README.md

metadata

license: apache-2.0
authors:
  - Beegbrain
task_categories:
  - robotics
tags:
  - tsfile
  - timeseries
  - tabular
  - robotics
  - lerobot
  - so100
  - manipulation
modality:
  - timeseries
  - tabular
pretty_name: SO100 Put Cube in Cup TsFile
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/beegbrain_so100_put_cube_cup.tsfile
size_categories:
  - 10K<n<100K

SO100 Put Cube in Cup TsFile

This Apache TsFile edition comes from Beegbrain/so100_put_cube_cup, a LeRobot v2.0 SO100 robot-manipulation dataset. It contains numeric robot trajectories, frame timing, and episode/task tags. Camera videos remain in the original repository.

Source Dataset and Attribution

• Original dataset: Beegbrain/so100_put_cube_cup
• Original author/uploader: Beegbrain
• Authorship evidence: the repository shows Beegbrain as its owner, uploader, and sole contributor. The source card gives no separate author list.
• License: Apache-2.0
• Task: "Put the red cube in the red cup."
• Robot: so100; LeRobot version: v2.0
• Split: train; sampling rate: 15 fps
• Scale: 50 episodes, 10,778 frames, 1 task
• Source layout: 50 Parquet shards at data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet

TsFile Data

• Path: data/beegbrain_so100_put_cube_cup.tsfile (293,817 bytes)
• Table: beegbrain_so100_put_cube_cup
• Rows: 10,778; episodes/devices: 50; tasks: 1
• Time precision: milliseconds
• TsFile/source-Parquet size ratio: 0.371
• Source metadata remains under meta/, with meta/info.json describing the TsFile layout.

Schema

Time = round(timestamp * 1000) milliseconds. Time restarts at zero in every episode. Source timestamp is omitted afterward because it equals Time / 1000 seconds.

TAG columns, stored through the TsFile table/device mechanism:

• episode_index
• task_index

Scalar FIELD columns:

• frame_index
• sample_index (renamed from source index)

Flattened FLOAT FIELD groups:

• action[6] -> action_0 ... action_5
• observation.state[6] -> observation_state_0 ... observation_state_5

The six dimensions are main_shoulder_pan, main_shoulder_lift, main_elbow_flex, main_wrist_flex, main_wrist_roll, and main_gripper. Dots in source vector names become underscores. No numeric row, episode, task, or vector dimension is dropped.

Encodings and Compression

• FLOAT/DOUBLE: GORILLA + LZ4
• INT32/INT64: TS_2DIFF + LZ4
• Time: TS_2DIFF + LZ4
• BOOLEAN: RLE + LZ4 (the source has no BOOLEAN field)
• TAG: TsFile table/device storage

The physical table schema, Time codec, every FIELD codec, and all 10,778 rows were read back with the Apache TsFile Java API.

Videos

Videos are not included in this TsFile repository. The source has 100 frame-aligned AV1 MP4 files (155,224,490 bytes, about 148.0 MiB), 640x480, 15 fps, no audio, in two streams:

• observation.images.realsense_side
• observation.images.realsense_top

Each stream uses videos/chunk-000/{video_key}/episode_{episode_index:06d}.mp4. episode_index, frame_index, and meta/episodes.jsonl preserve alignment.

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/beegbrain_so100_put_cube_cup.tsfile")
with reader.query_table(
    "beegbrain_so100_put_cube_cup",
    ["episode_index", "task_index", "frame_index", "sample_index",
     "action_0", "observation_state_0"],
    batch_size=65536,
) as result:
    print(result.read_arrow_batch().to_pandas().head())
reader.close()

Source Notes

The source card states that the dataset was created with LeRobot. It does not provide a paper or completed BibTeX citation.