Date of Award

Summer 2020

Document Type

Open Access Dissertation


Computer Science and Engineering

First Advisor

Ioannis Rekleitis


This dissertation addresses the problem of real-time Simultaneous Localization and Mapping (SLAM) in challenging environments. SLAM is one of the key enabling technologies for autonomous robots to navigate in unknown environments by processing information on their on-board computational units. In particular, we study the exploration of challenging GPS-denied underwater environments to enable a wide range of robotic applications, including historical studies, health monitoring of coral reefs, underwater infrastructure inspection e.g., bridges, hydroelectric dams, water supply systems, and oil rigs. Mapping underwater structures is important in several fields, such as marine archaeology, Search and Rescue (SaR), resource management, hydrogeology, and speleology. However, due to the highly unstructured nature of such environments, navigation by human divers could be extremely dangerous, tedious, and labor intensive. Hence, employing an underwater robot is an excellent fit to build the map of the environment while simultaneously localizing itself in the map.

The main contribution of this dissertation is the design and development of a real-time robust SLAM algorithm for small and large scale underwater environments. SVIn – a novel tightly-coupled keyframe-based non-linear optimization framework fusing Sonar, Visual, Inertial and water depth information with robust initialization, loop-closing, and relocalization capabilities has been presented. Introducing acoustic range information to aid the visual data, shows improved reconstruction and localization. The availability of depth information from water pressure enables a robust initialization and refines the scale factor, as well as assists to reduce the drift for the tightly-coupled integration. The complementary characteristics of these sensing v modalities provide accurate and robust localization in unstructured environments with low visibility and low visual features – as such make them the ideal choice for underwater navigation. The proposed system has been successfully tested and validated in both benchmark datasets and numerous real world scenarios. It has also been used for planning for underwater robot in the presence of obstacles. Experimental results on datasets collected with a custom-made underwater sensor suite and an autonomous underwater vehicle (AUV) Aqua2 in challenging underwater environments with poor visibility, demonstrate performance never achieved before in terms of accuracy and robustness. To aid the sparse reconstruction, a contour-based reconstruction approach utilizing the well defined edges between the well lit area and darkness has been developed. In particular, low lighting conditions, or even complete absence of natural light inside caves, results in strong lighting variations, e.g., the cone of the artificial video light intersecting underwater structures and the shadow contours. The proposed method utilizes these contours to provide additional features, resulting into a denser 3D point cloud than the usual point clouds from a visual odometry system. Experimental results in an underwater cave demonstrate the performance of our system. This enables more robust navigation of autonomous underwater vehicles using the denser 3D point cloud to detect obstacles and achieve higher resolution reconstructions.