An Outlier Robust Filter for Maritime Robotics Applications

Abstract

Navigation systems of autonomous vehicles often exploit range measurement information that may be affected by outliers. In marine application the presence of outliers in sonar bathymetry, for instance, can be particularly severe due to multipath phenomena in the acoustic propagation. This paper describes a possible approach to process range measurements highly contaminated by outliers. The proposed solution builds on a robust parameter identification algorithm minimizing a nonlinear cost function that exploits the mathematical properties of Gibbs entropy. Numerical examples on simulated data are provided to illustrate the method and its performance. The use of simulated data allows to vary the amount of noise and outliers contamination while knowing the ground truth values of the parameters to be identified. For the sake of experimental validation, the method is also applied to third party (publicly available) upward looking sonar ice draft data collected by submarines in the Arctic Ocean.

Keywords

• marine robotics
• least-squares identification
• signal processing
• parameter estimation
• identification algorithms
• robust estimation