Towards a Simple Approach to Multi-step Model-based Reinforcement Learning