Decision Tree based diagnosis for hybrid model-based/data-driven fault detection and exclusion of a decentralized multi-vehicle cooperative localization system*

Abstract

Cooperative navigation systems are one of the main topics of interest in multi-robot systems emerging nowadays, where the question of safety remains a very critical one preventing the actual integration of the technology. In this article, a multi-sensor multi-vehicle Cooperative Positioning System (CPS) is presented, with a hybrid fault detection method under a decentralized architecture, and that is tolerant to simultaneous sensor faults. In order to detect and isolate faults, a set of fault sensitive residuals are generated based on the divergence of Jensen Shannon (DJS) between the probability distributions predicted by the encoder based evolution model and the various observations obtained by sensors. Then, in order to detect a fault, a data-driven approach is applied, where the classification of faults is done by a pre-trained detection decision tree (D-DT) and isolation random forest (I-RF). The testing and evaluation of the approach is done on real data from three Turtlebot3 burger robot equipped with wheel encoders (for prediction), a gyroscope (for the yaw angle) and a Marvelmind (for the absolute position). A ground truth is also recorded using optitrack system.