Home▸Plasma Physics▸AI Plasma Shape Control (Tokamak)

AI Plasma Shape Control (Tokamak)

Reinforcement-learning-style feedback control of a tokamak's plasma cross-section: watch a PD controller drive poloidal-field coil currents to hold elongation and triangularity against a vertical-instability drift, or switch it off and watch the plasma run away toward the wall.

Plasma Physics3DAdvanced60 FPS

Real tokamaks don't just heat plasma — they have to hold its cross-sectional shape (elongation, triangularity, vertical position) against a genuine instability, tens of thousands of times a second, using currents in a ring of external magnetic coils. This simulation renders a 3D tokamak vessel with a Miller/D-shape plasma boundary swept around the torus, drives it with the same vertical instability that makes elongated plasmas hard to control, and lets a proportional–derivative feedback controller — the same job DeepMind and EPFL's TCV reinforcement-learning controller was trained to do — fight to hold the shape on target. Turn the AI gain to zero and watch the same plasma drift into the wall; turn it back up, change the target shape or disturbance level, and watch the controller re-stabilize it in real time.

⚙ Under the hood

A PD feedback controller drives poloidal-field coil currents to hold a tokamak plasma's elongation, triangularity and vertical position against a real vertical instability — turn the AI gain to zero and watch the same plasma drift into the wall.

tokamakplasma physicsreinforcement learningfeedback controlfusion energymagnetic confinement

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install