Challenges and Countermeasures for Adversarial Attacks on Deep Reinforcement Learning

Deep reinforcement learning (DRL) has numerous applications in the real world, thanks to its ability achieve high performance a range of environments with little manual oversight. Despite great advantages, DRL is susceptible adversarial attacks, which precludes use real-life critical systems and (e.g., smart grids, traffic controls, autonomous vehicles) unless vulnerabilities are addressed mitigated. To address this problem, we provide comprehensive survey that discusses emerging attacks on DRL-based potential countermeasures defend against these attacks. We first review fundamental background present machine techniques. then investigate an adversary can exploit attack along state-of-the-art prevent such Finally, highlight open issues research challenges for developing solutions deal intelligent systems.