Snail Radar Dataset

4D radars are increasingly favored for odometry and mapping of autonomous systems due to their robustness in harsh weather and dynamic environments. Existing datasets, however, often cover limited areas and are typically captured using a single platform. To address this gap, we present a diverse large-scale dataset specifically designed for 4D radar-based localization and mapping. This dataset was gathered using three different platforms: a handheld device, an e-bike, and an SUV, under a variety of environmental conditions, including clear days, nighttime, and heavy rain.

Brief description

A dataset collected by 4D radars, stereo cameras, and lidars. The preprint paper describing the dataset is at arXiv.

@misc{huai2024snailradar,
  title={SNAIL Radar: A large-scale diverse benchmark for evaluating 4D-radar-based SLAM},
  author={Jianzhu Huai and Binliang Wang and Yuan Zhuang and Yiwen Chen and Qipeng Li and Yulong Han},
  year={2024},
  eprint={2407.11705},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2407.11705},
}

Changelog

• 2025-5-1:
  • Update the ZED2i camera extrinsics to Lidar, and ZED2i IMU extrinsics to lidar, using direct visual lidar calib on new calib seqs and dataset seqs.
  • Upload the dataset with time offset corrections and updated extrinsics to Baidu NetDisk and OneDrive.
• 2025-4-6:
  • Provide the residual time offsets.
  • Publish the comparison of swift_vio and correlation of ORB-SLAM3 mono + zed2i IMU for time offset estimation.
• 2024-12-28:
  • Blur the faces and license plates with egoblur,
  • add static transform tf files into each zipped sequence folder,
  • add metadata descriptions on the website.
• 2024-08-08:
  • Publishes all sequences, the website, and the dataset paper.

Roadmap

• Sync all motion related messages
• Calibrate all sensor extrinsic parameters
• Recalibrate the ars548 radar extrinsics for the 20231105 morning sequences, and update the zipped folders
• Fix the Hesai PandarXT-32 Point cloud timestamps by publishing the time corrections.
• Update the Baidu Netdisk with the latest dataset on OneDrive
• Update swift_vio with learning-based feature matchers and recalibrate the image time offsets.
• Update the ref trajs and full trajs using the dataset with corrected time offsets.

Complementary datasets

While SNAIL radar features rainy days, we also collected datasets in smoke and snowy days. For those interested, please refer to

• SmokeBasement

• SnowXinghu

I am so grateful to the hosting service by zenodo.

Quick start

The dataset can be downloaded from either OneDrive or Baidu Netdisk(百度网盘) 提取码: 3141.

We provided zipped rosbags and zipped data folders.

The 44 sequences of our dataset are repeatedly collected at 8 places with three platforms in diverse weather/lighting conditions. The place shorthands bc=basketball court, sl=starlake, ss=software school, if=info faculty, iaf=info and arts faculty, iaef=info, arts, and engineering faculty, st=starlake tower, 81r=August 1 road. The platforms H=handheld, E=ebike, S=SUV. The weather and lighting condition is clear and daytime unless specified. Different to other sequences of the info and arts faculty, the 20240113/2 sequence's route is anticlockwise.

A more detailed table for these sequences with additional attributes can be found here.

SDK Tools

The SDK tools to load, visualize, convert the dataset sequences for ROS1 bags and file folders are available here.

Software Programs Developed for this Dataset

During creation this dataset, we have developed several useful tools.

Time offset and rotation calibration tool

Time offset and rotation calibration tool in matlab is released at here. The core of the algorithm is FFT-based correlation.

Cascaded pose graph optimization

Cascaded pose graph optimization is released at here. The core of the algorithm is weighted pose graph optimization. Features include right invariant pose error, IMU preintegration factor, and ceres solver.

These programs are made strong by running through hundreds of trials set up by this large dataset.

Issues and Support

Please submit your issues to the dataset_tools repository.

In the past, some issues were posted on the source repository for this website. Although these issues have been fixed and closed, they may serve as a useful reference.

License

This SNAIL radar dataset is made available under the Open Database License. Any rights in individual contents of the dataset are licensed under the Database Contents License.