Thesis (Ph. D.)--Joint Program in Applied Ocean Science and Engineering (Massachusetts Institute of Technology, Dept. of Ocean Engineering; and the Woods Hole Oceanographic Institution), 2005. / Bibliography: p. 119-129. / Obtaining accurate and repeatable navigation for robotic vehicles in the deep ocean is difficult and consequently a limiting factor when constructing vehicle-based bathymetric maps. This thesis presents a methodology to produce self-consistent maps and simultaneously improve vehicle position estimation by exploiting accurate local navigation and utilizing terrain relative Measurements. It is common for errors in the vehicle position estimate to far exceed the errors associated with the acoustic range sensor. This disparity creates inconsistency when an area is imaged multiple times and causes artifacts that distort map integrity. Our technique utilizes small terrain "sub-maps" that can be pairwise registered and used to additionally constrain the vehicle position estimates in accordance with actual bottom topography. A delayed state Kalman filter is used to incorporate these sub-map registrations as relative position measurements between previously visited vehicle locations. The archiving of previous positions in a filter state vector allows for continual adjustment of the sub-map locations. The terrain registration is accomplished using a two dimensional correlation and a six degree of freedom point cloud alignment method tailored for bathymetric data. / (cont.) The complete bathymetric map is then created from the union of all sub-maps that have been aligned in a consistent manner. Experimental results from the fully automated processing of a multibeamn survey over the TAG hydrothermal structure at the Mid-Atlantic ridge are presented to validate the proposed method. / by Christopher N. Roman. / Ph.D.

Thesis (Ph. D.)--Joint Program in Applied Ocean Science and Engineering (Massachusetts Institute of Technology, Dept. of Mechanical Engineering; and the Woods Hole Oceanographic Institution), 2012. / Cataloged from PDF version of thesis. / Includes bibliographical references (p. 177-187). / This dissertation presents three separate-but related-contributions to the art of underwater navigation. These methods may be used in postprocessing with a human in the loop, but the overarching goal is to enhance vehicle autonomy, so the emphasis is on automated approaches that can be used in realtime. The three research threads are: i) in situ navigation sensor alignment, ii) dead reckoning through the water column, and iii) model-driven delayed measurement fusion. Contributions to each of these areas have been demonstrated in simulation, with laboratory data, or in the field-some have been demonstrated in all three arenas. The solution to the in situ navigation sensor alignment problem is an asymptotically stable adaptive identifier formulated using rotors in Geometric Algebra. This identifier is applied to precisely estimate the unknown alignment between a gyrocompass and Doppler velocity log, with the goal of improving realtime dead reckoning navigation. Laboratory and field results show the identifier performs comparably to previously reported methods using rotation matrices, providing an alignment estimate that reduces the position residuals between dead reckoning and an external acoustic positioning system. The Geometric Algebra formulation also encourages a straightforward interpretation of the identifier as a proportional feedback regulator on the observable output error. Future applications of the identifier may include alignment between inertial, visual, and acoustic sensors. The ability to link the Global Positioning System at the surface to precision dead reckoning near the seafloor might enable new kinds of missions for autonomous underwater vehicles. This research introduces a method for dead reckoning through the water column using water current profile data collected by an onboard acoustic Doppler current profiler. Overlapping relative current profiles provide information to simultaneously estimate the vehicle velocity and local ocean current-the vehicle velocity is then integrated to estimate position. The method is applied to field data using online bin average, weighted least squares, and recursive least squares implementations. This demonstrates an autonomous navigation link between the surface and the seafloor without any dependence on a ship or external acoustic tracking systems. Finally, in many state estimation applications, delayed measurements present an interesting challenge. Underwater navigation is a particularly compelling case because of the relatively long delays inherent in all available position measurements. This research develops a flexible, model-driven approach to delayed measurement fusion in realtime Kalman filters. Using a priori estimates of delayed measurements as augmented states minimizes the computational cost of the delay treatment. Managing the augmented states with time-varying conditional process and measurement models ensures the approach works within the proven Kalman filter framework-without altering the filter structure or requiring any ad-hoc adjustments. The end result is a mathematically principled treatment of the delay that leads to more consistent estimates with lower error and uncertainty. Field results from dead reckoning aided by acoustic positioning systems demonstrate the applicability of this approach to real-world problems in underwater navigation. / by Michael Jordan Stanway. / Ph.D.

Thesis (S.M. in Oceanographic Engineering)--Joint Program in Applied Ocean Science and Engineering (Massachusetts Institute of Technology, Dept. of Ocean Engineering; and the Woods Hole Oceanographic Institution), 1999. / Includes bibliographical references (leaves 85-89). / by Qiang Wang. / S.M.in Oceanographic Engineering

Thesis (Ph. D.)--Joint Program in Applied Ocean Science and Engineering (Massachusetts Institute of Technology, Dept. of Ocean Engineering; and the Woods Hole Oceanographic Institution), 2005. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Includes bibliographical references (p. 173-187). / This thesis describes a vision-based, large-area, simultaneous localization and mapping (SLAM) algorithm that respects the low-overlap imagery constraints typical of autonomous underwater vehicles (AUVs) while exploiting the inertial sensor information that is routinely available on such platforms. We adopt a systems-level approach exploiting the complementary aspects of inertial sensing and visual perception from a calibrated pose-instrumented platform. This systems-level strategy yields a robust solution to underwater imaging that overcomes many of the unique challenges of a marine environment (e.g., unstructured terrain, low-overlap imagery, moving light source). Our large-area SLAM algorithm recursively incorporates relative-pose constraints using a view-based representation that exploits exact sparsity in the Gaussian canonical form. This sparsity allows for efficient O(n) update complexity in the number of images composing the view-based map by utilizing recent multilevel relaxation techniques. We show that our algorithmic formulation is inherently sparse unlike other feature-based canonical SLAM algorithms, which impose sparseness via pruning approximations. In particular, we investigate the sparsication methodology employed by sparse extended information filters (SEIFs) and offer new insight as to why, and how, its approximation can lead to inconsistencies in the estimated state errors. Lastly, we present a novel algorithm for efficiently extracting consistent marginal covariances useful for data association from the information matrix. / (cont.) In summary, this thesis advances the current state-of-the-art in underwater visual navigation by demonstrating end-to-end automatic processing of the largest visually navigated dataset to date using data collected from a survey of the RMS Titanic (path length over 3 km and 3100 m² of mapped area). This accomplishment embodies the summed contributions of this thesis to several current SLAM research issues including scalability, 6 degree of freedom motion, unstructured environments, and visual perception. / by Ryan M. Eustice. / Ph.D.

Visual Odometry (VO) is the process of estimating the relative motion of a vehicle by using solely image data gathered from the camera. In underwater environments, VO becomes extremely challenging but valuable since ordinary sensors for on-road localization are usually unpractical in these hostile environments. For years, VO methods have been purely based on Computer Vision (CV) principles. However, the recent advances in Deep Learning (DL) have ushered in a new era for VO approaches. These novel methods have achieved impressive performance with state-of-the-art results on urban datasets. Nevertheless, little effort has been made to push learning-based research towards natural environments, such as underwater. Consequently, this work aims to bridge the research gap by evaluating the effectiveness of the learning-based approach in the navigation of Autonomous Underwater Vehicles (AUVs). We compare two learning-based methods with a traditional feature-based method on the Underwater Caves dataset, a very challenging dataset collected in the unstructured environment of an underwater cave complex. Extensive experiments are thus conducted training the models on this dataset. Moreover, we investigate different aspects and propose several improvements, such as sub-sampling the video clips to emphasize the camera motion between consecutive frames, or training exclusively on images with relevant content discarding those with dark borders and representing solely sandy bottoms. Finally, during the training, we also leverage underwater images from other datasets, hence acquired from different cameras. However, the best improvement is obtained by penalizing rotations around the x-axis of the camera coordinate system. The three methods are evaluated on test sequences that cover different lighting conditions. In the most favorable environments, although learning-based methods are not up to par with the feature-based method, the results show great potential. Furthermore, in extreme lighting conditions, where the feature-based baseline sharply fails to bootstrap, one of the two learning-based methods produces instead qualitatively good trajectory results, revealing the power of the learning-based approach in this peculiar context. / Visuell Odometri (VO) används för att uppskatta den relativa rörelsen för ett fordon med hjälp av enbart bilddata från en eller flera kameror. I undervattensmiljöer blir VO extremt utmanande men värdefullt eftersom vanliga sensorer för lokalisering vanligtvis är opraktiska i dessa svåra miljöer. I åratal har VO-metoder enbart baserats på klassisk datorseende. De senaste framstegen inom djupinlärning har dock inlett en ny era för VO-metoder. Dessa nya metoder har uppnått imponerande prestanda på dataset urbana miljöer. Trots detta har ganska lite gjorts för att driva den inlärningsbaserad forskningen mot naturliga miljöer, till exempel under vattnet. Följaktligen syftar detta arbete till att överbrygga forskningsgapet genom att utvärdera effektiviteten hos det inlärningsbaserade tillvägagångssättet vid navigering av autonoma undervattensfordon (AUV). Vi jämför två inlärningsbaserade metoder med en traditionell nyckelpunktsbaserad metod som referens. Vi gör jämförelsen på Underwater Caves-datasetet, ett mycket utmanande dataset som samlats in i den ostrukturerade miljön i ett undervattensgrottkomplex. Omfattande experiment utförs för att träna modellerna på detta dataset. Vi undersöker också olika aspekter och föreslår flera förbättringar, till exempel, att delsampla videoklippen för att betona kamerarörelsen mellan på varandra följande bildrutor, eller att träna på en delmängd av datasetet bestående uteslutande på bilder med relevant innehåll för att förbättra skattningen av rörelsen. Under träningen utnyttjar vi också undervattensbilder från andra datamängder, och därmed från olika kameror. Den bästa förbättringen uppnås dock genom att straffa skattningar av stora rotationer runt kamerakoordinatsystemets x-axel. De tre metoderna utvärderas på testsekvenser som täcker olika ljusförhållanden. I de mest gynnsamma miljöerna visar resultaten stor potential, även om de inlärningsbaserade metoder inte är i nivå med den traditionella referensmetoden. Vid extrema ljusförhållanden, där referensmetoden misslyckas att ens initialisera, ger en av de två inlärningsbaserade metoderna istället kvalitativt bra resultat, vilket demonstrerar kraften i det inlärningsbaserade tillvägagångssättet i detta specifika sammanhang.