Tethered remotely operated vehicles (ROVs) are central to subsea inspection, but tethers add complexity and cost. This talk presents a practical path to tetherless ROV operations over acoustic links using a prior model-based image compression pipeline. The core idea: build a photorealistic novel-view-synthesis (NVS) model of the site during mapping; during reinspection, transmit only a compact representation and a compressed difference image. In this talk, we will present: (1) a scene-specific compression framework driven by trained NVS models; (2) systematic evaluation of underwater NVS rendering in controlled and real-world settings; (3) a pose estimation method enhanced with geometry-aware loss and NVS-based data augmentation; and (4) a gradient-descent refinement step that reduces view-synthesis error and, consequently, transmitted data. Field trials benchmark our approach against conventional codecs and learned baselines, showing substantial bandwidth savings while preserving inspection-grade image quality. Together, these results make real-time, low-bandwidth, tetherless ROV operation feasible for subsea inspection and intervention.

Biography:

Luyuan Peng is a research fellow at the Acoustic Research Laboratory, National University of Singapore (NUS). She recently defended her PhD in Electrical and Computer Engineering at NUS. Luyuan’s research lies at the intersection of artificial intelligence, marine robotics and underwater communication. She is particularly interested in developing AI-driven perception, localization, and path planning systems that make underwater robots more autonomous and communication-efficient.

Beyond research, Luyuan serves the IEEE Oceanic Engineering Society (OES) community: she is Chief Editor of the OES Student & Young Professional Newsletter and Chair of the Singapore AUV Challenge 2026. She is also an IEEE OES Women in Engineering Propel Laureate (2024–2025).