Outlier-robust Unscented Kalman Filters for multisensor autonomous underwater navigation

Abstract

Accurate and reliable localization and navigation systems are essential for modern mobile robots to successfully perform their missions. It is particularly difficult to establish localization and navigation solutions in the underwater environment, where Global Positioning Systems (GPSs) cannot be used. The Doppler Velocity Log (DVL) sensor, which provides highly accurate linear velocity estimates, is the basis of the most widely used techniques. In addition, payload sensors such as optical cameras or Forward-Looking SONARs (FLSs) are used for inspection and can serve as a reliable complement or replacement for the DVL. Given the availability of a large amount of velocity data, sensor fusion algorithms can improve estimation performance while the DVL, FLS, and camera are operating. However, they need to be modified to work with possible spurious data. Specifically, two Unscented Kalman Filters (UKFs) with an intrinsic outlier rejection strategy are discussed and an evaluation and comparison of the filters is performed. These filtering strategies differ in how the filter handles outliers. To explore the approaches discussed here, FeelHippo AUV was used to conduct an autonomous mission on Vulcano Island, Messina (Italy) and gather data to be processed for offline validation.