废水处理测试:AI 智能体连接实验规划与机器人控制

继续使用 Google、Facebook 或 Apple,即表示您同意使用条款。
自适应 AI 智能体在废水处理测试中将实验规划与机器人控制相连接
作者:Ingrid Fadelli,Phys.org
编辑:Lisa Lock,审阅:Robert Egan
本文已按照 Science X 的编辑流程与政策进行审阅。编辑在确保内容可信度的同时,重点强调了以下特征:
添加到首选来源
具身科学智能体(Embodied Science Agent,ESA)的架构,这是一种面向自主实验室的认知—具身混合范式,包含三个组成部分。来源:arXiv(2026)。DOI: 10.48550/arxiv.2611.12958
人工智能系统如今已广泛用于研究场景,主要用于分析数据、做出预测,以及发现有前景的材料、药物或解决方案。用于研究的智能体大多没有具身形态,也就是说它们基于计算机运行,需要人类或独立的自动化系统来实际执行实验。
最近,一些计算机科学家一直在探索开发既能规划又能执行科学实验的具身 AI 智能体的可能性。这类智能体通常将 AI 智能体与能够与周围环境交互的机器人系统相结合。
伦敦大学学院(UCL)、香港科技大学(广州)及其他机构的研究人员最近开发出一种用于规划和开展科学实验的新智能体,名为具身科学智能体(embodied science agent,ESA)。在一篇发布到 arXiv 预印本服务器的论文中,ESA 能够自主规划并运行涉及染料去除的废水处理实验。
此前开发的大多数用于科学发现的具身智能体,都依赖 AI 智能体来规划实验、依赖机器人系统来执行实验。但作者认为,其中许多智能体在将实验规划与物理反馈,以及与能够控制科学仪器的习得技能紧密结合起来方面,仍存在局限。
“当前架构将模型视为无实体的规划器,输出符号化工作流,交由各自独立的机器人系统执行,这从根本上阻碍了具身实验直觉的形成。”Yihang Chen、Zhongwei Yu 及其同事在论文中写道。“我们认为,要克服这一悖论,就需要沿着三条相互依存的轴线嵌入可学习性:认知可学习性(学习研究什么)、实验可学习性(学习在哪里搜索)以及具身可学习性(学习如何行动)。”
一个能开展实验的具身智能体
研究人员开发的这个新智能体包含三个主要部分:一个数字智能体、一个具身智能体,以及连接两者的接口。该 AI 智能体利用现有的科学数据和早期实验记录来决定下一步研究什么。
为了筛选出有前景的实验条件,该智能体依赖一种名为贝叶斯优化的数学方法。AI 智能体得出的信息被用于自适应地规划具身机器人系统的动作,同时考虑仪器位置的差异或实验环境中的其他变化。最后,具身统一接口连接非具身智能体与具身智能体,将前者制定的计划与后者执行的动作协调起来。
作者写道:“我们提出了 ESA,它通过三个协同组件将数字智能与物理具身统一起来:一个数字智能体,通过基于记忆的推理和贝叶斯优化提供认知与实验层面的可学习性;一个具身智能体,通过可泛化的仪器控制提供具身层面的可学习性;以及一个具身统一接口,实现无缝协调。”
Chen、Yu 及其同事通过让该系统自主规划和执行废水处理实验来对其进行测试。该智能体从超过 2 亿种可能组合中挑选实验条件,并利用每次实验的结果来指导后续测试。
ESA 连续开展了 15 轮废水处理实验。在另一项测试中,团队评估了该具身系统能否可靠地处理放置在不同位置的电极架。
Chen、Yu 及其同事写道:“经过 15 轮连续迭代,ESA 将最终溶液的透光率从 82% 提升至 99.7%,相当于残余染料浓度降低 98.5%,并接近完全脱色;同时,其习得的视觉运动策略能够泛化到不同的仪器摆放位置,这是迈向用于自主科学发现的下一代 SDL 的一个有前景的进展。”
初步结果与未来可能
作者开发的该智能体被成功用于规划和执行废水处理实验,并利用先前的结果来指导后续测试。研究人员发现,在 15 轮实验中,该智能体找到了能够生成更澄清溶液的处理条件。
该智能体似乎还能适应周围环境的变化,即使仪器被移动到不同位置,也能成功完成实验。值得注意的是,研究人员的测试衡量的是脱色程度,而脱色本身并不能证明染料被完全分解或其他污染物被去除。
尽管如此,这项研究的结果表明,开发更具适应性的科研智能体是可能的,这类智能体能够在实验室环境中既规划实验又执行实验。未来,该团队的这一系统有望得到改进,并在源自不同科学学科的更广泛实验中接受测试。
本文由我们的作者 Ingrid Fadelli 撰写,Lisa Lock 编辑,Robert Egan 进行事实核查与审阅——这篇文章是细致人工工作的成果。我们依赖像您这样的读者,让独立的科学新闻得以存续。如果这篇报道对您有意义,请考虑捐赠(尤其是按月捐赠)。作为感谢,您将获得一个无广告账户。
Yihang Chen 等,《具身科学智能体:让具身实验设计变得可学习》,arXiv(2026)。DOI: 10.48550/arxiv.2611.12958
期刊信息:arXiv
这篇报道的背后是谁?
自由记者,心理学学士、国际新闻学硕士。自 2018 年起报道 AI、机器人技术、神经科学和天体物理学。完整简介 →
艺术史学士,物质文化硕士。曾任博物馆编辑、急救员和移植协调员。自 2021 年起为 Science X 担任编辑。完整简介 →
数学生物学学士,创意写作硕士。游历广泛,对科学与语言有独特视角。完整简介 →
引用:自适应 AI 智能体在废水处理测试中将实验规划与机器人控制相连接(2026 年 11 月 20 日),2026 年 11 月 20 日检索自 https://techxplore.com/news/2026-11-ai-agent-experimental-robotic-wastewater.html
本文件受版权保护。除为私人学习或研究目的进行的任何合理处理外,未经书面许可,不得复制本文件的任何部分。内容仅供参考。
智能体式 AI 机器人帮助科学家与机器人对话,加速实验进程
“电脑,我的实验完成了吗?”研究人员探讨在研究中运用 AI 智能体
人工智能“Coscientist”实现科学发现自动化
无需睡眠或进食,AI 构建的“科学家”快速设计出针对 SARS-CoV-2 变异株的纳米抗体
AI 科学家自主生成并验证新的生物学发现
强强联合:两种金属与一种墨水如何重塑印刷储能
开源微控制器将传感、存储与无线工具集于一体,面向可穿戴设备与机器人
从仙人掌中收集电能
六英寸机器人有望检查常规摄像头无法触及的核废料管道
脑研究人员让 LLM 承受与中风患者相似的“损伤”,以解析 AI 语言机制
微量的锰对下一代电池影响巨大
获取即时摘要文本(Gist)
一个具身智能体利用贝叶斯优化和机器人控制开展了 15 次废水染料去除实验,将溶液透光率从 82% 提升至 99.7%。机器人控制能够适应仪器摆放位置的变化。脱色并不能证明染料被完全降解,也不能证明其他污染物被去除。
本摘要由 LLM 自动生成。完整免责声明
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(Adaptive AI agent connects experimental planning with robotic control in wastewater treatment test)
By continuing with Google, Facebook or Apple, you agree to the Terms of Use.
November 20, 2026 feature
Adaptive AI agent connects experimental planning with robotic control in wastewater treatment test
by Ingrid Fadelli, Phys.org
edited by Lisa Lock, reviewed by Robert Egan
This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:
Add to Preferred Sources
The architecture of the Embodied Science Agent (ESA), a hybrid cognitive–embodied paradigm for self-driving laboratories with three components. Credit: arXiv (2026). DOI: 10.48550/arxiv.2611.12958
Artificial intelligence systems are now widely used in research settings, primarily to analyze data, make predictions and discover promising materials, drugs or solutions. Most of the agents used for research are disembodied, which means that they are computer-based and require humans or separate automated systems to physically perform experiments.
More recently, some computer scientists have been exploring the possibility of developing embodied AI agents that can both plan and execute scientific experiments. These agents typically combine AI agents with robotic systems that can interact with their surroundings.
Researchers at University College London (UCL), the Hong Kong University of Science and Technology (Guangzhou) and other institutions recently developed a new agent for planning and conducting science experiments, called embodied science agent (ESA). Presented in a paper posted to the arXiv preprint server, ESA was able to autonomously plan and run wastewater treatment experiments involving dye removal.
Most previously developed embodied agents for scientific discovery rely on AI agents to plan experiments and robotic systems to execute them. However, the authors argue that many of them remain limited in how closely they connect the planning of experiments with physical feedback and learned skills enabling the control of scientific instruments.
"By treating models as disembodied planners that output symbolic workflows for separate robotic systems, current architectures inherently block the development of embodied experimental intuition," wrote Yihang Chen, Zhongwei Yu and their colleagues in their paper. "We argue that overcoming this paradox requires embedding learnability along three interdependent axes: cognitive learnability (learning what to investigate), experimental learnability (learning where to search) and embodied learnability (learning how to act)."
An embodied agent that runs experiments
The new agent developed by the researchers has three primary components: a digital agent, an embodied agent and an interface connecting them. The AI agent uses available scientific data and records of earlier experiments to decide what to investigate next.
To select promising experimental conditions, the agent relies on a mathematical approach called Bayesian optimization. The information derived by the AI agent is used to plan the actions of the embodied robotic system adaptively, accounting for differences in the position of instruments or other variations in the experimental environment. Finally, the embodied unified interface connects the disembodied and embodied agents, coordinating the plans produced by the first with the actions performed by the second.
"We introduce the ESA, which unifies digital intelligence with physical embodiment through three synergistic components: a Digital Agent that provides cognitive and experimental learnability via memory-based reasoning and Bayesian optimization, an Embodied Agent that provides embodied learnability via generalizable instrument control, and an Embodied Unified Interface for seamless coordination," wrote the authors.
Chen, Yu and their colleagues tested their system by asking it to autonomously plan and execute wastewater treatment experiments. The agent selected experimental conditions from more than 200 million possible combinations, using each experiment's results to guide subsequent tests.
ESA carried out 15 consecutive rounds of wastewater treatment experimentation. In a separate test, the team assessed whether the embodied system could reliably handle an electrode rack placed in different positions.
"Through 15 continuous rounds, ESA raised the final solution transmittance from 82% to 99.7%, corresponding to a 98.5% reduction in residual dye concentration and near-complete decolorization, while its learned visuomotor policies generalized across varying instrument placements, a promising step toward next-generation SDLs for autonomous scientific discovery," wrote Chen, Yu and their colleagues.
Initial results and future possibilities
The agent developed by the authors was successfully used to plan and carry out wastewater treatment experiments, using earlier results to guide subsequent tests. The researchers found that across the 15 experimental rounds, the agent identified treatment conditions that produced clearer solutions.
The agent also appeared to adapt to changes in its surroundings, successfully performing experiments even when instruments were moved to different positions. Notably, the researchers' tests measured decolorization, which does not by itself establish the complete breakdown of dye or the removal of other pollutants.
Nonetheless, the results of this study demonstrate the possibility of developing more adaptive agents for scientific research, which can both plan and run experiments in laboratory settings. In the future, the team's system could be improved and tested on a wider range of experiments rooted in different scientific disciplines.
Written for you by our author Ingrid Fadelli, edited by Lisa Lock, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You'll get an ad-free account as a thank-you.
Yihang Chen et al., Embodied Science Agents: Making Embodied Experimental Design Learnable, arXiv (2026). DOI: 10.48550/arxiv.2611.12958
Journal information:arXiv
Who's behind this story?
Freelance journalist with BSc Psychology and MA International Journalism. Covers AI, robotics, neuroscience, and astrophysics since 2018. Full profile →
BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021. Full profile →
Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →
© 2026 Science X Network
Citation: Adaptive AI agent connects experimental planning with robotic control in wastewater treatment test (2026, November 20) retrieved 20 November 2026 from https://techxplore.com/news/2026-11-ai-agent-experimental-robotic-wastewater.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
Agentic AI bot helps scientists speak to robots, speeding up experiments
Computer, is my experiment finished? Researchers discuss the use of AI agents in their research
Artificially intelligent 'Coscientist' automates scientific discovery
With no need for sleep or food, AI-built 'scientists' quickly design nanobodies against SARS-CoV-2 variants
AI scientist autonomously generates and validates new biological discoveries
Better together: How two metals and one ink are reshaping printed energy storage
Open-source microcontroller combines sensing, storage and wireless tools for wearables and robots
Harvesting electricity from prickly pear cacti
Six-inch robot could inspect nuclear waste pipes beyond reach of conventional cameras
Brain researchers afflict LLMs with 'injuries' similar to those of stroke victims to unpack AI language mechanisms
A little manganese makes a big difference for next-generation batteries
Get Instant Summarized Text (Gist)
An embodied AI agent used Bayesian optimization and robotic control to run 15 wastewater dye-removal experiments, raising solution transmittance from 82% to 99.7%. Robotic control adapted to changes in instrument placement. Decolorization did not establish complete dye degradation or removal of other pollutants.
This summary was automatically generated using LLM. Full disclaimer
This site uses cookies to assist with navigation, analyse your use of our services, collect data for ads personalisation and provide content from third parties. By using our site, you acknowledge that you have read and understand our Privacy Policy and Terms of Use.
原文为英文,正文由自动抓取并保留原文。
原文发布于 TechXplore(Adaptive AI agent connects experimental planning with robotic control in wastewater treatment test),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。
