Model Imitation for Model-Based Reinforcement Learning