Adversarial Intrinsic Motivation for Reinforcement Learning

It further shows that the policy that minimizes this Wasserstein-1 distance is the policy that reaches the goal in as few steps as possible.