Learning Representations in Model-Free Hierarchical Reinforcement Learning