Towards Accurate Deep-Sea Localization in Structured Environments based on Perception Quality Cues

Arturo Gomez Chavez, Qingwen Xu, Christian A. Mueller, Sören Schwertfeger, Andreas Birk

Abstract: In recent years, the number of maritime exploration and exploitation activities has rapidly increased, and with it the necessity to perform more complex tasks underwater, e.g., floating manipulation and mapping with Remote Operated Vehicles (ROVs). The first step to perform these activities in a reliable manner, is to obtain an accurate robot localization estimate. Localization approaches based on multi-robot systems or complex acoustic infrastructures have been favored in the literature, but alternatively visual modalities are pursued when these options are not feasible. In this work, we present a two-stage navigation scheme that initially generates a coarse probabilistic map of the workspace that is used to refine localization accuracy and filter noise in the second stage. Additionally, an adaptive decision-making approach is introduced that determines which perception cues to incorporate into the localization filter, i.e., tracked 2D features or plane representations, to ensure high accuracy and reduce computation times. Our approach is thoroughly investigated in simulation and validated with deep-sea field trial data originated from oil & gas commercial operations.