LLM-assisted Semantic Option Discovery for Facilitating Adaptive Deep Reinforcement Learning

Researchers have developed a new framework that combines Large Language Models (LLMs) with Deep Reinforcement Learning to improve data efficiency, interpretability, and cross-environment transferability. The approach uses LLMs to map natural language instructions into executable rules and create semantically annotated options for better skill reuse and constraint monitoring.

• →New LLM-driven framework addresses key challenges in Deep Reinforcement Learning including low data efficiency and limited transferability.
• →The system maps natural language instructions into executable rules for better interpretability and control.
• →Framework enables semantic-driven skill reuse across similar environments through automated option creation.
• →Experiments on Office World and Montezuma's Revenge demonstrate superior performance in data efficiency and cross-task transferability.
• →Integration of LLMs with symbolic planning shows promise for safer and more compliant AI behavior in practical applications.