Skip to main content
QUICK REVIEW

[论文解读] Grounding LLMs in Scientific Discovery via Embodied Actions

Bo Zhang, Jinfeng Zhou|arXiv (Cornell University)|Feb 24, 2026
Machine Learning in Materials Science被引用 0
一句话总结

EmbodiedAct 将 LLMs 绑定到具体现实动作中,在 MATLAB 的紧密感知-执行循环中实现连续监控和自主纠错,以便在长期科学仿真与建模中实现可靠性。

ABSTRACT

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.

研究动机与目标

  • 在过程导向的科学任务中弥合理论推理与可验证物理验证之间的差距。
  • 将现有科学软件转化为主动的具身代理,耦合执行与持续感知。
  • 证明运行时感知与热修复引导的重新规划能在长期仿真中提升可靠性、稳定性与准确性。
  • 展示在 MATLAB 与开源后端以及多种模型族中的普适性。
  • 提供一个可重复使用的框架与协议,用于科学-focused LLM 应用中的紧密感知-行动循环。

提出的方法

  • 提出 EmbodiedAct,这是一个闭环认知体系结构,将意图转化为可执行的仿真原语。
  • 将 LLMs 基于四个模块进行绑定:策略规划器、原语生成器、运行时感知引擎、反思决策者。
  • 使用异步状态同步协议来维持来自仿真环境的实时、流式观测。
  • 通过拓扑推理将高层科学意图翻译为软件特定的原语(如 MATLAB 求解器、Simulink 图操作)。
  • 实现对仿真轨迹与中间状态的实时监控,以检测异常并触发自主热修复。
  • 引入双环控制:快速内部环用于感知驱动的热修复,慢速外部环用于由物理信息反馈引导的重新规划。
Figure 1 : Comparison of EmbodiedAct with existing paradigms. EmbodiedAct integrates executable simulation primitives and continuous runtime perception, endowing the agent with the capacity for embodied action within physical simulation environments.
Figure 1 : Comparison of EmbodiedAct with existing paradigms. EmbodiedAct integrates executable simulation primitives and continuous runtime perception, endowing the agent with the capacity for embodied action within physical simulation environments.

实验结果

研究问题

  • RQ1具备实时感知的具身代理在面向过程的科学任务中,是否能超越传统的代码即行动的 LLM 代理?
  • RQ2运行时感知如何在跨工程与物理科学的长期仿真中影响可靠性、稳定性与准确性?
  • RQ3领域特定原语与拓扑感知的执行是否比通用代码生成能提升解的质量?
  • RQ4感知-行动循环对在 MATLAB 与开源仿真后端上的泛化有何影响?
  • RQ5双环体系结构(内部热修复环 + 外部反思规划器)对复杂系统是否有可观的收益?

主要发现

  • EmbodiedAct 在 EngDesign 与 SciBench-107 基准测试中达到业内领先水平,在多种模型族上超过生成式与基于代码的基线。
  • 主动的运行时感知实现对中间状态的实时监控并触发热修复,提升长期仿真的可靠性并减少故障模式。
  • 领域特定原语与对意图的拓扑感知翻译为仿真动作,尤其在数值敏感领域如 Mathematics 与 Physics 中提高了准确性。
  • 该方法可在后端(MATLAB 与开源)之间通用,持续保持性能优势,显示出强健的跨环境适用性。
  • 开源模型在与 EmbodiedAct 搭配时获得显著提升,缩小与专有模型的差距,凸显该框架的支架作用。
  • 消融研究确认运行时感知是性能的关键驱动因素,感知禁用的变体接近 CodeAct 的性能。
Figure 2 : Overview of EmbodiedAct, which bridges the LLM agent and simulation environment via the Asynchronous State Sync Protocol. EmbodiedAct orchestrates a fast inner loop driven by the Runtime Perception Engine to trigger immediate Hot-Fixes, and a slow outer loop driven by the Reflective Decis
Figure 2 : Overview of EmbodiedAct, which bridges the LLM agent and simulation environment via the Asynchronous State Sync Protocol. EmbodiedAct orchestrates a fast inner loop driven by the Runtime Perception Engine to trigger immediate Hot-Fixes, and a slow outer loop driven by the Reflective Decis

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。