Structure & Priors in Reinforcement Learning (SPiRL)