Orchestrating agentic systems at ALS beamline 5.3.1 with Osprey and Bluesky.

Authors

Ka Hung Chan, Yang Ha, Antoine Islegen-Wojdyla, Seij De Leon, Manuel Garces, Dylan McReynolds, Xiaoya Chong, Raja Vyshnavi Sriramoju, Johannes Mahl, Damian Guenzing, Wiebke Koepp, Alexander Hexemer, Thorsten Hellert, Tanny Chavez

Published in

Journal of synchrotron radiation. Sep 01, 2026. Epub Sep 01, 2026.

Abstract

Modern synchrotron beamlines support increasingly complex experiments, but their operation remains strongly dependent on facility-specific control interfaces and workflows. This reliance creates a significant barrier to entry for new users and limits the portability of experimental procedures across beamlines. Here, we present a minimal, agent-agnostic orchestration architecture for AI-assisted beamline operation, deployed at Advanced Light Source beamline 5.3.1. The approach introduces a capability-based abstraction layer that exposes beamline functionality as a set of predefined operations. Within this framework, AI agents translate natural-language user requests into structured experimental plans composed of these capabilities. All plans are executed through the Bluesky Queue Server, ensuring deterministic operation within existing control and safety constraints without modification of the underlying beamline control system. The framework is defined by three key elements: (i) a portable capability layer that decouples user intent from beamline-specific implementation, (ii) a hybrid interaction framework combining AI-assisted workflow composition with graphical user interface (GUI)-based monitoring and control, and (iii) constrained execution that restricts all actions to predefined capabilities and requires user approval prior to execution. We demonstrate the architecture using grazing-incidence scattering (GISAXS), multi-edge X-ray absorption spectroscopy (XANES), and cross-beamline deployment scenarios. In each case, the system generates physically meaningful experimental plan, adapts to local hardware constraints, and preserves experimental intent across different instruments. These results demonstrate a practical and scalable approach for integrating AI into synchrotron experimentation, enabling intent-driven, portable, and safe beamline operation.

PMID:
42677900
Bibliographic data and abstract were imported from PubMed on 01 Sep 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.