Author ORCID Identifier

Date of Award

Spring 5-2024

Document Type

Thesis (Ph.D.)

Department or Program

Computer Science

First Advisor

Alberto Quattrini Li


Before a robot can act, it must perceive its environment. Though, this is not a simple task when considering the challenges in underwater domains -- poor visibility conditions, limited sensor configurations, and lack of readily accessible localization. Underwater robots have, nevertheless, improved dramatically with more extensive sensor and navigation equipment. Robot and sensor use have enabled us to explore all reaches of our oceans. On the other hand, these same robots are not easily accessible or transferable to many practical tasks, including fishery management, infrastructure maintenance, disaster response, site conservation, and ecological surveys. There is a growing need for robots that are more scalable -- accessible (consisting of off-the-shelf equipment), easy to transport and deploy, and less of a monetary barrier (on the robot and mission). The biggest hurdle is that research on robust perception tools suitable for these scalable robots is still underdeveloped.

This thesis develops methods for improving the 3-D scene reconstruction capability of scalable underwater robots. These robots are modular -- integrated with low-cost sensors, including a monocular camera, multiple lights, a pressure depth sensor, and a single-beam echosounder. Perceiving and modeling complex underwater scenes with a robot requires precise localization, clean and informative sensory data, and a way to recover 3-D scene information. However, this is not directly possible with our simple sensory suite. Dynamic water conditions cause unique image color loss and the air-tight camera enclosures introduce additional image distortions. With only a single camera, vision odometry systems will generate unreliable 3-D scene and localization information. Tools presented throughout this thesis addresses the above challenges. Finally, leading to a novel 3-D scene reconstruction framework that exploits the camera-and-light setup, providing a means to model the unknown scene while simultaneously localizing the robot. Thorough assessments -- in various underwater environments, from controlled scenes in a swimming pool to real-world scenarios in the ocean -- of each tool and final framework validate its importance in underwater perception. This thesis demonstrates the potential and feasibility of scalable underwater robots to undertake challenging perception-based tasks, providing accessible and abundant means for humans to explore and work in the underwater world.