GAN-Enabled U-Shaped Network for Adversarial Attack Generation for Autonomous Unmanned Vehicles

With the rapid development of autonomous unmanned vehicle (AUV) technology, the requirements for sensor data integrity and reliability are increasing. Data-driven intrusion detection systems (IDSs) have been widely used as an effective module of defense to ensure AUV security. Once the IDS is destroyed, the impact on the normal driving of the AUV will be unimaginable. Unlike traditional research that seeks to improve the detection accuracy of IDS, this paper discusses the challenges posed by adversarial attacks on autonomous vehicles from the perspective of attackers. In this paper, an generative adversarial-enabled U-shaped network (GEUN) is proposed to generate adversarial sensor attacks in AUVs and reveal vulnerabilities in existing IDSs. Specifically, GEUN consists of a shared feature extraction encoder, an adversarial anomaly reconstruction decoder, a discriminator, and a detector. The feature extraction encoder is used to add multi-scale information to the input to strengthen the multi-scale feature extraction of sensor signals by the network. With the help of discriminators and detectors, the adversarial anomaly reconstruction decoder is used to generate anomalies based on the extracted multi-scale vehicle information. Through experimental comparison with other methods, it is shown that the proposed GEUN can effectively attack various common IDS, which provides guidance for future research and development in the field of security of AUVs.