VIP: Towards Universal Visual Reward and Representation
via
Value-Implicit Pre-Training
(ICLR 2023, SpotLight)