Adversarial Imitation Learning from Video using a State Observer