Robust Dead Reckoning: Calibration, Covariance Estimation, Fusion and Integrity Monitoring

TL;DR

This paper introduces a comprehensive approach for robust vehicle odometry estimation using sensor calibration, covariance estimation, data fusion, and integrity monitoring to improve accuracy and reliability in multi-sensor systems.

Contribution

It presents novel calibration procedures, an improved wheel diameter estimation algorithm, and a robust odometry fusion method with outlier detection for enhanced vehicle state estimation.

Findings

Effective sensor calibration and covariance estimation methods.

Robust odometry fusion with outlier detection improves accuracy.

Application of chi-squared test for measurement integrity.

Abstract

To measure system states and local environment directly with high precision, expensive sensors are required. However, highly accurate system states and environmental perception can also be achieved using data fusion techniques and digital maps. One crucial task of multi-sensor state estimation is to project different sensor measurements into the same temporal, spatial and physical domain, estimate their covariance matrices as well as the exclusion of erroneous measurements. This paper presents a generic approach for robust estimation of vehicle movement (odometry). We will shortly present our calibration procedure, including the estimation of sensor alignments, offset / scaling errors, covariances / correlations and time delays. An improved algorithm for wheel diameter estimation is presented. Additionally an approach for robust odometry will be shown as odometry estimations are fused under known covariances, while outliers are detected using a chi-squared test. Utilizing our robust odometry, local environmental views can be associated and fused. Furthermore our robust odometry can be used to detect and exclude erroneous position estimates.