Autonomous AI Design Unlocks a Faster Route to Custom Materials

Researchers often know what they want a material to do — conduct electricity, withstand heat, respond to stimuli or display a specific color — but turning those goals into a working chemical formula can take months or years of trial and error. The process is especially challenging for polymers, whose performance can change dramatically and unpredictably with even small tweaks to their molecular building blocks or how they are combined.

Researchers at the U.S. Department of Energy’s (DOE) Arogonne National Laboratory, the University of Chicago, and Purdue University have now demonstrated a faster path: An autonomous inverse-design workflow that helps scientists go from a target property to a polymer recipe with far fewer experiments. Autonomous workflow combines artificial intelligence, machine learning, and robotics to rapidly create polymers with precise, customizable properties.

The approach connects three pieces that are often separate in traditional research. First, it gathers prior knowledge by automatically extracting data from published scientific papers, including information that appears in both text and images. To do this, the team used AI “reading” tools, including large language models, to scan papers and pull out the details researchers normally collect by hand. These AI tools can identify and organize information buried in paragraphs, tables, and even images.

Argonne Scientist Jie Xu, also an Assistant Professor at the University of Chicago Pritzker School of Molecular Engineering.

Second, the approach uses machine learning to predict which combinations of building blocks are most likely to produce the desired result. Third, predictions are sent directly into an automated laboratory workflow that can synthesize the polymers, purify them, prepare samples, measure their properties, and feed the results back to improve the next round of predictions.

This workflow utilized Polybot, a self-driving laboratory platform housed in the Center for Nanoscale Materials, a DOE Office of Science user facility at Argonne. Designed to coordinate robots and instruments through an AI-driven system, Polybot allows experiments to run continuously with minimal human intervention. Polybot was not just used to automate a known procedure but also to carry out an inverse-design loop where each experiment is chosen to move closer to the desired result.

By turning materials design into a faster, more systematic process, this work points toward a future where scientists can request a property and rapidly receive a tailored recipe, ready for further development and real-world applications.

A vibrant array of electrochromic polymer samples showcases precise color tuning achieved through AI-driven inverse design, highlighting advancements in materials for smart display technologies. (Image: Argonne National Laboratory)

In this interview, Argonne Scientist Jie Xu, also an Assistant Professor at the University of Chicago Pritzker School of Molecular Engineering, discusses the results of this research.

Tech Briefs: What is the key difference between inverse design and forward design with respect to materials development? Can you explain with an example?

Jie Xu: Inverse design is a process that works backward from a target property to determine the optimal formulation. Instead of synthesizing a material and then testing its properties, one begins with the specific wanted properties and identify the best structure or composition to achieve them. For example, our team aimed to develop an electrochromic polymer with a precise target color. Rather than relying on trial-and-error, our Polybot system took the target color as input and automatically determined the building blocks and ratios among them required to synthesize the polymer.

Tech Briefs: What specific limitations of traditional polymer design does inverse design overcome, and where have you seen the most dramatic gains in speed or efficiency so far?

Xu: Traditionally, polymer design takes a massive amount of time that relies heavily on extensive trial-and-error tests experimentation to develop new materials. Because the development process is based on empirical testing, even minor modifications to a polymer’s structure can lead to significant changes in its properties, making optimization challenging and inefficient. This limitation can be addressed through inverse design, which leverages design rules derived from machine learning applied to literature data mining and well-designed physical experiments, substantially reducing development cycles.

Tech Briefs: What are the biggest technical or data-related challenges in ensuring that AI-guided predictions translate reliably into real, reproducible materials?

Xu: The two biggest challenges are data quantity and data quality. A robust AI prediction model typically requires a large amount of training data, but generating such dataset in the lab is time-consuming and labor-intensive. Small-data ML approaches can help address this challenge, such as physics-informed models. High data quality is another critical requirement that strongly influences model accuracy and consistency. However, obtaining high-quality data is non-trivial, especially when it originates from diverse sources.

Tech Briefs: Which real-life applications such as smart windows, electronics, or energy materials are best positioned to benefit from this inverse-design workflow first, and why?

Xu: Based on our in-house evaluation, smart windows are currently one of the most promising application areas. This is because conventional development requires substantial effort to identify optimal formulations of color-changing materials. Using our new AI-enabled workflow, this process can be reduced to less than a single day. We believe transferring this approach to electronics or energy materials will also yield significant benefits. In fact, similar approaches have been explored in the broader material science community.

Tech Briefs: How do you plan to scale this approach to move beyond a single application and what are your next steps in this research?

Xu: While we demonstrated this concept using color-changing polymers, the resulting AI models, along with our established autonomous laboratory (the Polybot lab) and workflow, can be readily extended to other material systems and applications due to the generality of the models and the flexible, modular design of the Polybot lab. We are currently working on several new projects, including plastics recycling, reusable energy materials, and smart electronics.

Tech Briefs: As more of the design– synthesis–testing loop becomes autonomous, how do you see the role of human intuition and expertise evolving in materials research labs?

Xu: As robots take over the heavy-lifting and routine tasks in the lab, researchers will have more time for high-level thinking. Humans will focus on setting ambitious goals, defining research scope, and developing more effective ways to guide AI systems. At the same time, researchers will need to become comfortable working side-by-side with AI tools.

Tech Briefs: With AI expediting design and development, what’s the future outlook for data-driven manufacturing of advanced materials?

Xu: Materials discovery is expected to accelerate significantly, and the field must prepare for a more open and collaborative mindset. Researchers will need to adapt their workflows and decision-making processes to incorporate AI. To make this future a reality, it will also be important to share software and data as open-source resources to continue advancing scientific progress.

Other contributors to this work include Doga Ozgulbas, Subramanian Sankaranarayanan, Maria Chan and Qiaomu Yang from Argonne; Yukun Wu and Jianing Zhou from Argonne and Purdue University; Zhiyang Wang from Purdue University; and Anna Österholm and John Reynolds from Georgia Institute of Technology. Shiyu Hu, Rafael Vescovi, and Aikaterini Vriza were at Argonne when this research was conducted.

For more information, visit here  or contact Jie Xue at This email address is being protected from spambots. You need JavaScript enabled to view it.; This email address is being protected from spambots. You need JavaScript enabled to view it..