Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces EmbodiedAct, a framework that grounds LLMs in embodied actions within scientific software to improve their ability to perform scientific discovery tasks. EmbodiedAct transforms scientific software like MATLAB into an active embodied agent with a tight perception-execution loop, allowing the LLM to perceive and react to transient anomalies during simulations. Experiments on engineering design and scientific modeling demonstrate that EmbodiedAct achieves state-of-the-art performance by improving reliability, stability, and accuracy compared to existing passive "execute-then-response" approaches.
LLMs can now actively perceive and react to anomalies during scientific simulations, leading to more reliable and accurate results in complex engineering and modeling tasks.
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.
Tsinghua AI