3D Reconstruction of Underwater Features in Lake Tahoe with an Autonomous Underwater Vehicle

Selena Sun, Rohan Tan Bhowmik, Elizabeth McElhinney

Abstract: We developed and deployed an autonomous underwater vehicle (AUV) for environmental monitoring at Lake Tahoe, California. Our system performs end-to-end autonomous underwater video collection and 3D reconstruction using Neural Radiance Fields (NeRF), combined with real-time environmental data collection. Our preprocessing pipeline enabled COLMAP to register 70-85\% of camera poses from underwater video sequences. Compared to traditional human-diver photogrammetry operations costing approximately 2,500 USD for a 10 ft $ \times$ 10 ft underwater survey, our AUV achieves comparable reconstruction quality for under \$600 per deployment—a 76\% cost reduction. Additionally, our AUV autonomously collects environmental monitoring data including eDNA samples, temperature, salinity, and pressure measurements. To our knowledge, this represents the first fully autonomous underwater vehicle capable of complete video-to-3D reconstruction workflows. Our design, open-sourced at stanfordrobosub.org, enables scalable, recurring ecological monitoring for climate research and biodiversity assessment.

Submission Number: 12