11/04/2025
By Danielle Fretwell

The Francis College of Engineering, Department of Electrical and Computer Engineering, invites you to attend a Master's Thesis defense by Eric Huynh on: "Sonar-Centric Perception for Underwater Robots: Mapping the Deep, One Angle at a Time."

Candidate Name: Eric Huynh
Degree: Master’s
Defense Date: Tuesday, November 18, 2025
Time: 11 a.m.-12:30 p.m.
Location: Ball 313

Committee:

  • Advisor: Paul Robinette, Ph.D., Department of Electrical and Computer Engineering, University of Massachusetts Lowell
  • Maru Cabrera, Ph.D., Miner School of Computer and Information Sciences, University of Massachusetts Lowell
  • Jean-Francois Millithaler, Ph.D., Department of Electrical and Computer Engineering, University of Massachusetts Lowell

Abstract:
Reliable perception in underwater environments remain challenging due to poor visibility and the limitations of optical sensors. Acoustic sensing, particularly sonar, provides a robust alternative for navigation and situational awareness in such conditions. To address this, we propose a sonar-centric perception and SLAM framework for the low-cost BlueROV2 platform equipped with a Ping360 scanning imaging sonar. The goal is to advance affordable, reliable mapping and localization methods for underwater robots operating in low-visibility environments.

The first phase of this research investigates the generalization of deep learning–based perception methods across different sonar modalities. Using simulated and real-world sonar data, we evaluate object detection and tracking performance to understand how synthetic pretraining transfers to real sonar imagery. The results show that models trained on simulated multibeam sonar data perform well on real multibeam sonar but struggle with scanning imaging sonar due to modality-specific differences in object representation and acoustic noise patterns. These findings emphasize the importance of sonar modality when designing perception algorithms and highlight the challenges of domain adaptation between synthetic and real-world acoustic data.

Building on these insights, the next phase of this work focuses on improving the robustness and accuracy of sonar-based mapping. A Fourier-domain scan registration framework, based on the Fourier-SOFT (FS2D) method, is enhanced with multi-ring spectral whitening, phase-only correlation, and confidence-based fusion metrics to achieve more reliable rotation and translation estimates. These spectral refinements result in more stable registrations, particularly in environments with sparse or repetitive sonar returns. Experiments conducted using synthetic ground-truth transforms, simulated 3D environments in Gazebo simulation, and real-world pool trials with a BlueROV2 equipped with a Ping360 scanning imaging sonar demonstrate improved registration accuracy and stable map consistency. These results establish a practical and efficient foundation for sonar-based SLAM in underwater environments.

In the final stage of this research, a SLAM framework is extended to improve efficiency, scalability, and long-term reliability for real-world underwater missions. A hierarchical submapping architecture will organize sequential sonar scan-lines into compact local maps, enabling faster optimization and consistent performance over extended trajectories. To enhance global consistency, an information-driven loop closure strategy is implemented to selectively revisit and re-align past scans based on their potential to reduce pose uncertainty to allow the system to reason more intelligently about when and where to close loops. Finally, an uncertainty-aware sonar map representation is introduced to convey not only the reconstructed structures but also the confidence associated with each region. This offers a more interpretable and trustworthy depiction of the environment. Together, these developments aim to make sonar-based SLAM smarter, faster, and more dependable for long-duration underwater mapping.