SAGE-LLM: Towards Safe and Generalizable LLM Controller with Fuzzy-CBF Verification and Graph-Structured Knowledge Retrieval for UAV Decision

Researchers propose SAGE-LLM, a novel framework that combines Large Language Models with Control Barrier Functions for safe UAV autonomous decision-making. The system addresses LLM safety limitations through formal verification mechanisms and graph-based knowledge retrieval, demonstrating improved safety and generalization in drone control scenarios.

• →SAGE-LLM introduces a train-free two-layer architecture combining LLMs with formal safety verification for UAV control.
• →The framework uses fuzzy Control Barrier Functions to provide provable safety certification for LLM-generated actions.
• →A star-hierarchical graph-based retrieval system enables efficient scene adaptation without online training.
• →Experimental validation shows maintained performance with significantly enhanced safety in pursuit-evasion scenarios.
• →The framework demonstrates strong extensibility potential for broader embodied intelligence and safety-critical control systems.