Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Google DeepMind and EPFL researchers trained an AI controller in simulation, then tested it on the TCV experimental tokamak in Lausanne. The system controlled magnetic coils to shape plasma—including two separate plasma droplets at once—but the result was a plasma-control demonstration, not a demonstration of fusion power or net energy.
How does AI control plasma in a fusion reactor?
A tokamak uses magnetic fields to confine hot plasma. In the 2022 experiment, a deep reinforcement-learning controller learned how to adjust those fields by commanding magnetic coils. The work was carried out by Google DeepMind and EPFL’s Swiss Plasma Center and reported in a peer-reviewed Nature paper published February 16, 2022.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Heavy Industry: Tokamak Fusion Plant | $22.95 | Buy on Amazon |
Train in simulation, then test on the machine
The researchers first trained the controller through interaction with a tokamak simulator. They then tested the learned control on EPFL’s Tokamak à Configuration Variable (TCV), a research tokamak in Lausanne, Switzerland. Simulation made it possible to train without relying solely on limited, costly machine experiments; the controller still had to be validated on the real device. EPFL’s account of the collaboration describes the simulator as based on more than 20 years of research and continuously updated.
One controller coordinated magnetic coils
DeepMind describes the demonstrated architecture as one neural network that took sensor inputs and control targets and issued voltage commands to all 19 magnetic coils. Its account contrasts this with TCV’s existing arrangement of separate controllers for the coils. This describes the experiment’s controller design; it does not establish that one network can replace control systems on every tokamak. DeepMind’s explanation also reports that TCV plasma experiments could last up to three seconds, followed by roughly 15 minutes for cooling and reset. Those are operational details for TCV as reported by DeepMind, not universal limits for tokamaks.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
- Miniatures Pack
What plasma shapes did the experiment control?
The original work tested a range of plasma shapes and configurations, including elongated shapes, negative triangularity and snowflake plasmas. A particularly unusual demonstration sustained two separate plasma droplets simultaneously in the vessel. These results showed that the controller could handle varied magnetic-shaping tasks on TCV, rather than only one target shape. The Nature paper reports the experiments and their configurations.
Did Google AI achieve fusion power?
No. The experiment addressed magnetic control of plasma in a fusion research device. It did not demonstrate commercial electricity generation, nor did it report net energy from the AI-controlled experiment. The result matters as a possible tool for fusion research: better plasma control could help researchers explore configurations, but control performance is distinct from producing usable fusion power.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What changed in the 2024 follow-up?
In a March 1, 2024 summary of “Towards Practical Reinforcement Learning for Tokamak Magnetic Control,” DeepMind described follow-up work intended to address drawbacks relative to traditional feedback control, including shape accuracy, steady-state error and the time needed to learn new tasks. The summary separates simulation metrics from experimental validation:
- Simulation: the upgraded approach achieved up to 65% better shape accuracy and reduced training time for new tasks by at least a factor of three.
- Simulation: the work also reported a substantial reduction in long-term plasma-current bias.
- TCV experiment: DeepMind says upgraded reinforcement-learning controllers were tested on TCV. The cited summary does not present the 65% accuracy or training-time figures as experimental TCV measurements.
DeepMind also says it released TORAX in May 2024, an open-source simulator that models the plasma core and predicts changes in temperature, density and electric current. TORAX is research software, not evidence of deployment in a commercial fusion reactor. The 2024 results extend the research-control story; they do not establish commercial-reactor use.
Quick Recap
What the results do—and do not—establish
- Established: a reinforcement-learning controller trained in simulation could be tested on TCV and control multiple plasma configurations.
- Established: a later controller was tested on TCV, while specific accuracy and training-time gains reported in the 2024 summary were simulation results.
- Not established: that the controller can transfer unchanged to other tokamaks, replace their control systems, or operate a commercial fusion plant.
- Not established: that this experiment generated net energy or electricity for the grid.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
