Magnetic control of tokamak plasmas through deep reinforcement learning
Problem
Shaping and maintaining high-temperature plasma in a tokamak requires high-dimensional, high-frequency magnetic control that is difficult to design.
Why it matters
Stable plasma control is a core challenge on the path to fusion as sustainable energy.
What the AI did
A deep reinforcement learning architecture autonomously learned to command the full set of control coils, meeting high-level objectives under physical and operational constraints.
Results
Successfully produced and controlled a diverse set of plasma configurations; no quantitative metrics stated in abstract.
Impact
Shows RL-designed controllers can flexibly produce diverse plasma configurations with reduced design effort.
Record
| Venue | Nature |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Jonas Degrave, Federico Felici, Jonas Buchli, Michael Neunert, Brendan Tracey, F. Carpanese, Timo Ewalds, Roland Hafner et al. |
| Citations | 739 (OpenAlex, at import time) |
| References | 51 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-021-04301-9 |
| Open access | hybrid |
| Keywords | tokamak, plasma control, deep reinforcement learning, nuclear fusion, magnetic confinement |
Importance rating
significant — thousands to millions affected
Abstract frames magnetic confinement fusion as a promising path to sustainable energy and plasma shaping as a core challenge; controller design advance is significant but no scale of impact is quantified.
More in Physics & Astronomy
| # | Paper | Year | Citations | Importance |
|---|---|---|---|---|
| 1 | Solving the quantum many-body problem with artificial neural networks | 2017 | 2,205 |
7
|
| 2 | Learning high-accuracy error decoding for quantum processors | 2024 | 90 |
7
|
| 3 | Optimizing Quantum Error Correction Codes with Reinforcement Learning | 2019 | 158 |
7
|
| 4 | Machine-learning-assisted correction of correlated qubit errors in a topological code | 2018 | 141 |
7
|
| 5 | Quantum Error Correction with Quantum Autoencoders | 2023 | 36 |
7
|