Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations

A deep reinforcement learning (DRL) agent observes its states through observations, which may contain natural measurement errors or adversarial noises.