A study of natural robustness of deep reinforcement learning algorithms towards adversarial perturbations

Qisai Liu, Xian Yeow Lee, Soumik Sarkar

Abstract: Highlights•Evaluated RL agents’ performance in various attack scenarios and gym environments.•Identified optimal perturbation thresholds to ensure robustness against adversarial attacks.•Ranked RL agents by sensitivity and robustness after adversarial attacks.•Analyzed diverse attack strategies to assess their impact on system integrity.