(169cz) AI Based Exploration on Synthesizable Space for Autonomous Laboratory

Autonomy laboratories have been recently conducted widely for accelerating material development. In traditional research, humans directly define the initial experimental conditions, which are constraints for solving such optimization problems, thereby slowing down the development of materials. Furthermore, under the initial experimental conditions as defined synthesizable space found by humans, it may not be possible to find the global optimal. Therefore, we solve the problem of requiring a human to set the initial experimental conditions by developing a model that breaks the hardware constraints and explores the domain using active learning. Our research demonstrated that superior performance compared to other baseline models in an unexplored synthesizable area before conducting experiments, enabling efficient exploration in fewer experiment iterations. This study proves the justification of applying exploration on synthesizable space which can applicability to new environments and experimental conditions, so that our work verified the conventional autonomous lab even more flexible.