A regional point descriptor for global topological localization in flooded subterranean environments

Abstract

A regional point descriptor for global localization using natural landmarks in flooded subterranean environments is presented. Global localization in underwater environments is complicated by a lack of sensors that land robots depend on for position estimation such as GPS, LADAR, and wheel odometry. This descriptor, the slide image, is designed to take advantage of orientation references available in subterranean voids while tolerating expected pose estimation errors in the horizontal plane. It serves as the basis of a robust topological navigation system for Minefish, an untethered, borehole deployable AUV designed to map flooded mines. Results are presented from tests on sonar data collected in the Wakulla springs tunnel system in Florida.