“As part of its reinforcement learning process, Physical Intelligence trains a value function to predict the probability of future success or failure from any given state in a task.”