Experimental evaluation of outliers filtering techniques in networked acoustic localisation systems

Localisation-aware underwater networks are gaining increasing attention in the marine robotics community thanks to their ability of providing navigational services. This can be beneficial in a number of applications, as for instance to support the navigation of Autonomous Underwater Vehicles (AUVs) when traditional aiding systems are impractical or not cost effective. However, the unreliability of the acoustic channel, together with the additional overhead and constraints introduced by the network itself, result in localisation measurements that are intrinsically sporadic and unpredictable. This makes the outlier filtering problem of localisation measurements obtained through networked underwater systems particularly important and challenging. The presence of one or more wrong observations can heavily impact on the performance of the localisation system and hence resulting in poor navigation. This paper uses experimental data to compare the integration of two different outlier filtering methodologies in an existing network-aided AUV navigation filter. The first method aims at pre-filtering the measurements to identify and discard potential outliers before they are fused in the navigation filter. The second one modifies the correction step of the Kalman filter to integrate measurements in an outlier-robust way. Results shows that when the navigation filter is made outlier-robust the navigation performance increases and the system becomes less sensitive to tuning, a key characters tic for fielded systems.