GAAD: GAN-Enabled Autoencoder for Real-Time Sensor Anomaly Detection and Recovery in Autonomous Driving

Autonomous vehicles are an essential component of the intelligent transportation system, and their safe operation depends on reliable data from their sensors. However, these vehicles are vulnerable to cyberattacks and sensor failures that can generate anomalous data and potentially result in fatal crashes. Therefore, there is a critical need for a real-time anomaly detection and recovery approach to mitigate these risks. This study proposes a novel unsupervised/semi-supervised machine learning approach, which we refer to as GAN-enabled autoencoder for anomaly detection (GAAD). GAAD aims to detect different types of anomalies and recover anomaly-free data streams in autonomous driving. GAAD incorporates an additional autoencoder module into a generative adversarial network (GAN), which enables effective learning of the distribution of nonanomalous data. We evaluate GAAD using the Lyft Level 5 dataset and demonstrate its superior performance compared to state-of-the-art benchmark. Our results indicate that GAAD can effectively detect anomalies and recover true signals from anomalous data, enhancing the safety and reliability of autonomous vehicles.