Outlier Rejection in Underwater Acoustic Position Measurements Based on Prediction Errors

Abstract

Outlier detection and rejection is an important step toward more robust underwater navigation systems. More specifically, acoustic positioning measurements can be notoriously intractable by introducing spikes, or freezing for periods of time, hence driving the navigation filter state estimates to wrong values. In this paper, a simple approach for detecting and rejecting outliers for underwater operations is presented. Acoustic positioning measurements are combined with absolute velocity measurements from a DVL (Doppler Velocity Log) sensor in order to robustify the filter when the vehicle navigates in the operative range of the DVL. Simple x2 statistic tests are employed in order to evaluate every new measurement and result in discarding those measurements which give large residuals compared to the predicted value from the Extended Kalman Filter. Moreover, the sum of the prediction errors is computed over a fixed number of valid acoustic measurements for filter divergence monitoring purposes. The efficacy of the approach is demonstrated using experimental data acquired during ROV operations.