This research introduces a hybrid framework combining LLMs with formal verification to ensure robot safety with mathematical guarantees.

• Develops a reachability analysis method that verifies if robot trajectories remain within safe operational bounds
• Creates a safety filter that redirects unsafe LLM commands while preserving intended functionality
• Demonstrates effectiveness through real-world experiments with mobile robots navigating complex environments
• Achieves 100% collision avoidance while maintaining task performance

This breakthrough addresses critical safety concerns in autonomous systems, enabling more reliable deployment of LLM-controlled robots in industrial, healthcare, and consumer applications where failures could have serious consequences.

Safe LLM-Controlled Robots with Formal Guarantees via Reachability Analysis