Factor graphs and submap simultaneous localization and mapping for microbathymetry
The utility of high-resolution bathymetric surveys important to many problems in oceanography is often limited by the poor navigation options available to Unmanned Underwater Vehicles. This thesis presents a novel method to integrate conventional dead-reckoning navigation with recent advances in factor graph Simultaneous Localization and Mapping (SLAM) algorithms. Dead-reckoning is represented in the factor graph as a new type of factor using a linearized state transition model derived from the Extended Kalman Filter commonly used for dead reckoning. The new factor graph submap-SLAM is faster and more scalable than prior methods and is shown to properly represent navigation uncertainty. This new method is used to evaluate change detection algorithms using surveys before and after excavation of the Monterrey A shipwreck. Factor graph submap-SLAM is shown to significantly reduce change detection artifacts caused by navigation error. Finally, a derivation using the Cramer Rao Lower Bound demonstrates that all of the navigation improvement provided by SLAM over dead reckoning results from the quality of submap matches. This result leads to a metric that may be evaluated online during a survey to assess the terrain matching potential and may be used in the future to optimize survey trajectories for post-processing.^
Geographic information science and geodesy|Ocean engineering|Robotics
James Ian S Vaughn,
"Factor graphs and submap simultaneous localization and mapping for microbathymetry"
Dissertations and Master's Theses (Campus Access).