[R] On the role of planning in model-based deep reinforcement learning