Loop Closure Detection Through Environmental Indicators In Underwater SLAM

Abstract

A key step in Simultaneous Localization and Mapping (SLAM) with autonomous robots is the ability to recognize when the vehicle is revisiting an area. The recognition of salient features allows re-localization to a previously visited place, forming a so-called loop closure. Loop closures enable autonomous underwater vehicles (AUVs) to reduce the unbounded navigation drift experienced when mapping unknown terrain. However, the scarcity of recognizable features in unstructured sub-sea environments makes disambiguation of places a challenging problem. In this work we study the application of water analysis sensors mounted on an AUV to construct descriptors of underwater environments. Since these sensors measure very different properties of the areas compared to terrain sensors, these descriptors could increase robustness against perceptual aliasing on place recognition and thus improve loop closure detection in a SLAM framework. We present a method to create and compare these spatial descriptors, test it with several data sets collected with an AUV in two different geographical locations and analyze their potential use for loop closure detection.