Trajectory data provides an important data source for understanding human mobility patterns and transforming urban intelligence. However, existing trajectory modeling methods have limitations in terms of task specificity, regional dependency, and data sensitivity. The construction of the WorldTrace dataset aims to address these challenges by providing unprecedented geographical diversity and data scale, thereby promoting the development of region-agnostic universal trajectory models. This dataset contains 2.45 million trajectories collected from 70 countries worldwide, totaling 8.8 billion raw GPS points, making it the most geographically comprehensive and diverse trajectory dataset currently available.

📊 Key Statistics

Trajectory Information	Statistical Values
Number of Trajectories	2.45 million
Number of Trajectory Points	880 million (after processing)
Geographic Coverage	70 countries
Time Span	August 2021 - December 2023
Sampling Interval	1 second (after standardization)
Average Trajectory Duration	~6 minutes
Average Trajectory Distance	5.73 kilometers
Average Trajectory Speed	48.0 km/h
Average Trajectory Points	~358 points

📁 File Structure and Data Format

We provide two types of files: processed trajectory data (Trajectory.zip) and metadata (Meta.zip). Each record in the dataset typically contains the following fields:

📄 Trajectory: csv files

time: Timestamp in ISO 8601 format
latitude: Latitude coordinates under WGS84 standard
longitude: Longitude coordinates under WGS84 standard
altitude: Altitude of trajectory points (optional)
osm_way_id: Way ID in OpenStreetMap
matched_latitude: Matched latitude coordinates
matched_longitude: Matched longitude coordinates
matched_distance: Distance between trajectory point and matched coordinates
matched_type: Type of matched coordinates

📝 Meta: json files

Filename: Original trajectory file name
Uploaded: Upload time of trajectory file
Points: Number of data points in trajectory file
Start coordinate: Starting coordinates of trajectory (longitude, latitude)
End coordinate: Ending coordinates of trajectory (longitude, latitude)
Owner: Uploader of trajectory data
Description: Description information of trajectory
Tags: Tags related to trajectory
Visibility: Visibility description of trajectory
Distance: Total distance of trajectory
Time: Total duration of trajectory
geometry: Geometric representation of trajectory, typically an array of [longitude, latitude]

⏬ Download

Using Command Line

hf download OpenTrace/WorldTrace --repo-type dataset --local-dir <Your_dir>

Only download Trajectory:

hf download OpenTrace/WorldTrace --repo-type dataset --local-dir <Your_dir> --include "Trajectory.zip"

For other download methods, please refer to https://huggingface.co/docs/huggingface_hub/en/guides/download and https://huggingface.co/docs/huggingface_hub/en/guides/cli

⚖️ Privacy and License

Privacy: To protect user privacy and comply with international data protection regulations, all data collection strictly adheres to privacy regulations and ethical guidelines. Trajectory data has been anonymized, and any personally identifiable information has been excluded.

License: The original data of this dataset follows the OpenStreetMap Open Database License (ODbL) version 1.0. Any derivative datasets we release will also follow the same license terms.

If you use the WorldTrace dataset in your research, please cite our paper:

@article{unitraj2025,
  title={UniTraj: Learning a Universal Trajectory Foundation Model from Billion-Scale Worldwide Traces},
  author={Zhu, Yuanshao and Yu, James Jianqiao and Zhao, Xiangyu and Zhou, Xun and Han, Liang and Wei, Xuetao and Liang, Yuxuan},
  journal={Advances in Neural Information Processing Systems},
  volume={38},
  year={2025}
}

Downloads last month: 346