Fault Tolerant multi-sensor Data Fusion for vehicle localisation using Maximum Correntropy Unscented Information Filter and α-Rényi Divergence
Makkawi Khoder, Nourdine Ait-Tmazirte, El Badaoui El Najjar Maan, Nazih Moubayed · 2020
The paper presents a fault-tolerant multi-sensor fusion approach with Fault Detection and Exclusion (FDE) based on information theory. The Maximum Correntropy Criterion (MCC) in Unscented Information Filter (UIF) form, called (MCCUIF), is used as estimator. The Unscented Transformation (UT) provides an efficient tool to restrict the non-linear state estimation problem. However, the UIF works well with Gaussian noises, where its performance may decrease when dealing with non-Gaussian noises. The MCC is used to deal with non-Gaussian noises (for instance shot noises or Gaussian mixture noises). For detection and exclusion of erroneous measurements, a residual is designed using a- Rényi Divergence (a-RD) between a priori and a posteriori probabilities distributions. Then α-Rényi criterion (a-Rc) is used in the decision part of the proposed approach in order to calculate an adaptive threshold for FDE. In order to target both high integrity and accuracy of the navigation function of an autonomous vehicle in stringent environments (urban canyon, forests ...), this paper presents a tightly coupled architecture by merging raw data of a Global Navigation Satellite System (GNSS) with odometer (odo) measurements through the proposed approach. The main contributions of this paper are: - the proposition of a multisensor fusion approach using MCCUIF, - the development of an FDE method using a residual based on a-RD with an adequate choice of a value and adaptive thresholding, - the validation of the proposed approach with real experimental data.