Simultaneous policy learning and latent state inference for imitating driver behavior

Human driving depends on latent states, such as aggression and intent, that cannot be directly observed. In this work, we propose a method for learning driver models that can account for unobserved states. When trained on a synthetic dataset, our model is able to learn encodings for vehicle trajectories that distinguish between four distinct classes of driver behavior. Such encodings are learned without any knowledge of the number of driver classes or any objective that directly requires the model to learn encodings for each class. We show that latent state driving policies outperform baseline methods at replicating driver behavior. Furthermore, we demonstrate that the actions chosen by our policy are heavily influenced by its assigned latent state.