Research

Materials science relies on processes such as powder handling, mixing, coating, and emulsification that are common yet demand great skill. Handling unpredictable, nonlinear materials like powders and non-Newtonian fluids requires both precise robotic control and the tacit knowledge — intuition, skill, and experience — of expert practitioners. We aim to robotize and automate these challenging operations and complex processes, while studying the underlying techniques from first principles to build ultra-compact, high-speed autonomous experimental systems and fully autonomous measurement systems capable of "ultimate mixing" and "ultimate grinding" beyond human ability.

What makes an experiment good? Where should we measure? When should we stop? Such judgments have traditionally relied on the experience of skilled researchers. We are building the theoretical foundations of autonomous experimentation, including Bayesian experimental design that incorporates prior knowledge, optimal experimental design for rational measurement selection, and optimal stopping for deciding when to end an experiment. We are also developing physics-based mathematical models for optimal measurement and process control, based on the idea that automated data analysis requires mechanisms grounded in physical models.

We analyze large-scale data obtained from quantum-beam measurements such as synchrotron radiation and neutrons using AI and machine learning, mapping out the landscape of material space to reveal properties that were previously invisible. Building on these insights, we also develop physical and mathematical models for inverse design — working backward from desired properties to discover new materials — contributing to the creation of next-generation materials.