model-based reinforcement learning (RL) is a sample-efficient way of learning
complex behaviors by leveraging a learned single-step dynamics model to plan
actions in imagination. However, planning every action for long-horizon tasks
is not practical, akin to a human planning out every