Autonomous Alignment at Synchrotron Facilities Using Bayesian Optimization

Autonomous alignment will be an indispensable component of autonomous experimentation at synchrotron facilities, as beamlines drift over time and must be constantly re-aligned. Bayesian optimization is a machine learning-based algorithm well-suited for high-dimensional, expensive-to-sample, and potentially noisy optimization problems, and it has been successfully implemented to autonomously align beamlines at several facilities. However, there are beamline-specific obstacles that can hinder the robustness and efficiency of these efforts, meaning that most are tailored to a narrow context of optimization problems. In this talk, I outline these obstacles and their solutions, and show that it is possible to construct an adaptable, generalized framework for autonomous alignment that performs well at many different kinds of beamlines across different facilities. I outline the application of Bayesian optimization to different optimization problems at several light sources (ALS, APS, NSLS-II, LCLS). I discuss the prospect of a unified, collaborative approach to beamline automation.