Model-free and Bayesian Ensembling Model-based Deep Reinforcement Learning for Particle Accelerator Control Demonstrated on the FERMI FEL

TL;DR

This paper compares model-free and model-based deep reinforcement learning methods for particle accelerator control, demonstrating that model-based approaches offer higher sample efficiency and better representational power, with practical implementation on the FERMI FEL system.

Contribution

It introduces a DYNA-style model-based reinforcement learning algorithm with uncertainty awareness for accelerator control, and compares it to a deep Q-learning approach, highlighting their respective advantages.

Findings

Model-based RL shows higher sample efficiency.

Model-free RL achieves slightly better asymptotic performance.

Both methods exhibit increased noise robustness in accelerator control.

Abstract

Reinforcement learning holds tremendous promise in accelerator controls. The primary goal of this paper is to show how this approach can be utilised on an operational level on accelerator physics problems. Despite the success of model-free reinforcement learning in several domains, sample-efficiency still is a bottle-neck, which might be encompassed by model-based methods. We compare well-suited purely model-based to model-free reinforcement learning applied to the intensity optimisation on the FERMI FEL system. We find that the model-based approach demonstrates higher representational power and sample-efficiency, while the asymptotic performance of the model-free method is slightly superior. The model-based algorithm is implemented in a DYNA-style using an uncertainty aware model, and the model-free algorithm is based on tailored deep Q-learning. In both cases, the algorithms were implemented in a way, which presents increased noise robustness as omnipresent in accelerator control problems. Code is released in https://github.com/MathPhysSim/FERMI_RL_Paper.