AI Learns to Sculpt Superhot Fusion Fuel, A Big Step for Clean Energy
Scientists used a special type of AI, called reinforcement learning, to control the superheated gas inside a fusion reactor. This AI learned to shape and stabilize the plasma, which is the fuel for fusion energy, in ways that are very difficult for human-designed systems to maintain. This breakthrough could help make fusion power, a clean energy source, more practical.
Better control over fusion fuel brings us closer to a future powered by clean, virtually limitless fusion energy.
Confining plasma at fusion temperatures requires constant, millisecond adjustments to dozens of magnetic coils. Researchers trained a reinforcement-learning agent in simulation, then deployed it on a real tokamak where it controlled the plasma's shape and position directly.
The learned controller sustained elongated and droplet-like plasma configurations and reduced the engineering effort normally needed to design bespoke control laws for each scenario.
Fusion remains hard, but autonomous AI control of the plasma itself removes one persistent obstacle on the path to sustained net-energy reactions.
🔬 Read the technical version
Reinforcement learning holds fusion plasma in shapes human controllers struggle to sustain
A control policy trained with reinforcement learning steered the magnetic coils of a tokamak in real time, sculpting and stabilizing the superheated plasma into configurations that are difficult to maintain with conventional controllers.