1–6 Jun 2025
Taipei International Convention Center (TICC)
Asia/Taipei timezone

Towards autonomous accelerator control at ALS: a multi-agent LLM approach

THPS049
5 Jun 2025, 15:30
2h
Exhibiton Hall A _Salmon (TWTC)

Exhibiton Hall A _Salmon

TWTC

Poster Presentation MC6.T33 Online Modelling and Software Tools Thursday Poster Session

Speaker

Antonin Sulc (Helmholtz-Zentrum Berlin für Materialien und Energie)

Description

This paper demonstrates progress towards a novel approach to particle accelerator control at ALS using a decentralized multi-agent framework powered by Large Language Models (LLMs). Our distributed control system deploys specialized autonomous agents to manage critical accelerator subsystems while maintaining coordinated operation through LLM-driven communication protocols.

The system demonstrates fundamental capabilities essential for next-generation accelerator operations, given the flexible nature of agent specialization, this framework provides a robust platform for integrating and coordinating diverse control algorithms and approaches already established in accelerator operations. Our prototype shows the system's ability to autonomously diagnose and resolve basic operational issues using existing control infrastructure.

This work represents an advancement toward practical autonomous accelerator operation, establishing a scalable foundation for managing increasingly sophisticated accelerator configurations. The demonstrated success of this distributed control architecture opens new possibilities for improving operational efficiency across various accelerator facilities.

Region represented America
Paper preparation format LaTeX

Author

Antonin Sulc (Helmholtz-Zentrum Berlin für Materialien und Energie)

Co-authors

Raimund Kammering (Deutsches Elektronen-Synchrotron DESY) Thorsten Hellert (Lawrence Berkeley National Laboratory)

Presentation materials

There are no materials yet.