About this document

ROBOGUARD: Safety for LLM Robots by tye46 is a document available to read on EtoBox.

The paper presents ROBO GUARD, a two-stage safety architecture designed to address the unique safety challenges posed by large language model (LLM)-enabled robots, particularly in dynamic environments. It contextualizes safety rules and resolves conflicts between these rules and potentially unsafe plans, significantly reducing the execution of unsafe plans from 92% to below 2.5%. The proposed system is shown to be robust against adversarial attacks while maintaining performance on safe plans, highlighting i

Author

tye46

Language

EN