2023 R&D 100 Award for Physics-Informed, Active Learning–Driven Autonomous Microscopy for Science Discovery

Researchers from ORNL and the University of Tennessee, Knoxville, have created a physics-informed, active learning method for autonomous experiments. This software suite is composed of active learning algorithms, control software for microscopes and other experimental tools that expedite scientific discovery.

The advancement of microscopy has transformed the way scientists and researcher study materials and biological systems. However, there are multiple challenges in developing autonomous microscopy including automating microscopy (such as automating data acquisition and transfer protocols), development of task-specific machine learning methods, understanding the interplay between physics discovery and machine learning and end-to-end definition of workflows.

The research addresses current challenges in enabling autonomous microscopy workflows by balancing the required physical intuition, prior knowledge of scientists and experimental goals with machine learning algorithms capable of translating these to specific experimental protocols.

Funding for this project was provided by DOE Office of Science.

Co-development was led by ORNL’s Yongtao Liu and Maxim Ziatdinov with Sergei Kalinin from UT Knoxville. ORNL’s Kevin Roccapriore, Rama Vasudevan, Kyle Kelley and Stephen Jesse contributed to development.