Vision-Based Control and Navigation of an Autonomous Underwater Vehicle in Crowded Environments

Abstract

Autonomous underwater vehicles (AUVs) are used for a variety of purposes in the realms of research, industry, and the ocean as a whole. They are versatile and powerful, allowing for work to be done on tasks that are too inaccessible, costly, or time consuming for humans to complete directly. Researchers are constantly working to advance technologies that could further increase the capabilities of AUVs. One such frontier of AUV research is the use of underwater vision. This project sought to design an AUV system that utilizes vision-based sensing to explore unknown crowded environments. This involves the AUV mapping an environment, localizing itself in the map, and creating and following a path for comprehensive exploration. All of this would use vision as the primary sensing mechanism. In addition, a robot that could be used to test this system was designed and constructed.

The system design is explored first, with algorithms chosen for use in each element of the mission. Methods for image processing, stereo camera depth sensing, visual odometry, Simultaneous Localization and Mapping (SLAM), path planning, and vehicle control were each evaluated on the bases of industry and research context, effectiveness, and role in the overall system. The resulting network of algorithms can be implemented using the Robot Operating System (ROS). For the physical robot design, components were selected and used to build an effective, inexpensive test vehicle for the system. This design used prior thesis work as a starting point. The dynamics of the vehicle were derived, and PID control was introduced for use in testing. Though the overall system was not able to be tested comprehensively, modular pieces were tested in both simulation and on the test vehicle, and these experimental results are presented. At the end, suggestions are given for future work on this and related projects.