Enan, Sadman Sakib2024-07-242024-07-242024-05https://hdl.handle.net/11299/264301University of Minnesota Ph.D. dissertation. May 2024. Major: Computer Science. Advisor: Junaed Sattar. 1 computer file (PDF); xi, 135 pages.Numerous important tasks, such as environmental monitoring, cable or wreckage inspection, search-and-rescue, and oil drilling or spillage monitoring, are conducted underwater. A team of human divers typically carries out these challenging and often dangerous tasks, occasionally receiving assistance from a Remotely Operated Vehicle (ROV). However, an ROV is mainly controlled by someone on the surface which leads to inefficient collaboration due to the indirect engagement among the divers and the robot. In contrast, an Autonomous Underwater Vehicle (AUV) does not require a surface operator to operate and can significantly enhance task efficiency by actively engaging with a team of human divers and other AUVs during the tasks. Thus, it is imperative for the AUVs to have robust multi-Human-Robot Interaction (mHRI) capability. In this dissertation, we present a set of vision-based computational methods for AUV perception to facilitate effective underwater mHRI to allow successful collaboration among multiple divers and AUVs. Furthermore, we provide several novel underwater datasets designed to facilitate learning about robot motion, diver identity, their pose information, and multi-human-robot collaborative scenarios. Our proposed methods allow the AUV to enable human-comprehensible interaction between multiple AUVs, identify unique divers for secure interaction and collaboration, reposition itself for interaction by determining whether their human partners are attentive, and identify the current activity of divers to make informed decisions. However, the general operation of AUVs is severely impacted by various factors, such as water turbidity, rapid currents, varying lighting conditions, and signal attenuation. AUVs also have several platform-specific constraints, such as finite battery life, limited on-board processing power, and real-time operational requirements. We have taken these challenges into consideration while designing and implementing our proposed algorithms on-board physical AUV platforms. We have elaborated on the rationale behind the specific design choices made for each system. Experimentalvalidations on proposed datasets as well as through numerous robot trials, performed in both closed- and open-water environments (e.g., swimming pools and oceans), show the efficacy of each proposed system.enHuman-Robot InteractionMachine VisionThesis or Dissertation