Camera Relocalization using 3D Point Clouds for Enhanced Underwater Situational Awareness
PI: Leonidas Guibas
Department: Computer Science
Sponsor: United States Navy (USN) ONR NEPTUNE Program
Underwater missions are critical in a variety of naval affairs such as marine science, oceanography,search-and-rescue missions, salvage, exploration, facility inspection and warfare. Despite this wide outreach, utilization of underwater vehicles is still limited due to the physiological and technological challenges posed by the deep sea operation. Recently, the nuisances fostered significant research in autonomous underwater vehicles (AUVs). Unfortunately, water behaves significantly differently than thin air and the advanced technology readily available to self driving cars cannot be immediately utilized in deep waters. In this work, by leveraging recent advances in underwater 3D scanning, we propose to adopt and deploy 3D perception algorithms such as segmentation, object detection and mapping, developed for the land into submarines, equipping them with intelligent scene understanding and autonomy. This has many desirable uses: (i) navigation without GPS and external sensors which are unavailable in the ocean, (ii) autonomous docking, target detection and task-driven decision making, and (iii) autonomous exploration without the need of human operators. Unfortunately, the state of the art AI solutions would not generalize to underwater scenes which are fundamentally different in appearance, geometry and physics. Bridging this gap constitutes the majority of the efforts we conduct as part of this project.
H4D Focus Areas: AI/ML, Autonomy, Technology Transition