THULab/language_table_train_15000_20000_augmented

Language Table Train 15000 20000 Augmented TsFile

This dataset is a TsFile conversion of the Hugging Face dataset oxe-auge/language_table_train_15000_20000_augmented, an OXE-AugE / LeRobot v3.0 Language Table dataset.

Modalities: Time-series. The original dataset also includes eight video streams; those videos are not included in this repository and remain available in the source dataset.

Source Dataset

• Original dataset: oxe-auge/language_table_train_15000_20000_augmented
• Project page: https://oxe-auge.github.io/
• Project repository: https://github.com/GuanhuaJi/oxe-auge
• Paper: https://arxiv.org/abs/2210.06407
• License: CC-BY-4.0
• Format: LeRobot v3.0
• Robot type: mixed
• Robots: google_robot, images, jaco, kinova3, kuka_iiwa, panda, sawyer, and ur5e
• Split: single train split (0:5000)
• Scale: 5,000 episodes, 79,558 frames, 4,214 tasks
• Sampling rate: 10 fps
• Source data layout: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
• Source video layout: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4

Converted Files

• TsFile: data/language_table_train_15000_20000_augmented.tsfile
• Rows: 79,558
• Episodes: 5,000
• Tasks: 4,214
• Table name: language_table_train_15000_20000_augmented
• Time precision: milliseconds
• Metadata: meta/ is mirrored from the source dataset, with meta/info.json rewritten to describe the TsFile artifact and conversion mapping.

Schema

Column	Role	Type	Notes
Time	TIME	INT64	round(timestamp * 1000), in milliseconds; restarts per episode
episode_index	TAG	INT64	Source episode identifier
task_index	TAG	INT64	Source task identifier; join with meta/tasks.parquet for task text
frame_index	FIELD	INT64	Source frame index, preserved
sample_index	FIELD	INT64	Renamed from source index
observation_state_0 ... observation_state_1	FIELD	FLOAT	Flattened from observation.state[2]
observation_joints_0 ... observation_joints_7	FIELD	FLOAT	Flattened from source robot joints
observation_ee_pose_0 ... observation_ee_pose_6	FIELD	FLOAT	Flattened from source end-effector pose
observation_<robot>_joints_*	FIELD	FLOAT	Flattened per-robot joint positions
observation_<robot>_ee_pose_*	FIELD	FLOAT	Flattened per-robot end-effector poses
observation_<robot>_base_position_*	FIELD	FLOAT	Flattened per-robot base translations
observation_<robot>_base_orientation	FIELD	FLOAT	Per-robot base orientation scalar
observation_<robot>_ee_error_*	FIELD	FLOAT	Flattened per-robot end-effector error vectors

The per-robot field groups cover google_robot, jaco, kinova3, kuka_iiwa, panda, sawyer, and ur5e where present in the source schema.

episode_index and task_index are TAG columns, so they form the TsFile device dimension. To read one episode, filter by episode_index.

Conversion Notes

• The LeRobot v3 frame Parquet file under data/chunk-000/ was converted into one TsFile for the train split.
• Floating-point vector columns were flattened by preserving the source column name, replacing . with _, and appending the element index.
• Scalar columns containing . were renamed by replacing . with _.
• The source timestamp column is dropped because it is redundant with Time / 1000 seconds.
• The source index column is renamed to sample_index.
• The source natural_language_instruction[512] token vector is not stored in the TsFile table. The task text is preserved through task_index and the mirrored meta/tasks.parquet table. This keeps the TsFile table readable by the current TsFile SDK while preserving the language metadata in meta/.
• Videos are not uploaded here. Use the original dataset videos: https://huggingface.co/datasets/oxe-auge/language_table_train_15000_20000_augmented/tree/main/videos

Read Example

from tsfile import TsFileReader

path = "data/language_table_train_15000_20000_augmented.tsfile"
table = "language_table_train_15000_20000_augmented"

with TsFileReader(path) as reader:
    columns = [
        "episode_index",
        "task_index",
        "frame_index",
        "sample_index",
        "observation_state_0",
        "observation_google_robot_joints_0",
    ]
    with reader.query_table(table, columns, batch_size=4096) as result:
        batch = result.read_arrow_batch()
        print(batch)

Citation

If you use OXE-AugE datasets, the source dataset card asks users to cite both the OXE-AugE dataset and the upstream dataset.

@article{lynch2022interactive,
  title   = {Interactive Language: Talking to Robots in Real Time},
  author  = {Corey Lynch and Ayzaan Wahid and Jonathan Tompson and Tianli Ding and James Betker and Robert Baruch and Travis Armstrong and Pete Florence},
  journal = {arXiv preprint arXiv:2210.06407},
  year    = {2022},
  url     = {https://arxiv.org/abs/2210.06407}
}

@misc{
  ji2025oxeaug,
  title  = {OXE-AugE: A Large-Scale Robot Augmentation of OXE for Scaling Cross-Embodiment Policy Learning},
  author = {Ji, Guanhua and Polavaram, Harsha and Chen, Lawrence Yunliang and Bajamahal, Sandeep and Ma, Zehan and Adebola, Simeon and Xu, Chenfeng and Goldberg, Ken},
  year   = {2025},
  note   = {Manuscript}
}