Generative Adversarial Imitation Learning