code-as-policies

This paper presents RECENT, a framework that enables efficient skill grounding in embodied agents using small language models (sLMs) by refactoring code-based skills rather than regenerating them from scratch, achieving performance comparable to LLM-based methods.