Selected work

Representative Papers

CORAL-AUV: CFD Oriented Reinforcement Learning for Autonomous Underwater Vehicles

Jul 10, 2026

This study addresses the poor sim-to-real transfer performance of conventional autonomous underwater vehicle (AUV) control methods, which rely on oversimplified hydrodynamic models and struggle to adapt to configuration changes and environmental disturbances. To overcome this limitation, the authors propose a novel framework that integrates computational fluid dynamics (CFD) with reinforcement learning. Specifically, high-fidelity yet computationally efficient surrogate drag models (SDMs) are constructed from CFD data and embedded within a six-degree-of-freedom simulation environment to train control policies. Remarkably, the resulting policy is deployed on a physical AUV without any fine-tuning—demonstrating, for the first time, zero-shot sim-to-real transfer. Compared to controllers based on simplified models, the proposed approach reduces energy consumption by 31%, increases waypoint-to-waypoint speed by 11%, decreases trajectory error by 19%, and is the only method to successfully generalize under parameter perturbations, substantially enhancing both robustness and task performance.

0 citationsRead paper

Sonar-GPS Fusion for Seabed Mapping in Turbid Shallow Waters with an Autonomous Surface Vehicle

May 03, 2026

This study addresses the challenge of high-precision, long-range seafloor mapping in turbid shallow waters, where conventional optical methods are ineffective and existing sonar systems suffer from trajectory drift. The authors propose a drift-resistant seabed mapping framework that integrates local frame alignment using forward-looking sonar with global trajectory optimization leveraging multi-sensor data (GPS, IMU, and compass). Local registration is achieved via Fourier–Mellin transform (FMT), while an extended Kalman filter refines the global trajectory. Additionally, variance-weighted image fusion is introduced to suppress stitching artifacts. Field tests conducted in an oyster farm demonstrate a 9.5% reduction in trajectory RMSE compared to a baseline FMT-only approach, achieving sub-meter reconstruction accuracy while preserving high-resolution texture—enabling precise oyster stock estimation.

0 citationsRead paper

ReefMapGS: Enabling Large-Scale Underwater Reconstruction by Closing the Loop Between Multimodal SLAM and Gaussian Splatting

Apr 13, 2026

This work addresses the challenge of high computational cost in pose estimation and the difficulty of deploying underwater large-scale 3D reconstruction on field robots. The authors propose ReefMapGS, a novel framework that, for the first time, integrates multimodal SLAM with 3D Gaussian splatting in a closed-loop manner, enabling incremental reconstruction without relying on COLMAP. The method leverages acoustic, inertial, pressure, and visual sensors to construct a pose-graph SLAM system, initializes Gaussian primitives in high-confidence regions, and alternates between local image-based tracking and global Gaussian optimization. Evaluated on two complex coral reef environments, the system achieves COLMAP-free reconstruction over 700-meter AUV trajectories, significantly improving both global pose accuracy and reconstruction efficiency.

0 citationsRead paper

Sonar-MASt3R: Real-Time Opti-Acoustic Fusion in Turbid, Unstructured Environments

Mar 13, 2026

This work addresses the challenge of robust real-time 3D reconstruction in turbid or low-light underwater environments, where optical cameras alone often fail. To overcome this limitation, the authors propose a hand-eye photometric-acoustic fusion system that, for the first time, enables MASt3R-based real-time dense reconstruction in real-world high-turbidity waters (0.5–12 NTU). The approach leverages MASt3R to extract dense correspondences from optical images and integrates geometric constraints provided by 3D sonar to significantly enhance both robustness and real-time performance. Evaluated in unstructured turbid settings, the method demonstrates superior accuracy and stability compared to existing baselines, establishing a new benchmark for underwater dense reconstruction under adverse visibility conditions.

0 citationsRead paper

Autonomous Search for Sparsely Distributed Visual Phenomena through Environmental Context Modeling

Mar 10, 2026

This work addresses the challenge of efficiently locating sparse biological targets—such as specific coral species—in energy-constrained autonomous underwater vehicle (AUV) missions within visually sparse environments like coral reefs. The study introduces a novel approach that leverages environmental visual context, including co-occurring habitat features, as a guiding signal to drive adaptive search when direct target observations are absent. By integrating DINOv2 embeddings for patch-level one-shot online detection, the method simultaneously identifies both targets and their contextual cues, enabling dynamic path planning informed by real-time scene understanding. Experiments on real AUV imagery demonstrate that the proposed strategy discovers up to 75% of sparse targets in approximately half the time required by exhaustive coverage approaches, significantly outperforming baseline methods that rely solely on direct target detection.

0 citationsRead paper

Recent publications

Latest Papers

CORAL-AUV: CFD Oriented Reinforcement Learning for Autonomous Underwater Vehicles

Jul 10, 2026

This study addresses the poor sim-to-real transfer performance of conventional autonomous underwater vehicle (AUV) control methods, which rely on oversimplified hydrodynamic models and struggle to adapt to configuration changes and environmental disturbances. To overcome this limitation, the authors propose a novel framework that integrates computational fluid dynamics (CFD) with reinforcement learning. Specifically, high-fidelity yet computationally efficient surrogate drag models (SDMs) are constructed from CFD data and embedded within a six-degree-of-freedom simulation environment to train control policies. Remarkably, the resulting policy is deployed on a physical AUV without any fine-tuning—demonstrating, for the first time, zero-shot sim-to-real transfer. Compared to controllers based on simplified models, the proposed approach reduces energy consumption by 31%, increases waypoint-to-waypoint speed by 11%, decreases trajectory error by 19%, and is the only method to successfully generalize under parameter perturbations, substantially enhancing both robustness and task performance.

0 citationsRead paper

Sonar-GPS Fusion for Seabed Mapping in Turbid Shallow Waters with an Autonomous Surface Vehicle

May 03, 2026

This study addresses the challenge of high-precision, long-range seafloor mapping in turbid shallow waters, where conventional optical methods are ineffective and existing sonar systems suffer from trajectory drift. The authors propose a drift-resistant seabed mapping framework that integrates local frame alignment using forward-looking sonar with global trajectory optimization leveraging multi-sensor data (GPS, IMU, and compass). Local registration is achieved via Fourier–Mellin transform (FMT), while an extended Kalman filter refines the global trajectory. Additionally, variance-weighted image fusion is introduced to suppress stitching artifacts. Field tests conducted in an oyster farm demonstrate a 9.5% reduction in trajectory RMSE compared to a baseline FMT-only approach, achieving sub-meter reconstruction accuracy while preserving high-resolution texture—enabling precise oyster stock estimation.

0 citationsRead paper

ReefMapGS: Enabling Large-Scale Underwater Reconstruction by Closing the Loop Between Multimodal SLAM and Gaussian Splatting

Apr 13, 2026

This work addresses the challenge of high computational cost in pose estimation and the difficulty of deploying underwater large-scale 3D reconstruction on field robots. The authors propose ReefMapGS, a novel framework that, for the first time, integrates multimodal SLAM with 3D Gaussian splatting in a closed-loop manner, enabling incremental reconstruction without relying on COLMAP. The method leverages acoustic, inertial, pressure, and visual sensors to construct a pose-graph SLAM system, initializes Gaussian primitives in high-confidence regions, and alternates between local image-based tracking and global Gaussian optimization. Evaluated on two complex coral reef environments, the system achieves COLMAP-free reconstruction over 700-meter AUV trajectories, significantly improving both global pose accuracy and reconstruction efficiency.

0 citationsRead paper

Sonar-MASt3R: Real-Time Opti-Acoustic Fusion in Turbid, Unstructured Environments

Mar 13, 2026

This work addresses the challenge of robust real-time 3D reconstruction in turbid or low-light underwater environments, where optical cameras alone often fail. To overcome this limitation, the authors propose a hand-eye photometric-acoustic fusion system that, for the first time, enables MASt3R-based real-time dense reconstruction in real-world high-turbidity waters (0.5–12 NTU). The approach leverages MASt3R to extract dense correspondences from optical images and integrates geometric constraints provided by 3D sonar to significantly enhance both robustness and real-time performance. Evaluated in unstructured turbid settings, the method demonstrates superior accuracy and stability compared to existing baselines, establishing a new benchmark for underwater dense reconstruction under adverse visibility conditions.

0 citationsRead paper

Autonomous Search for Sparsely Distributed Visual Phenomena through Environmental Context Modeling

Mar 10, 2026

This work addresses the challenge of efficiently locating sparse biological targets—such as specific coral species—in energy-constrained autonomous underwater vehicle (AUV) missions within visually sparse environments like coral reefs. The study introduces a novel approach that leverages environmental visual context, including co-occurring habitat features, as a guiding signal to drive adaptive search when direct target observations are absent. By integrating DINOv2 embeddings for patch-level one-shot online detection, the method simultaneously identifies both targets and their contextual cues, enabling dynamic path planning informed by real-time scene understanding. Experiments on real AUV imagery demonstrate that the proposed strategy discovers up to 75% of sparse targets in approximately half the time required by exhaustive coverage approaches, significantly outperforming baseline methods that rely solely on direct target detection.

0 citationsRead paper