The autopilot system has been a hot research direction in recent years. As deep learning develops rapidly, the autopilot perception system based on deep neural network has become a research hotspot and an essential direction of the autopilot system. However, there is a potential risk of adversarial attacks on DNN model that may harm the perception system of the car and endanger the safety of passengers’ lives and property. The research on the adversarial attacks of the autopilot perception system is conducive to enhancing the robustness of the perception system and further enhancing the safety of the autopilot system. For this reason, it is beneficial to analyze and study the attack and defense techniques of adversarial examples on neural network systems and explore the security of neural network-based perception systems to build a defense system for the autonomous driving system. The paper first introduces the background knowledge of neural networks and adversarial examples applied to automatic driving perception systems. Secondly, we classify and introduce the existing classical attack methods of adversarial examples. Then, the current classic defense methods are introduced. Finally, we outlook the future research directions and related challenges of adversarial examples.