Quantitative control and recording of materials-synthesis processes using an automated experimentation platform

The quest to build better materials—stronger alloys, more efficient batteries, or smarter catalysts—has long relied on a slow, human process of trial and error. Scientists mix chemicals, heat them, and observe the results, hoping to stumble upon a combination that works. But human hands are not perfectly steady, and human descriptions like "mix slowly" or "add until cloudy" are open to interpretation. When one researcher says "slowly," another might mean something entirely different. This lack of precision makes it difficult to repeat experiments exactly, which is a major hurdle for the modern field of data-driven materials science. To teach computers how to predict new materials, researchers need vast amounts of high-quality data where every single step of the experiment is recorded with exact numbers. If the data is fuzzy, the computer's predictions will be fuzzy too. The challenge, then, is not just to build machines that can mix chemicals, but to build systems that can measure and record the invisible details of the process that humans often overlook.

In a recent study, a team of researchers at Tohoku University in Japan tackled this problem by building a simple, open-source platform designed to automate the mixing of chemical solutions with extreme precision. Instead of aiming for a fully autonomous robot that can design and run any experiment from scratch, they focused on a more practical goal: creating a reliable system that can perform a specific task over and over again while recording every tiny detail of how it was done. They combined off-the-shelf tools, such as robotic arms and electric pipettes, with custom-made 3D-printed parts and a camera system. A key innovation was the use of an artificial intelligence agent to write the computer code that controls the robots. This allowed the system to be easily adapted to different setups without needing a team of expert programmers. The researchers tested this platform by synthesizing ZIF-8, a type of metal-organic framework, which is a porous material useful for gas storage and catalysis. The goal was to see if the machine could control the mixing process better than a human and, more importantly, if it could reveal how specific, hard-to-measure factors affect the final product.

The experiment involved mixing two liquid solutions to create the ZIF-8 material. In a traditional lab, a scientist might use a manual pipette to add the second solution to the first. The speed at which they squeeze the pipette bulb is usually a matter of feel, described in lab notebooks only as "slow" or "fast." In this automated system, the researchers used electric pipettes that could be set to nine distinct speed levels, ranging from a very slow pour taking about ten seconds to a rapid dump taking less than one second. They ran the synthesis multiple times, alternating between the slowest, a medium, and the fastest settings. The results were striking. While the chemical composition of the final white powder was the same in every run, the size of the particles changed dramatically depending on how fast the liquid was added. When the solution was added slowly, the resulting particles were larger. When it was added quickly, the particles were smaller. This relationship was so consistent that the researchers could predict the particle size simply by knowing the speed setting used during the mix.

What made this discovery possible was the system's ability to quantify a variable that is usually ignored. In manual experiments, the speed of addition is rarely measured or recorded; it is just a vague part of the procedure. Because it wasn't tracked, its effect on the material was hidden. The automated platform, however, treated the dispensing speed as a precise number, just like temperature or weight. By recording this number every time, the researchers could prove that the speed of mixing was a dominant factor in determining the final size of the particles. They confirmed this by measuring the particle sizes with specialized equipment and found that the trend held true across repeated runs. The system also captured images of the mixture as it formed, showing how the white suspension settled differently depending on the speed, providing a visual record that matched the numerical data. This level of detail revealed that the "slow" and "fast" instructions in old lab manuals were actually controlling the fundamental physics of how the crystals formed, a detail that was previously lost in translation.

The study also demonstrated that this level of control and recording could be achieved without building a complex, expensive super-robot. The entire setup was constructed using commercially available components and 3D-printed holders, with all the design files and computer code made freely available to other scientists. This approach lowers the barrier for other laboratories to adopt similar automation. By sharing the blueprints and the software, the researchers hope to encourage a shift in how experiments are conducted. Instead of relying on the intuition of a single expert, the scientific community can move toward standardized, highly reproducible processes where every variable is known and recorded. The researchers noted that while the system did not replace human judgment entirely, it successfully handled the repetitive, precision-critical tasks that humans struggle to perform consistently. This cooperation between human oversight and machine precision allows for the collection of the kind of high-quality data needed to train artificial intelligence models to discover new materials faster and more reliably than ever before.

Ultimately, the work shows that the path to better materials may not lie in more complex robots, but in better data. By turning vague instructions into exact numbers, the researchers uncovered a direct link between how a chemical is mixed and what the final material looks like. They proved that the speed of adding a liquid is not just a minor detail, but a powerful lever for controlling the properties of a material. This finding suggests that many other materials might have hidden dependencies on process details that have been overlooked for decades. With a platform that can record these details automatically, scientists can begin to build a comprehensive library of how materials are made, paving the way for a future where the design of new substances is guided by precise, reproducible data rather than guesswork. The study concludes that the future of materials science depends on capturing the full story of the experiment, not just the final result, and that simple, transparent automation is the key to writing that story clearly.