Discrete-time total stability: learning robust flux-profile controllers for tokamak reactors

Abstract: Recent advances in Artificial Intelligence-based control provide a suitable framework for plasma profile control. However, most of these learning algorithms lack fundamental robustness and generalization guarantees. This impairs their applicability outside of simulation. In this talk, we present novel total stability results for discrete-time nonlinear systems, allowing the study of stability properties via model comparison. These theoretical foundations justify the use of Deep Reinforcement Learning algorithms to train controllers in a simulated environment, and their subsequent application in real-world scenarios. We present experimental results showing the effectiveness of such learned agents equipped with an error integrator in robustly controlling the safety factor profile in Tokamak reactors. 

Bio: Samuele Zoboli received the B.S. degree in electronics engineering from the University of Modena and Reggio Emilia, Italy, in 2016 and the M.S degree with honors in automation engineering from University of Bologna, Italy, in 2019. He currently is a Ph.D. candidate at LAGEPP, University of Lyon 1, France. His research interests include stabilization of discrete-time nonlinear systems, multi-agent systems, reinforcement learning and control-applied artificial intelligence.