On Effective Scheduling of Model-based Reinforcement Learning