← Back ICRA 2026

Side-Scan Sonar SLAM Using Ping-Level Landmark Detection in Feature-Poor Seabed Environments

Jinho Im, Seonghun Hong

PDF

AI summary

Key figure (auto-extracted from paper)

Directly extracting landmarks from raw sonar ping signals cuts localization drift to under 5 meters in feature-poor seabeds, outperforming both dead-reckoning and image-based SLAM.

Side-scan sonar SLAM ping-level detection underwater robotics feature-poor environments drift correction

Problem

Dead-reckoning accumulates severe drift during long UUV missions, while existing side-scan sonar SLAM methods rely on image-level processing that fails in feature-poor environments.

Approach

The method processes raw one-dimensional backscatter intensity profiles to detect salient acoustic peaks as landmarks, using a range-adaptive RANSAC decay model and DBSCAN clustering to extract features without forming sonar images.

Key results

4.47 m final position error in real sea trials
84% drift reduction compared to dead-reckoning
77% accuracy improvement over image-level SLAM
Robust landmark extraction without acoustic image formation

Why it matters

Enables reliable long-range UUV navigation in challenging seabed terrains where traditional acoustic mapping and dead-reckoning fail.

Abstract

No abstract on file.

Index terms

Marine Robotics SLAM Localization