Robust Variational-Based Kalman Filter for Outlier Rejection With Correlated Measurements

State estimation is a fundamental task in many engineering fields, and therefore robust nonlinear filtering techniques able to cope with misspecified, uncertain and/or corrupted models must be designed for real-life applicability. In this contribution we explore nonlinear Gaussian filtering problems where measurements may be corrupted by outliers, and propose a new robust variational-based filtering methodology able to detect and mitigate their impact. This method generalizes previous contributions to the case of multiple outlier indicators for both independent and dependent observation models. An illustrative example is provided to support the discussion and show the performance improvement.

Automated summary

State estimation is a crucial task in engineering fields, requiring robust nonlinear filtering techniques to handle uncertainties and corrupted models. This study presents a new robust variational-based filtering methodology for detecting and mitigating the impact of outliers, contributing to performance improvement.