Priors, Hierarchy, and Information Asymmetry for Skill Transfer in Reinforcement Learning