Fault-Tolerant Fusion Using α-Rényi Divergence for Autonomous Vehicle Localization
Résumé
The use of a satellites localization system (GNSS: Global Navigation Satellites System) has become essential in an outdoor environment. However, suffering from several degradations such as satellite masking, NLOS/multipath, or interferences, GNSS alone is not able to ensure the availability, continuity, and integrity of a safety-critical localization function. Therefore, a multi-sensor fusion step of satellite measurements with proprioceptive data is necessary. Adding also, the erroneous measurements should be detected and excluded from the fusion procedure in order to ensure the high level of position estimation integrity. In this work, a tightly coupled architecture (GNSS/Odometer) method is presented by applying a Nonlinear Information Filter (NIF) integrating a Fault Detection and Exclusion (FDE) stage based on α-Rényi Divergence (α-RD). An appropriate fixed threshold is used based on a Receiver operating characteristic (ROC) study. Field-obtained GNSS and Odometer data are used in experimental studies to show the performance of our proposed algorithm.KeywordsLocalizationMulti-sensors fusionα-Rényi divergenceInformation metricsFault detection and exclusionGNSS