| dc.description.abstract | In recent years, underwater robots have become cheaper and more readily accessible, allowing smaller companies and even private individuals to conduct their own research and explore the underwater domain. This thesis presents two main contributions to the field of underwater robotics. The first contribution is the development of an underwater simulation environment using Unreal Engine and Gazebo, designed for visual SLAM and other robotic applications. The environment generates visually realistic scenarios that closely resemble real-world underwater conditions, serving as a valuable tool for research and development in robotics. The second contribution is the adaptation of ORB-SLAM3 for underwater environments using visual-pressure and visual-inertial-pressure configurations. These configurations demonstrate improved performance in the underwater domain through modifications to the visual component, pose estimation, local bundle adjustment algorithms, new vocabularies, and various initialization procedures. While both the simulation environment and the modified ORB-SLAM configurations show promising results, there remain several challenges and opportunities for future work. For the simulation environment, usability and sensor data synchronization are areas for improvement. Expanding the evaluation, implementing new environments, additional sensors, and a GUI would further enhance the simulation’s value. For the adapted ORB-SLAM configurations, refining scale estimation, addressing trajectory drift, and improving loop closure capabilities are crucial next steps. More tests in different environments and potential applications of deep learning techniques can also enhance the system’s performance. By making the simulation and SLAM system implementations openly available online, this thesis aims to increase accessibility to underwater robotics for the public, fostering continued development and innovation in the field. | en_US |
