Grounding Large Language Models In Embodied Environment With Imperfect World Models

Dataset Analysis

GLIMO uses proxy world models (simulators) to generate training data via an LLM agent with self-refinement, improving LLM performance on embodied tasks.