Toward autonomous underwater mapping in partially structured 3D environments

Abstract

Motivated by inspection of complex underwater environments, we have developed a system for multi-sensor SLAM utilizing both structured and unstructured environmental features. We present a system for deriving planar constraints from sonar data, and jointly optimizing the vehicle and plane positions as nodes in a factor graph. We also present a system for outlier rejection and smoothing of 3D sonar data, and for generating loop closure constraints based on the alignment of smoothed submaps. Our factor graph SLAM backend combines loop closure constraints from sonar data with detections of visual fiducial markers from camera imagery, and produces an online estimate of the full vehicle trajectory and landmark positions. We evaluate our technique on an inspection of a decomissioned aircraft carrier, as well as synthetic data and controlled indoor experiments, demonstrating improved trajectory estimates and reduced reprojection error in the final 3D map. My research supervisor, John Leonard, has provided invaluable advice, mentoring, and access to an overwhelming array of research projects and other opportunities. He would be the village wise man or chief, in constant contact with everyone, always available for advice both personal and professional, identifying problems to be overcome in our hunting and gathering activities, and coordinating trade with nearby villages to ensure we had adequate supplies for the winter.To my advisor Dana Yoerger, whose contagious energy sparked a passion for marine robotics, whose battle-tested wisdom guided me at land and at sea, I owe my entire career at MIT. I see him as the village shaman, dispensing sage advice while magically solving any problem with his beyond-mortal mastery of arcane technical systems.Michael Kaess has been absolutely essential to my academic progress, and I imagine him as the village's technical wizard. He would be master of emerging technologies like metal tools and mechanical advantage, and would share his knowledge generously to ensure the success of the entire community.Franz Hover, the chief of the village next door, has provided both long-term opportunities and day-to-day mentoring and advice. I am immensely grateful for his help, and have very much enjoyed our collaboration on both research and operations.My family is always close to my heart, and their support would be crucial to surviving the harsh alpine environment (although the closeness might be su↵ocating if we were holed up in a tiny yurt made of animal skins all winter, with only a sooty fire for warmth and the village passtime of pebble-tossing for entertainment). My mother Diane would certainly stay up all night helping me study the various edible roots and practice animal-skinning technique for my rite of passage into manhood, a 5 twelve-day journey into the icy mountains with nothing but a flint knife and alpacaskin blanket. Over meals of small game, my father Rex and I would discuss the finer points of glacier navigation, and occasionally get into heated debates about village politi...

Dense, Sonar-based Reconstruction of Underwater Scenes

“…In section 3. firing in the same cycle, the received signal will be "contaminated" by other signals in the same cycle. This results in the "tunnelling" artifact that can be seen in many profiling scans, and can often prove challenging for the extraction of range measurements [115].…”

“…To reduce the number of outliers in a submap, so as to improve the accuracy of the model and mitigate their effect on the registration process, the submap must be filtered before it is added to the map. It is important to note that any filtering process must take into account the features of interest in the scene; certain filtering processes may be very effective at removing outliers, but do so at the expense of small-scale detail in the scene [115].…”

“…Other approaches have also added some other form of post-segmentation outlier rejection, usually by looking at the agreement between the range measurements of neighboring beams to improve the accuracy of the range estimates[58,61,82]. Recognizing the difficulty in obtaining an accurate and robust segmentation using threshold-type methods, other approaches have instead addressed these limitations later in the pipeline[115]. This is achieved by employing outlier rejection and smoothing techniques on the point clouds that are obtained by spatially registering the range measurements.…”

Underwater inspection using sonar-based volumetric submaps

“…Random error is caused both by the acoustic noise in the environment and electronic noise in the transducer and signal processing circuitry. Systematic error manifests itself in several different ways, and can be traced to a combination of causes, including non-zero beamwidth, cross-talk, vehicle motion, and multi-path scattering [12]. Figure 3(a) shows a corner of a tank, as seen by the multibeam sonar used in our experiments (covered in greater detail in section V).…”

An evaluation of artificial fiducial markers in underwater environments

“…They can also be employed to label specific structures [3], such as a docking station or areas of interest, such as sampling points. In underwater environments, fiducial markers can be also applied to robot mapping and localization [4] [5], to monitoring underwater sensors [6], to assist ROV operators in robot control [7] and as an interface for human-robot communication [8] [9].…”