Non-intrusive Anomaly Detection of Industrial Robot Operations by Exploiting Nonlinear Effect

Abstract

With the development of Internet of Robotic Things concept, low-cost radio technologies open up many opportunities to facilitate the monitoring system of industrial robots, while the openness of wireless medium exposes robots to replay and man-in-the-middle attackers, who send pre-recorded movement data to mislead the system. Recent advances advocate the use of high-resolution sensors to monitor robot operations, which however require invasive retrofit to the robots. To overcome this predicament, we present RobotScatter, a non-intrusive system that exploits the nonlinear effect of RF circuits to fuse the propagation of backscatter tags attached to the robot to defend against active attacks. Specifically, the backscatter propagation interacted by the tags significantly depends on various movement operations, which can be captured with the nonlinearity at the receiver to uniquely determine its identity and the spatial movement trajectory. RobotScatter then profiles the robot movements to verify whether the received movement information matches the backscatter signatures, and thus detects the threat. We implement RobotScatter on two common robotic platforms, Universal Robot and iRobot Create, with over 1,500 operation cycles. The experiment results show that RobotScatter detects up to 94% of anomalies against small movement deviations of 10mm/s in velocity, and 2.6cm in distance.