AI system anticipates fusion disruptions milliseconds ahead
Princeton researchers developed PACMAN, an AI framework that monitors and adjusts fusion plasma in real time, predicting instabilities about 200 ms before they occur. In tests on a tokamak, it successfully prevented damaging disruptions. The system allows human operators to set objectives while AI handles rapid responses.
The PACMAN framework, detailed in *Nuclear Fusion*, was validated across five separate trials on a working tokamak. It coordinates multiple machine learning models that share outputs, addressing the previous fragmentation of AI control efforts.
Traditional physics simulations require days or months, rendering them useless for millisecond-scale reactions. By contrast, PACMAN's integrated models operate in real time, allowing human operators to define overarching goals while the AI executes the rapid, necessary adjustments to maintain plasma stability.
This advancement could accelerate the timeline for practical fusion energy by solving the critical bottleneck of real-time plasma control. If successfully scaled, it may reduce the need for human intervention in hazardous high-speed environments, potentially lowering operational costs and improving reactor reliability. Ultimately, society could benefit from a more stable path toward clean, abundant baseload electricity, though significant engineering hurdles remain before commercial deployment.