Autonomous Bayesian Optimization for Physics-Guided Discovery in Functional Materials

Adaptive experimentation is transforming materials research by enabling real-time decision-making informed by machine learning and domain physics. We present a framework that integrates scanning probe microscopy with multi-objective Bayesian optimization (MOBO) to autonomously explore compositional and structural landscapes in functional materials. The framework models each measurable property such as domain morphology, magnetic contrast, or surface correlation length as an objective within a Gaussian process surrogate, while physics-based acquisition functions guide the selection of new measurement sites along a Pareto frontier of competing targets. This approach dynamically balances exploration and exploitation, capturing interdependencies between structure, composition, and function. Applied to composition-spread magnetic thin films of , it efficiently reconstructs property landscapes and reveals coupled morphological–magnetic behavior, including domain coarsening and magnetic ordering trends. The method establishes a generalizable strategy for closed-loop, physics-informed discovery across imaging, spectroscopy, and synthesis, bridging explainable AI and experimental automation.