A Modular Framework for AI-Driven, Multimodal Synchrotron Experiments

Multimodal characterization, which combines information from different measurement techniques, often greatly benefits material studies. However, these campaigns can be challenging due to the experimental logistics and analysis expertise required to integrate multiple methods. Typically, the multimodal aspects of materials research can only be considered after measurements are complete, during the combined analysis to derive deeper insights. Recently, we demonstrated AI-driven simultaneous multi-beamline experiments using combined synchrotron total scattering and X-ray Absorption Fine Structure (XAFS) measurements. This world-first achievement opens new opportunities for human/AI collaborative operational modes, where AI algorithms work alongside facility users to enhance scientific output.<br/>Using the Bluesky data acquisition platform, we conducted fully and partially autonomous coordinated experiments in real-time across spatially distant beamlines with asynchronous measurements at vastly different collection rates. An ensemble of AI agents is crucial for these autonomous experiments, processing both raw data and results from scientific analysis (e.g., Rietveld refinement, XAFS modeling) from asynchronous multimodal measurements to drive real-time decision-making. This highly extensible framework offers plug-and-play capabilities, allowing AI agents for data reduction, analysis, and decision-making to be easily swapped for optimal solutions. Additionally, we have enabled <i>in silico</i> experiments, providing an environment for offline testing of AI agent performance. We will discuss the design and implementation of this AI-driven experiment framework and demonstrate the enhanced efficiency for sample exploration through several use cases.