robots in safety-critical environments, offering a pathway to utilize advanced language models without compromising operational safety standards.

Safe LLM-Controlled Robots with Formal Guarantees via Reachability Analysis