Controlling the nuclear fusion plasma in a tokamak with reinforcement learning