Decision Tree Based Diagnosis for Hybrid Model-Based/Data-Driven Fault Detection and Exclusion of a Decentralized Multi-Vehicle Cooperative Localization System - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Decision Tree Based Diagnosis for Hybrid Model-Based/Data-Driven Fault Detection and Exclusion of a Decentralized Multi-Vehicle Cooperative Localization System

Résumé

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 acquired by three Turtlebot3 equipped with wheel encoders (for odometry), a gyroscope (for the yaw angle) and a Marvelmind localization system (for the global position), and a ground truth is recorded using optitrack system.
Fichier non déposé

Dates et versions

hal-04278249 , version 1 (09-11-2023)

Identifiants

  • HAL Id : hal-04278249 , version 1

Citer

Zaynab El Mawas, Cindy Cappelle, Maan El Badaoui El Najjar. Decision Tree Based Diagnosis for Hybrid Model-Based/Data-Driven Fault Detection and Exclusion of a Decentralized Multi-Vehicle Cooperative Localization System. IFAC World Congress, Jul 2023, Yokohama, Japan. ⟨hal-04278249⟩
21 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More