Review for NeurIPS paper: Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations