Deep reinforcement learning controls fusion plasma in the TCV tokamak
DeepMind and EPFL (Nature, Feb 2022) trained a single deep-RL policy in simulation that commanded all of TCV's magnetic control coils on the real machine. It produced and held elongated, negative-triangularity and 'snowflake' plasmas, and even two separate 'droplet' plasmas at once.
Key facts
- Nature 602 (Feb 2022)
- Zero-shot sim-to-real transfer: trained in a simulator, deployed directly on the tokamak
- One neural controller replaced a set of hand-designed feedback loops for 19 magnetic coils
Science result
- Field
- physics / nuclear fusion / plasma control
- Problem
- Magnetic confinement and shaping of tokamak plasmas
- Result
- First deep-RL controller to shape and sustain diverse plasma configurations on a real tokamak.
- AI system
- deep reinforcement learning (MPO actor-critic)
- Human role
- Humans built the simulator, specified targets and rewards, supervised experiments
- Verification
- Peer-reviewed in Nature; demonstrated on hardware
- Status
- confirmed
What happened
A neural network learned to steer a hot plasma by adjusting magnetic coils thousands of times per second, first in simulation and then on the real reactor.
Why it matters
It showed that RL could replace complex hand-engineered control in fusion devices and opened the way to AI-designed plasma scenarios.
Changelog
- 2026-09-29: created
Related events
- AI controller predicts and avoids tearing instabilities in the DIII-D fusion reactor ★★★
- PPPL's PACMAN framework lets multiple AI models control a tokamak in ~20 ms, preventing a tearing mode ★★
- DeepMind's Deep Loop Shaping cuts LIGO control noise 30–100× ★★★
- Google DeepMind partners with Commonwealth Fusion Systems to optimise and control the SPARC tokamak with AI ★★★
Sources (2)
- paperMagnetic control of tokamak plasmas through deep reinforcement learning (Nature)
- officialDeepMind: Accelerating fusion science through learned plasma control
id: 2022-02-16-deepmind-tokamak-plasma-control · updated 2026-09-29 · open in the interactive timeline