Automatic Reward Design via Learning Motivation-Consistent Intrinsic Rewards