Phys.org — Physics · September 7, 2026

PACMAN AI framework for controlling fusion systems safely makes key decisions in milliseconds

PhysicsElectrical EngineeringComputer Science

THE AI ANGLE

Predicting plasma instabilities and driving real-time tokamak actuators

Researchers at Princeton University and the Princeton Plasma Physics Laboratory developed PACMAN, a modular AI framework that executes real-time tokamak control loops in roughly 20 milliseconds. Successfully validated across five experiments on the DIII-D tokamak, the system uses multiple machine learning models to forecast disruptions—such as predicting tearing modes 200 milliseconds in advance—and adjust plasma heating systems while strictly enforcing hardware safety limits. This architecture demonstrates how integrated machine learning can replace slow, traditional simulations with millisecond-scale prediction and control in complex cyber-physical environments.

THE TEACHING ANGLE

Instructors can explore the architectural tension between unconstrained machine learning optimization and deterministic safety-limit enforcement in safety-critical, millisecond-scale feedback control systems.

Read the original at phys.org Generate teaching or study materials

Instructors get discussion guides, assignments, and mini-cases. Students and readers get a plain summary, class prep, and an exercise. All built from the full article. Three are free with an account.

PACMAN AI framework for controlling fusion systems safely makes key decisions in milliseconds

More in Physics