The objective of this career development plan is to investigate new probabilistic methods for simultaneous localization and mapping (SLAM) that will improve the precision and scale of robotic mapping in ocean science. The proposed methodology will scale to multiple heterogeneous vehicles, allow for extended exploration over long time durations and over multiple spatial scales, and be robust to the challenging limitations of the underwater environment. Many land/air SLAM methods are largely inapplicable underwater because of (a) a lack of point features of unstructured seafloor, and (b) the rapid attenuation of electromagnetic, optical, and acoustic radiation in comparison to land, air, and space. The proposed multi-vehicle multi-scalar SLAM navigation framework overcomes these challenges by fusing information from three disparate technologies: 1) real-time vision-based seafloor navigation, 2) inertial navigation systems, and 3) acoustic modem-based communication, to create a flexible navigation framework that allows for inter-nodal ranging and data sharing among heterogeneous nodes. By leveraging the perception and localization capability of neighboring vehicles via a (low-bandwidth) distributed estimation framework, the proposed methodology will provide a cooperative navigation framework that will yield improved precision and scale in underwater robotic mapping for ocean science.
The goal of this award was to facilitate the practical use of cooperative acoustic localization methods in underwater robotic mapping and navigation over small- and large-scale environments. This technology can be used for seafloor mapping for ocean science, search and rescue operations for sunken targets, or for oil and gas exploration of deep-sea resources, to name just a few of the practical applications. Acoustic one-way-travel-time navigation hardware and methods were developed that support time-of-flight ranging between all sender and receiver vehicles who can passively hear the sender's transmission. Additionally, mathematical algorithms for decentralized cooperative state estimation were developed that are amenable to the low data rates and lossy communication medium of underwater acoustic modems. The practical outcome is that underwater vehicles can make use of data and range constraints from other vehicles in the network to improve their own navigation estimate in a mathematically principled way that is robust to the peculiarities of underwater acoustic communication. Additionally, this project contributed to new graph reduction tools that can be used in robotic map building. This is important because previous tools did not allow for principled graph sparsification when measurements were not full rank, such as from a camera, while the newly developed tools do. This project also supported collaboration between the PI at the University of Michigan and the NOAA Thunder Bay National Marine Sanctuary (TBNMS) to map targets within the sanctuary waters using the autonomous underwater vehicle (AUV) technology developed under this project award. The research findings from this project were reported in over 25 journal and conference papers, a project website, a new AUV technology display at the NOAA-TBNMS Great Lakes Maritime Heritage Center, and several K-12 outreach web telecasts in cooperation with NOAA-TBNMS.