Training a model to imitate expert demonstrations by learning to predict the same actions given the same observations.