THULab/pack2_starts_slow

Pack2 Starts Slow TsFile

This dataset contains the numeric time-series portion of andlyu/pack2_starts_slow converted to Apache TsFile format.

Modalities: Time-series. The original dataset also includes video streams, which remain in the source Hugging Face repository and are not duplicated here.

Source Dataset

• Original dataset: andlyu/pack2_starts_slow
• Source format: LeRobot v2.1 Parquet frame data plus MP4 videos
• Robot type: so100_follower
• Task: Pack berries from plate to shell
• Sampling rate: 30 fps
• Source scale: 150 episodes, 60,756 frames, 1 task, 1 chunk, 1,200 videos
• License: not specified in the source metadata

Converted Files

• data/pack2_starts_slow_train.tsfile: 54,675 rows from 135 episodes
• data/pack2_starts_slow_val.tsfile: 6,081 rows from 15 validation episodes
• Total converted rows: 60,756

The split assignment follows the source meta/info.json split metadata. The source val split is exposed as the validation file.

TsFile Schema

• Time: generated timestamp in milliseconds
• TAG columns: episode_index, task_index
• FIELD columns: frame_index, sample_index
• FIELD columns from action[6]: action_0 through action_5
• FIELD columns from observation.state[6]: observation_state_0 through observation_state_5

Conversion Notes

• Time = round(timestamp * 1000) milliseconds, restarting within each episode.
• The source timestamp column is dropped because it is represented by Time / 1000 seconds.
• The source index column is renamed to sample_index.
• Vector columns are flattened by preserving the source column name, replacing . with _, and appending the element index.
• Video features under observation.images.* are omitted from TsFile; use the original dataset's videos/ directory for frame-aligned MP4 streams.
• Source meta/ files are mirrored, with meta/info.json extended by a tsfile_conversion object documenting the converted schema and video policy.

Minimal Read Example

from tsfile import TsFileReader

path = "data/pack2_starts_slow_train.tsfile"
with TsFileReader(path) as reader:
    tables = reader.get_all_table_schemas()
    print(tables.keys())