跳转至

文章背景与核心概要

本文介绍了一种由大语言模型和迭代代码生成驱动的自进化科学智能体工作流,旨在自动化构建复杂物理系统的可解释白盒控制器。针对非线性流固耦合难题——即引导欠驱动的双关节狗鱼游动器穿过非定常尾流,该智能体能够在无需重新训练、重新调参或针对特定案例进行分支的情况下,自主设计出统一且数学上可读的控制策略,并在各种泛化场景中成功实现目标捕获。

该研究确立了一种具备可追溯性和物理推理能力的自动化控制设计新范式。通过将候选白盒控制器部署到物理仿真中,智能体能够根据多模态证据主动诊断动态行为,并将这些观测结果转化为逐步推进的源代码改进,从而在保持严格可解释性的同时,实现了复杂控制策略的自动化演进。


自进化科学智能体设计具备物理推理能力的白盒流体控制

摘要

This paper introduces a self-evolving scientific agent workflow driven by large language models and iterative code generation to automate the construction of interpretable whitebox controllers for complex physical systems. Tested on a non-linear fluid-structure interaction problem—navigating an underactuated two-joint dogfish swimmer through an unsteady wake—the agent autonomously designs a unified, mathematically readable control policy. Without requiring retraining, retuning, or case-specific branching, the final controller successfully achieves target capture across diverse generalization scenarios, establishing a traceable and physically-reasoned paradigm for automated control design.

本文介绍了一种由大语言模型和迭代代码生成驱动的自进化科学智能体工作流,用于自动化构建复杂物理系统的可解释白盒控制器。该方法在非线性流固耦合问题(即引导欠驱动的双关节狗鱼游动器穿过非定常尾流)上进行了测试,智能体自主设计了一种统一的、数学上可读的控制策略。无需重新训练、重新调参或针对特定案例进行分支,最终的控制器在各种泛化场景中均成功实现了目标捕获,为自动化控制设计建立了一种可追溯且具备物理推理能力的范式。


文档概览

  • arXiv ID: arXiv:2606.08405 [cs.AI]
  • 学科分类: 人工智能 (cs.AI); 流体力学 (physics.flu-dyn)
  • 作者: Boai Sun, Wenjin Guo, Zongmin Yu, Liu Yang
  • 提交时间: 2026年6月7日 (最后修订: 2026年8月25日, 版本 v3)

摘要正文

While data-intensive deep reinforcement learning can optimize complex control policies, scientific control design in physical systems fundamentally requires an interpretable chain of reasoning that connecting physical evidence to structured control architectures.

尽管数据密集型深度强化学习可以优化复杂的控制策略,但物理系统中的科学控制设计从根本上需要一条可解释的推理链,将物理证据与结构化控制架构连接起来。

Here, we present a self-evolving scientific agent workflow, driven by large language models and iterative code generation, that automates controller construction while preserving strict interpretability and rigorous physical reasoning. Instead of adjusting weights, the agent deploys candidate whitebox controllers into physical simulations, actively diagnoses dynamic behaviors from multimodal evidence, and translates these observations into progressive source-code refinements.

在此,我们提出了一种由大语言模型和迭代代码生成驱动的自进化科学智能体工作流,它能够在保持严格的可解释性和严密的物理推理的同时,实现控制器构建的自动化。智能体不是去调整权重,而是将候选白盒控制器部署到物理仿真中,从多模态证据中主动诊断动态行为,并将这些观测结果转化为渐进式的源代码改进。

We demonstrate this framework on a highly non-linear fluid-structure interaction problem: an underactuated, two-joint dogfish swimmer tasked with spatial target reaching in an unsteady flow using only joint angular accelerations. Starting from a target-blind propulsive seed, the agent autonomously designs and refines a unified controller that reaches a target embedded in an unsteady four-cylinder wake.

我们在一个高度非线性的流固耦合问题上展示了该框架:一个欠驱动的双关节狗鱼游动器,其任务是仅使用关节角加速度在非定常流中实现空间目标到达。从一个对目标无感(target-blind)的推进种子开始,智能体自主设计并优化了一个统一的控制器,该控制器能够到达嵌入在非定常四圆柱尾流中的目标。

Without retraining, retuning or case-specific branching, the retained controller achieves target capture across the full generalization test matrix, spanning variations in target position, rear-row geometry, cylinder count and inflow speed. The auditable evolution log reveals an emergent control architecture built upon travelling-wave propulsion, body-frame bearing guidance, phase-selective steering, corrective burst and adaptive relief.

在无需重新训练、重新调参或针对特定案例进行分支的情况下,保留下来的控制器在整个泛化测试矩阵(涵盖目标位置、后排几何形状、圆柱数量和来流速度的变化)中均实现了目标捕获。可审计的演进日志揭示了一个涌现的控制架构,该架构建立在行波推进、体坐标系方位引导、相位选择性转向、修正爆发(corrective burst)和自适应释放(adaptive relief)的基础之上。

Our results show that an autonomous scientific agent can successfully transform accumulated physical evidence into a robust, mathematically readable control policy, while maintaining a fully traceable process of scientific control design.

我们的结果表明,自主科学智能体能够成功地将积累的物理证据转化为稳健且数学上可读的控制策略,同时保持完全可追溯的科学控制设计过程。


访问与全文链接