Grounding LLMs in Scientific Discovery via Embodied Actions
This addresses the challenge for researchers and engineers in scientific fields where LLMs need to interact dynamically with simulations, though it appears incremental as it builds on existing software and methods.
The paper tackles the problem of LLMs struggling to connect theoretical reasoning with physical simulation in scientific discovery by introducing EmbodiedAct, a framework that grounds LLMs in embodied actions with a perception-execution loop, resulting in SOTA performance with improved reliability, stability, and accuracy in tasks like engineering design and scientific modeling.
Large Language Models (LLMs) have shown significant potential in scientific discovery but struggle to bridge the gap between theoretical reasoning and verifiable physical simulation. Existing solutions operate in a passive "execute-then-response" loop and thus lacks runtime perception, obscuring agents to transient anomalies (e.g., numerical instability or diverging oscillations). To address this limitation, we propose EmbodiedAct, a framework that transforms established scientific software into active embodied agents by grounding LLMs in embodied actions with a tight perception-execution loop. We instantiate EmbodiedAct within MATLAB and evaluate it on complex engineering design and scientific modeling tasks. Extensive experiments show that EmbodiedAct significantly outperforms existing baselines, achieving SOTA performance by ensuring satisfactory reliability and stability in long-horizon simulations and enhanced accuracy in scientific modeling.