Keith A. Brown (Boston University), John Dunlap (UES, Inc. and Air Force Research Laboratory), Robert Epps (National Renewable Energy Laboratory), Jason Hattrick-Simpers (University of Toronto) and Kiran Vaddi (University of Washington)
How to Build A Self-Driving Lab
Written by Sophia Chen
John Dunlap's PhD research could be tedious. The chemist, now working with UES, Inc., and Air Force Research Laboratory (AFRL) contractor in Ohio, was developing polymer-coated quantum dots for biomedical applications at the University of South Carolina. The synthesis process was partially automated, but on many days, he would have to sit around and wait to press a button every 15 minutes. To test the samples, he would have to walk down three floors to use the instruments. And then he would do this over and over again, zeroing in on the recipe he sought. "It was a lot of blood, sweat, and tears on my end," he told the crowd during Thursday's Symposium X - MRS/The Kavli Foundation Frontiers of Materials.
By the time Dunlap obtained his PhD degree in 2022, materials science researchers had begun adopting new automation strategies that pushes the exhausting repetitive laboratory work to robots and computers. During the panel discussion on “How To Build A Self-Driving Lab,” Dunlap, along with Keith A. Brown of Boston University, Jason Hattrick-Simpers of the University of Toronto, Kiran Vaddi of the University of Washington, and Robert Epps of the National Renewable Energy Laboratory, discussed their experiences developing and using laboratory techniques that do not require any human intervention.
Not all lab processes are suitable for full automation. It may not be worth it to automate experiments that are too short or too long, said Brown. “There’s a sweet spot in terms of experiment length,” he says.
In 2016, Brown’s group began developing a self-driving lab based on 3D-printing for making and testing mechanical materials that can efficiently absorb energy. These materials might be useful in designing helmets or the crumple zone of a car, for example. Brown’s system consists of six 3D printers arranged in a circle with a robotic arm in the middle, along with instruments that can weigh and perform compression tests on the 3D-printed elements. They’ve named it Mama Bear, which stands for “Mechanics of Additively Manufactured Architectures Bayesian Experimental Autonomous Researcher.”
Brown’s research team tasked Mama Bear with finding a 3D-printed structure with optimal energy absorption efficiency. “Over the span of about two years of continuous study, we’re able to find structures that exceeded the limits that have been previously found,” he said.
Dunlap, collaborating with AFRL, now has a fully automated continuous flow setup for synthesizing small molecules and modifying polymers. This self-driving setup can explore solid state reactions and photochemical reactions, among others, and perform NMR measurements for testing. The researchers are moving toward updating the setup to be capable of high-throughput experiments.
The panelists emphasized the importance of using modular machines that run on open-source software. This allows researchers to assemble and customize components to make a self-driving lab for their scientific needs. Vaddi cautioned that the commercially available machines he uses for designing colloidal nanoparticles are becoming increasingly less open-source and more expensive. “We are trying to move away from these highly expensive systems and build low-cost modular hardware,” he said.
Hattrick-Simpers talked about how to assemble a team with the necessary skills and mindset to build these self-driving labs. People can sometimes have a “trust barrier” to automation, and it’s crucial that team members fully buy into the concept. Otherwise “you're not going to make a lot of progress,” he said.
In addition, building a self-driving lab requires an interdisciplinary team with “broad range and skill set,” said Epps. Hattrick-Simpers advised researchers to be realistic about what a single person can do. “You can’t expect one person to build the AI … and have the bandwidth to become a subject matter expert,” he said.
Several panelists have moved beyond simply showing that their systems work. Brown and Hattrick-Simpers are developing user facilities at Boston University and the University of Toronto, their respective institutions.
Symposium X—MRS/The Kavli Foundation Frontiers of Materials features lectures aimed at a broad audience to provide meeting attendees with an overview of leading-edge topics.