Datasets for Maritime Robotics

Open-access datasets are essential for developing and validating perception, navigation, and control algorithms. This page catalogs available datasets for maritime robotics research.

Multi-Sensor Datasets

REMARO Data

REMARO Data Collections

The REMARO (Reliable AI for Marine Robotics) network provides various datasets for marine robotics.

Contents: - Multi-sensor recordings - Underwater vehicle data - Environmental conditions - Ground truth when provided by the dataset

Use Cases: - Algorithm benchmarking - Sensor fusion research - Environmental monitoring studies

Format: ROS bags, raw sensor data

Sonar Datasets

OpenSonarDatasets

OpenSonarDatasets

Curated collection of sonar datasets from various sources.

Contents: - Imaging sonar data - Side-scan sonar - Multibeam sonar - Various underwater environments

Use Cases: - Sonar SLAM development - Object detection and classification - Acoustic image processing

Annotations: Varies by dataset

MIMIR-UW

MIMIR-UW - A Multipurpose Synthetic Dataset for Underwater Navigation and Inspection

Type: Synthetic dataset (simulation-generated)

Contents: - Synthetic underwater imagery - Sonar data - Ground truth poses - Various underwater scenarios

Advantages: - Ground truth available for synthetic scenes - Controllable conditions - Large-scale data generation - Multiple sensor modalities

Use Cases: - Training deep learning models - Algorithm validation - Simulation-to-real transfer studies

Object Detection Datasets

Underwater Trash Detection

LearnOpenCV Underwater Trash Dataset

Contents: - Underwater images with trash - Bounding box annotations - Various marine environments

Use Cases: - Object detection training - Marine conservation applications - Environmental monitoring

Format: Images with YOLO-format annotations

Marine Structure Datasets

Datasets for inspecting underwater structures:

Common Targets: - Pipelines - Cables - Ship hulls - Offshore platforms - Underwater infrastructure

Characteristics: - Often proprietary or restricted - Some academic releases available - Competition datasets (e.g., from MBZIRC)

Visual Odometry and SLAM Datasets

Marine Visual Datasets

Datasets for visual navigation in underwater environments:

Challenges: - Limited visibility - Color distortion - Artificial lighting - Feature-sparse environments

Available Datasets: - Research institution releases - Competition datasets - Simulated environments

Surface Vehicle Datasets

Maritime Survey Data

Datasets from autonomous surface vehicles:

Contents: - GNSS/GPS tracks - Radar data - Camera imagery - Meteorological data - Wave conditions

Sources: - Research projects - Maritime competitions (VRX) - Academic institutions

Simulation-Generated Datasets

Gazebo-Based Datasets

Data generated from maritime simulators:

Sources: - VRX competition scenarios - DAVE simulation - Custom simulation environments

Advantages: - Perfect ground truth - Reproducibility - Controllable conditions - Unlimited data generation

Use Cases: - Initial algorithm development - Sim-to-real transfer research - Benchmarking

Acoustic Communication Datasets

Underwater acoustic channel data:

Contents: - Channel impulse responses - Communication performance metrics - Environmental parameters

Sources: - Oceanographic institutions - Research projects - ns-3 UAN simulations

Environmental Datasets

Oceanographic Data

Publicly available ocean data:

Sources: - NOAA databases - Copernicus Marine Service - World Ocean Database - Argo float data

Contents: - Temperature and salinity profiles - Current measurements - Wave conditions - Bathymetric data

Use Cases: - Environment modeling - Mission planning - Algorithm validation

Dataset Repositories

General Repositories

  • IEEE DataPort: Hosts various maritime robotics datasets
  • Kaggle: User-contributed datasets including underwater imagery
  • Academic Institutions: Many labs publish datasets with papers

MGDS (Marine Geoscience Data System)

MGDS provides public marine geoscience datasets and archives with broad coverage and consistent metadata practices.

Data Sources:

Access: - Public datasets with downloads and metadata - DOI-backed dataset references where available - Publications often link to related datasets

Why MGDS Data Matters: - Publicly archived data with traceable provenance - Useful for validating mapping, navigation, and perception pipelines

Competition Datasets

Some maritime robotics competitions release datasets or logs; check individual competition sites and repositories for availability and licensing (for example, VRX simulation logs).

Creating and Sharing Datasets

Best Practices

When creating datasets for the community:

  1. Documentation:
  2. Sensor specifications
  3. Calibration parameters
  4. Environmental conditions

  5. Data collection procedures

  6. Formats:

  7. ROS bags for multi-sensor data
  8. Standard formats (images, point clouds)

  9. Metadata in human-readable format (YAML, JSON)

  10. Annotations:

  11. Ground truth when available
  12. Annotation methodology

  13. Quality assessment

  14. Licensing:

  15. Clear license (Creative Commons, MIT, etc.)
  16. Citation requirements

  17. Usage restrictions if any

  18. Accessibility:

  19. Permanent hosting (university servers, Zenodo, etc.)
  20. DOI for citation
  21. README with quick-start guide

Contributing to This List

If you have a maritime robotics dataset to share:

  1. Add it to this page via pull request
  2. Include: name, link, brief description, use cases
  3. Specify data type and format
  4. Indicate if annotations/ground truth are available
  5. Provide license information

Dataset Comparison

Dataset Type Sensors Size Ground Truth License Format
REMARO Data Multi-sensor Various Varies Partial Open ROS bags
OpenSonarDatasets Sonar Imaging/Side-scan Varies Partial Open Mixed
MIMIR-UW Synthetic Camera/Sonar Large Full Open Simulation
Trash Detection Vision Camera Medium Bounding boxes Open Images

This page was last updated: December 30, 2025