Protecting Reward Function of Reinforcement Learning via Minimal and Non-catastrophic Adversarial Trajectory | IEEE Conference Publication | IEEE Xplore