Multimodal Deep Generative Models for Trajectory Prediction: A Conditional Variational Autoencoder Approach

Human behavior prediction models enable robots to anticipate how humans may\nreact to their actions, and hence are instrumental to devising safe and\nproactive robot planning algorithms. However, modeling complex interaction\ndynamics and capturing the possibility of many possible outcomes in such\ninteractive settings is very challenging, which has recently prompted the study\nof several different approaches. In this work, we provide a self-contained\ntutorial on a conditional variational autoencoder (CVAE) approach to human\nbehavior prediction which, at its core, can produce a multimodal probability\ndistribution over future human trajectories conditioned on past interactions\nand candidate robot future actions. Specifically, the goals of this tutorial\npaper are to review and build a taxonomy of state-of-the-art methods in human\nbehavior prediction, from physics-based to purely data-driven methods, provide\na rigorous yet easily accessible description of a data-driven, CVAE-based\napproach, highlight important design characteristics that make this an\nattractive model to use in the context of model-based planning for human-robot\ninteractions, and provide important design considerations when using this class\nof models.\n