Engineering & computing · OECD FoS 2.7 · China MoE 0827 核科学与技术
Energy & fusion
能源与核聚变
L2Local breakthroughs ↑ Rising Score 1.55 · Last reviewed 2026-10-22 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2022-02-16 AI leapField · major
Deep RL magnetic control of tokamak plasmas (Nature) ↗
First deep-RL control of all tokamak coils
Why this level
Deep RL controls tokamak plasmas (Nature 2022) and AI avoids tearing instabilities (Nature 2024); DeepMind and CFS use TORAX to plan SPARC operations.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
- nature.com/articles/s41586-021-04301-9
- nature.com/articles/s41586-024-07024-9
- deepmind.google/blog/bringing-ai-to-the-next-generation-of-fusion-energy
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
Plasma controlFusion digital twinsGrid forecasting
Papers
- Enabling integrated AI control on DIII-D: a control system design with state-of-the-art experiments
AI: modular framework enforces safety limits before actuation, run in five DIII-D experiments. Field: peer-reviewed engineering path for combining AI controllers on a major tokamak
2026-07-02 · Nuclear Fusion
- Reconstruction-free magnetic control of DIII-D plasma with deep reinforcement learning
RL control validated on DIII-D
2025-06-16 · arXiv
- TORAX: A Fast and Differentiable Tokamak Transport Simulator in JAX
Fast simulator used to plan SPARC
2024-06-10 · arXiv
- Avoiding fusion plasma tearing instability with deep reinforcement learning
Predicts and avoids tearing instabilities ~300 ms ahead
2024-02-21 · Nature
-
AI leap
Magnetic control of tokamak plasmas through deep reinforcement learningFirst deep-RL control of all tokamak coils
2022-02-16 · Nature
- Reinforcement Learning Controllers for Fusion Scenario Development
A single goal-conditioned RL controller replaces per-scenario tuning in tokamak scenario development.
2026-10-19 · arXiv · Evgeny Adishchev, Maxim Nurgaliev, Dmitry Sorokin, Alexander Granovsky
- Reconstruction of Unobserved Core Electron Density in Tokamak Plasmas with a Neural Operator Transport Surrogate and Uncertainty Quantification
A Fourier neural operator transport surrogate reconstructs unobserved core electron density in tokamaks with uncertainty.
2026-10-17 · arXiv · D. Sorokin, A. Granovskiy, M. Stokolesov, E. Adishchev
- ModeCast: One-Shot Prediction of Linear MHD Eigenmodes and Growth Rates in Tokamaks
ModeCast, trained on 45,000+ JOREK runs, predicts full linear MHD stability for n=1–4 modes.
2026-10-16 · arXiv · Alex Higginbottom, Zara Z. Ercan, Abetharan Antony, Guzman Sanchez Gonzalez
- Multiple Instance Learning for Interpretable Plasmoid Detection in the FLARE Device
Multiple-instance learning detects plasmoids in FLARE reconnection experiments with interpretable timing.
2026-10-16 · arXiv · Charles Marsden, Adam Robbins, Hantao Ji
- Physical-state supervision for design-only Gaussian process surrogates of implosion simulations
A Gaussian-process surrogate of implosion simulations supervised by physical diagnostics learned from design variables.
2026-10-09 · arXiv · Chang Liu, Hao Shen, Chuanying Li
- A neural characteristic mapping method: Lagrangian PINNs based on flow maps for transport-dominated problems
Lagrangian PINNs approximating flow maps for transport-dominated problems.
2026-10-08 · arXiv · Martin Campos Pinto, Victor Fournet, Emmanuel Franck, Philipp Krah
- WEST advanced wall protection achievements toward long pulse operation
WEST monitors plasma-facing components with indicators including ML-based processes to avoid hot spots.
2026-10-08 · arXiv · R. Mitteau, M. H. Aumeunier, L. Dubus, J. Gerardin
- Quantitative schlieren imaging of a laser-ionized plasma channel in atomic vapor using symbolic regression
Symbolic regression yields compact formulas linking schlieren signals to plasma-channel parameters in the AWAKE experiment.
2026-10-01 · arXiv · Gabor Demeter
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
Milestones 7· 1 leaps
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Policy
CFS leads DOE Genesis Mission phase-2 AI digital twin for SPARCNational programme puts AI twins into SPARC operations planning
Commonwealth Fusion SystemsVendor claim -
System
Enabling integrated AI control on DIII-D (PACMAN): real-time ML control with safety limits, ~20 ms cycle (Nuclear Fusion)AI: modular framework enforces safety limits before actuation, run in five DIII-D experiments. Field: peer-reviewed engineering path for combining AI controllers on a major tokamak
Nuclear Fusion -
Paper
Reconstruction-free magnetic control of DIII-D plasma with deep RLRL control validated on DIII-D
arXivPreprint -
System
TORAX: fast differentiable tokamak transport simulatorFast simulator used to plan SPARC
arXivPreprint -
Paper
AI avoids tearing instabilities in fusion plasma (Nature)Predicts and avoids tearing instabilities ~300 ms ahead
Nature -
PaperAI leap
Deep RL magnetic control of tokamak plasmas (Nature)First deep-RL control of all tokamak coils
Nature
No milestones match this filter.
More papers & repos 5
| Type | Title | Date |
|---|---|---|
| Paper | Reconstruction-free magnetic control of DIII-D plasma with deep reinforcement learning Preprint | 2025-06-16 |
| Paper | TORAX: A Fast and Differentiable Tokamak Transport Simulator in JAX Preprint | 2024-06-10 |
| Paper | Avoiding fusion plasma tearing instability with deep reinforcement learning | 2024-02-21 |
| Paper | Magnetic control of tokamak plasmas through deep reinforcement learning | 2022-02-16 |
| Repo | google-deepmind/torax To be verified | — |
Latest news 6
- US restarts three shut-down nuclear reactors to feed AI data centersTom's Hardware AIen
- CFS to lead AI-enabled digital twin fusion project in second phase of Genesis MissionCommonwealth Fusion Systemsen
- The U.S. Nuclear Plant Fleet Is Leaning Into AIIEEE Spectrumen
- Forecasting space weather risks on power gridsMicrosoft Researchen