Anomaly Detection in Unmanned Vehicles: Thesis
Eliahu Khalastchi · 2010
Autonomy implies robustness. The use of unmanned (autonomous) vehicles is appealing for tasks which are dangerous or dull. However, increased reliance on autonomous robots increases reliance on their robustness. Even with validated software, physical faults can cause the controlling software to perceive the environment incorrectly, and thus to make decisions that lead to task failure. Anomaly detection can also be applied to medical monitors; alarms will be raised whenever the values of the measurements of the patient are anomalous with respect to the patient’s current condition. This anomaly detection is particularly useful for medical devices that monitor patients in recovery after a surgery, where the assigned nurses or physicians are not near to watch the monitor. Model-based diagnosis and fault-detection systems have been proposed to recognize failures. However, these rely on the capabilities of the underlying model, which necessarily abstracts away from the physical reality of the robot. We present two novel, model-free, domain independent approaches for detecting anomalies in unmanned autonomous vehicles, based on their sensor readings (internal and external). Both approaches use the familiar Mahalanobis Distance for the online anomaly detection. The first approach uses an offline training process. With this approach, we show the importance of a training process, which enables the Mahalanobis Distance to detect anomalies successfully. The second approach uses an online training process, in a way that is light-weight, and is able to take into account a large number of monitored sensors and internal measurements. These properties make the approach a “plug & play” anomaly detection mechanism for different robotic platforms. We demonstrate a specialization of the Mahalanobis Distance for robot use, and also show how it can be used even with very large dimensions, by online selection of correlated measurements for its use. We empirically evaluate these contributions in different domains: commercial Unmanned Aerial Vehicles (UAVs), a vacuum-cleaning robot, a high-fidelity flight simulator, and an electrical power system. We find that the online Mahalanobis distance technique, presented here, is superior to previous methods.