Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations Hongge Chen *,2 Bo Li

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