Outlier robust state estimation through smoothing on a sliding window

Measurement outliers can severely impact on the performance of conventional state estimators. The design of state estimators exhibiting enhanced robustness to measurement outliers is of interest in many areas of systems and control engineering. In marine robotics applications the issue is particularly relevant for navigation and model identification tasks exploiting acoustic based positioning and velocity sensors that are subject to relatively high rates of outliers. A sliding window state estimator is designed by minimizing the Least Median of Squares cost function evaluated by running a Rauch-Tung-Striebel smoother on the current window. The resulting estimator is tested on Doppler Velocity Log navigation data acquired on an underwater robot. Although these are only preliminary results, they confirm that the approach can be successfully used online.