Enhanced Intrusion Detection in Robot Operating Systems via Grid Search Based Multi-Head Attention Stacked Convolutional Network

Abstract

This study presents a novel intrusion detection system (IDS) for Robot Operating Systems (ROS), utilising a hybrid neural network combining 1D Convolutional Neural Networks (CNNs) with Multi-head Attention (MHA). This approach effectively captures both local and global data features, essential for detecting security threats in ROS. The model architecture includes layers of 1D-CNNs for detailed temporal feature extraction, complemented by MHA to identify complex intrusion patterns. Extensive hyperparameter optimisation through grid search ensures optimal model performance. A key aspect of this research is the use of the recently introduced ROSIDS23 dataset, which provides a comprehensive and realistic benchmark for testing. The model demonstrated exceptional accuracy, achieving 99% in training and greater than 97% in testing, highlighting its efficacy in ROS security enhancement. These results and the utilisation of ROSIDS23 dataset mark significant advancements in the field of robotic security.