PACMAN AI Controls DIII-D Fusion Plasma
Analysis based on 7 articles · First reported Sep 02, 2026 · Last updated Sep 15, 2026
The breakthrough is a positive but early-stage development for the fusion energy sector, potentially accelerating progress toward commercial fusion power and supporting long-term investment interest in fusion startups and related AI and energy technology companies. Near-term market impact is limited because the technology remains at the research stage and is far from commercial deployment.
Researchers at the United States — United States Department of Energy's United States — Princeton Plasma Physics Laboratory and Princeton University developed a new software framework called PACMAN (Prediction And Control using MAchiNe learning) that uses artificial intelligence to control fusion plasma in milliseconds, far faster than a human operator can react. The framework was successfully tested in five experiments on the DIII-D National Fusion Facility tokamak in San Diego. PACMAN combines multiple machine learning models in a repeating control loop that runs in about 20 milliseconds, gathering real-time measurements of temperature, density and magnetic signals, predicting plasma behavior, and issuing commands to heating systems, magnets and gas injectors within strict hardware safety limits. During the tests, PACMAN allowed a reinforcement-learning model to take full control of the heating systems, predicted edge bursts of energy, detected and controlled fast-particle-driven waves, adjusted plasma density and rotation, and predicted a tearing mode instability about 200 milliseconds in advance, preventing it before it started. It also coordinated all six of DIII-D's gyrotrons simultaneously, retargeting mirrors and adjusting power in real time. The design and initial results were published in the journal Nuclear Fusion. Researchers emphasize that humans remain in charge of setting objectives and reviewing results, and that the modular framework could be adapted to future tokamaks of different shapes and sizes. The work was supported by the DOE Office of Science and the United States — National Science Foundation Graduate Research Fellowship.
Set up alerts, explore entity relationships, search across thousands of events, and build custom intelligence feeds.
Open Dashboard