Dataset Card for DexCap-Data

Dataset Summary

This is the official dataset collected by the DexCap system to train dexterous robot manipulation using human hand motion capture data, as presented in the paper. It contains 30 minutes of mocap data for the wiping task and 60 minutes of in-the-wild mocap data for the packaging task.

Dataset Structure

Both raw data (.zip) and postprocessed data (.hdf5) are provided. The raw data is structured as follows:

save_data_scenario_1
├── frame_0
│   ├── color_image.jpg           # Chest camera RGB image
│   ├── depth_image.png           # Chest camera depth image
│   ├── pose.txt                  # Chest camera 6-DoF pose in world frame
│   ├── pose_2.txt                # Left hand 6-DoF pose in world frame
│   ├── pose_3.txt                # Right hand 6_DoF pose in world frame
│   ├── left_hand_joint.txt       # Left hand joint positions (3D) in the palm frame
│   └── right_hand_joint.txt      # Right hand joint positions (3D) in the palm frame
├── frame_1
└── ...

Details on how to postprocess the raw data into a training dataset (.hdf5) are introduced in the GitHub repo. The training scripts for policy learning with the HDF5 dataset are also included in the repo.

Dataset Creation

All data are collected by the DexCap system.

Additional Information

Licensing Information

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Citation Information

@article{wang2024dexcap,
  title = {DexCap: Scalable and Portable Mocap Data Collection System for Dexterous Manipulation},
  author = {Wang, Chen and Shi, Haochen and Wang, Weizhuo and Zhang, Ruohan and Fei-Fei, Li and Liu, C. Karen},
  journal = {arXiv preprint arXiv:2403.07788},
  year = {2024}
}