AN ADAPTIVE ROBUSTIZING APPROACH TO KALMAN FILTERING

`This paper considers the problem of robustifying the Kalman filter. Starting from the equivalence between the Kalman filter and the feast-squares regression problem, the authors propose a statistical approach called M-estimation to solve the regression problem robustly. In addition, robust M-estimators are derived for adaptive estimation of the unknown a priori state and observation noise statistics, simultaneously with the system state. The results of simulation demonstrating the robustness of the proposed estimators are also included.