Detecting Data Injection Attacks in ROS Systems using Machine Learning

Abstract

In recent decades, there have been numerous technological advances that have allowed robots to share more and more space with humans. However, these systems are built on top of traditional computing platforms and are susceptible to the same cyber-attacks. In addition, they also introduce a new set of security issues that can result in a breach of privacy or even physical harm. In this context, a new generation of robotics software has gained momentum. The Robot Operating System (ROS) is one of the most popular frameworks for robot researchers and developers. However, several studies have shown that it brings vulnerabilities that can compromise its security and reliability. This work aims to evaluate the application of anomaly-based intrusion detection techniques to recognize data injection attacks. A model was proposed using the support vector machine (SVM) algorithm, which was trained from the network traffic characteristics of a ROS application. Results obtained through experiments conducted in a simulated environment demonstrated an accuracy of about 92% in the detection of these attacks.