AggreVaTe is a model-based imitation learning algorithm that combines online learning with a reduction to no-regret online learning, distinguished by its theoretical guarantee of sublinear regret relative to an expert policy when the latter is accessible during training.