Unsupervised Reinforcement Learning with Contrastive Intrinsic Control