A research team has shown for the first time that reinforcement learning—i.e., a neural network that learns the best action to perform at each moment based on a series of rewards—allows autonomous vehicles and underwater robots to locate and carefully track marine objects and animals.