A team has shown 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.