Autonomous experimentation enabled byartificial intelligence offers a new paradigm for accelerating scientificdiscovery. Nonequilibrium materials synthesis is emblematic of complex,resource-intensive experimentation whose acceleration would be a watershed formaterials discovery. We demonstrate accelerated exploration of metastablematerials through hierarchical autonomous experimentation governed by theScientific Autonomous Reasoning Agent (SARA). SARA integrates robotic materialssynthesis using lateral gradient laser spike annealing and opticalcharacterization along with a hierarchy of AI methods to map out processingphase diagrams. Efficient exploration of the multidimensional parameter spaceis achieved with nested active learning cycles built upon advanced machinelearning models that incorporate the underlying physics of the experiments andend-to-end uncertainty quantification. We demonstrate SARA’s performance byautonomously mapping synthesis phase boundaries for the Bi2O3 system, leadingto orders-of-magnitude acceleration in the establishment of a synthesis phasediagram that includes conditions for stabilizing d-Bi2O3 at room temperature, a criticaldevelopment for electrochemical technologies.

图 1.SARA 的闭环自主材料合成和发现周期。

图 2. 表征 AL 循环，加速采集相界检测所需的反射光谱。

图 3. 加速材料探索的 AL 合成回路。

图 4 选定迭代次数 n 时 Bi2O3 系统主动学习梯度相位图的演变。

Autonomous materials synthesis via hierarchical active learning of nonequilibrium phase diagrams