Self-Evolving Scientific Agent Designs Physically Reasoned White-Box Fluid Control
Abstract
While neural networks excel in autonomous control, their black-box nature makes control decisions difficult to interpret and diagnose in dynamic fluids. Here, we show how self-evolving scientific agents can design explicit, neural-network-free white-box controllers by iteratively interpreting simulation evidence, accumulating control knowledge and refining controller code. We demonstrate this approach on an underactuated two-joint swimmer navigating unsteady flows via joint angular accelerations. Starting from a target-blind propulsive controller, the agent gradually constructs key mechanisms, including travelling-wave propulsion, body-frame guidance, phase-selective steering, redirect bursts and adaptive relief. The resulting controllers reach targets and generalize across changes in target position, wake geometry, cylinder count, and inflow speed without revision. Moreover, 2D control priors transfer successfully to accelerate 3D adaptation. Our work demonstrates that self-evolving agents can autonomously design physically reasoned and generalizable white-box fluid control, showing a promising paradigm beyond traditional reinforcement learning and black-box neural network control.
Related
- ADAPT for Science: A Multi-Agent System for Self-Evolving Scientific Optimization
- Older adults’ evaluations of smart home control screen designs: Appearance dimensions and behavioral intentions in a multi-stage study
- Non-Invasive Identification of Blue-And-White Porcelain: Comparison of Vietnamese, Yunnan and Jingdezhen Wares Using Reasoned Visualization and PCA
- Modeling and simulation of time dependent ion migration process and bending of ion-doped rubber films using a minimal set of physically estimated parameters
- A complete quantum process model for physically realistic photo-detection in the joint photon-number and spectral degrees of freedom
BibTeX
@misc{sun2026selfevolving,
title = {{Self-Evolving Scientific Agent Designs Physically Reasoned White-Box Fluid Control}},
author = {Boai Sun and Wenjin Guo and Zongmin Yu and Liu Yang},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.08405},
eprint = {2606.08405},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.08405},
}
RIS
TY - UNPB
TI - Self-Evolving Scientific Agent Designs Physically Reasoned White-Box Fluid Control
AU - Boai Sun
AU - Wenjin Guo
AU - Zongmin Yu
AU - Liu Yang
PY - 2026
DA - 2026/09/09
PB - arXiv
DO - 10.48550/arxiv.2606.08405
UR - https://arxiv.org/abs/2606.08405
AB - While neural networks excel in autonomous control, their black-box nature makes control decisions difficult to interpret and diagnose in dynamic fluids. Here, we show how self-evolving scientific agents can design explicit, neural-network-free white-box controllers by iteratively interpreting simulation evidence, accumulating control knowledge and refining controller code. We demonstrate this approach on an underactuated two-joint swimmer navigating unsteady flows via joint angular accelerations. Starting from a target-blind propulsive controller, the agent gradually constructs key mechanisms, including travelling-wave propulsion, body-frame guidance, phase-selective steering, redirect bursts and adaptive relief. The resulting controllers reach targets and generalize across changes in target position, wake geometry, cylinder count, and inflow speed without revision. Moreover, 2D control priors transfer successfully to accelerate 3D adaptation. Our work demonstrates that self-evolving agents can autonomously design physically reasoned and generalizable white-box fluid control, showing a promising paradigm beyond traditional reinforcement learning and black-box neural network control.
ER -