azithromycin/oumeanin_lekiwi-pickcube-test

LeKiwi PickCube Test TsFile

This dataset is an Apache TsFile conversion of oumeanin/lekiwi-pickcube-test, a LeRobot v3.0 LeKiwi mobile-manipulator dataset.

The converted artifact contains tabular time-series robot state, action, frame, episode, and task data. Camera videos remain in the original Hugging Face dataset and are not duplicated here.

Source Dataset and Author

• Original dataset: oumeanin/lekiwi-pickcube-test
• Pinned source revision: d904800f111d7f6724cc7ab21ab5bc4a3ad60ddf
• Original repository owner and uploader: oumeanin
• License: Apache-2.0
• Robot type: lekiwi_soarm
• LeRobot codebase version: v3.0
• Task: LeKiwi pick-cube test demonstrations
• Split: train
• Sampling rate: 30 fps
• Scale: 50 episodes, 185,664 frame rows, 1 task, 1 Parquet data shard
• 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

The source card does not provide a paper, completed citation, or a separate formal author list. Repository ownership and commit history identify oumeanin as the source publisher.

Converted Files

• TsFile: data/oumeanin_lekiwi_pickcube_test_train.tsfile (975,363 bytes)
• Table: oumeanin_lekiwi_pickcube_test_train
• Rows: 185,664
• Episode/task devices: 50
• Time precision: milliseconds
• Metadata: meta/ is mirrored from the source, with meta/info.json updated for the TsFile artifact

TsFile Schema

Time is INT64 milliseconds computed as round(timestamp * 1000). It restarts at zero for each episode.

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

• episode_index
• task_index

Scalar INT64 FIELD columns:

• frame_index
• sample_index, renamed from source index

Flattened FLOAT FIELD groups:

• action[9] -> action_0 ... action_8
• observation.state[9] -> observation_state_0 ... observation_state_8

The nine dimensions in both groups are: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos, base_vx_mps, base_vy_mps, base_theta_degps.

Encoding and Conversion Notes

• FLOAT/DOUBLE fields use GORILLA + LZ4.
• INT32/INT64 fields and Time use TS_2DIFF + LZ4.
• BOOLEAN fields use RLE + LZ4 when present; this source has no BOOLEAN field.
• Source episode_index and task_index are TsFile TAGs.
• All source rows are merged into one train TsFile and sorted by TAGs then Time.
• Vector columns are flattened to scalar fields; dots become underscores.
• Source timestamp is dropped after Time synthesis because it equals Time / 1000 seconds.
• Source index becomes sample_index; frame_index is kept.
• No numeric row or state/action dimension is dropped.
• The TsFile is 0.3252 times the source Parquet size.

Videos

Videos are not included in this TsFile repository. The source contains two frame-aligned AV1, 640x480, 30 fps streams without audio, each with ten MP4 shards:

• observation.images.base - file-000.mp4 through file-009.mp4
• observation.images.wrist - file-000.mp4 through file-009.mp4

The source path template is videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4. episode_index, frame_index, and meta/episodes/ preserve alignment between numeric rows and the original video frames.

Validation

Apache TsFile Java 2.2.1 performed a full query readback of all 185,664 rows. The validator also checked the table schema, TAG columns, per-column physical encoding/compression, Time encoding/compression, unique TAG+Time keys, vector widths, and monotonic Time within every episode. Reports remain local and are not part of this upload-ready directory.

Minimal Read Example

from tsfile import TsFileReader

reader = TsFileReader("data/oumeanin_lekiwi_pickcube_test_train.tsfile")
table_name = "oumeanin_lekiwi_pickcube_test_train"
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_0",
    "observation_state_0",
]

with reader.query_table(table_name, columns, batch_size=65536) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())
reader.close()

Citation

The source card provides no paper or completed BibTeX entry. Cite the original Hugging Face dataset and its publisher, oumeanin, when using this converted artifact.