Learning to Walk and Fly with Adversarial Motion Priors