Anomaly Detection in Robot Operating System(ROS) Topic Data for Unmanned Ground Vehicles(UGV) Using a Transformer

Key Points

• The study aims to detect anomalies in data from Unmanned Ground Vehicles' ROS frameworks due to potential cyber threats.
• Collected ROS topic data from unmanned ground vehicles
• Utilized a transformer-based autoencoder to model complex data relationships
• Learned temporal relationships between normal operation data and anomalies
• Validated the model with attack data from executed cyber attacks
• Successfully detected anomalies through reconstruction error in the dataset
• Model effectively identified abnormal patterns during simulated cyber attacks
• The approach demonstrates enhanced security against denial of service attacks and spoofing.