cyberagent/VLG-Loc-Dataset

Vision-Language Global Localization (VLG-Loc) Dataset

This dataset is for evaluation of Vision-Language Global Localization (VLG-Loc).

Dataset Structure

Each dataset directory contains the following files. Note: All camera images (.png) are pre-corrected for lens distortion.

• left_camera_image.png: Image from the rear-left camera of the robot.
• center_camera_image.png: Image from the front-facing camera of the robot.
• right_camera_image.png: Image from the rear-right camera of the robot.
• data.yaml: Contains metadata, file paths, and ground truth information for the sample.
  • Format: A YAML file storing key-value pairs.
  • Example Content:

    task_label: global_localization
    ground_truth_pose: # Ground truth pose (x, y, yaw) in the map frame
      x: 1.4948672925333364
      y: -0.2408375545348086
      theta: -0.19592457986226722
    left_camera_image_path: left_camera_image.png
    center_camera_image_path: center_camera_image.png
    right_camera_image_path: right_camera_image.png
    pointcloud_path: pointcloud.npy
    timestamp: 1756980768.4454215
    

• pointcloud.npy: Point cloud data from the 2D LiDAR scan, saved as a NumPy array.
  • Format: A NumPy ndarray.
  • Shape: (N, 2), where N is the number of points.
  • Data Type: float32.
  • Content: Each row represents a 2D point (x, y) in the robot's base coordinate frame.

Environments

The table below details the environments included in this dataset.

Environment Name	Directory Name	Description
UG/UA (Uniform Geometry, Uniform Appearance)	env_ug_ua	An environment of identical columns arranged in a regular pattern.
UG/DA (Uniform Geometry, Diverse Appearance)	env_ug_da	An environment consisting of regularly placed bookshelves.
DG/UA (Diverse Geometry, Uniform Appearance)	env_dg_ua	An indoor scene with many pieces of furniture and frequent item repetition.
DG/DA (Diverse Geometry, Diverse Appearance)	env_dg_da	An indoor environment populated with numerous objects, where a high variety of furniture reduces item repetition.
Retail Store (Real)	env_retail_store_real	Real retail store environment (EZOHUB TOKYO).
Retail Store (Sim)	env_retail_store_sim	A simulated retail environment where appearances are substituted with alphanumeric labels.

Acknowledgement

It was collected in several simulation environments and at the real-world retail store "EZOHUB TOKYO", in cooperation with SATUDORA HOLDINGS CO.,LTD.