Diminishing Return of Value Expansion Methods in Model-Based Reinforcement Learning.