azithromycin/seonixx_so100_test_v3

SO100 Test V3 TsFile

This dataset is an Apache TsFile conversion of seonixx/so100_test_v3, a LeRobot v2.1 SO100 robot-manipulation dataset. It contains numeric trajectories, timing, episode/task tags, and source metadata. Videos remain in the original Hugging Face repository.

Source Dataset and Attribution

• Original dataset: seonixx/so100_test_v3
• Pinned revision: 77c33dbab49adaf3d12cf15551d0cd591fbcc2e3
• Original author/uploader: Simon White (seonixx)
• Authorship evidence: the repository is owned by seonixx; its four commits are attributed to seonixx or the profile name Simon White.
• License: Apache-2.0
• Task: "Pick up the blue block and place it in the box."
• Robot: so100; LeRobot version: v2.1
• Split: train; sampling rate: 30 fps
• Scale: 100 episodes, 59,693 frames, 1 task
• Source shards: 100 Parquet files under data/chunk-000/episode_{episode_index:06d}.parquet
• Paper, homepage, and completed citation: not provided by the source card.

Converted File

• TsFile: data/seonixx_so100_test_v3.tsfile (844,063 bytes)
• Table: seonixx_so100_test_v3
• Rows: 59,693; episodes/devices: 100; tasks: 1
• Time precision: milliseconds
• Source Parquet files: 2,667,794 bytes; TsFile/source ratio: 0.316
• meta/ is preserved, with meta/info.json updated for the TsFile artifact.

Schema and Mapping

Time = round(timestamp * 1000) milliseconds. Time restarts at zero in every episode. The source timestamp is dropped afterward because, at the selected millisecond precision, it is represented by Time / 1000 seconds.

TsFile column	Role	Type	Source mapping
Time	TIME	TIMESTAMP	round(timestamp * 1000) ms
episode_index	TAG	STRING	Original INT64 episode index
task_index	TAG	STRING	Original INT64 task index
frame_index	FIELD	INT64	Preserved
sample_index	FIELD	INT64	Renamed from index
action_0 ... action_5	FIELD	FLOAT	Flattened from action[6]
observation_state_0 ... observation_state_5	FIELD	FLOAT	Flattened from observation.state[6]

The six vector dimensions are main_shoulder_pan, main_shoulder_lift, main_elbow_flex, main_wrist_flex, main_wrist_roll, and main_gripper. Dots in source names are replaced by underscores. No numeric row, episode, task, state dimension, or action 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 TAG storage

The on-disk schema, physical FIELD codecs, Time codec, TAG roles, and all 59,693 rows were read back with the Apache TsFile Java API.

Videos

Videos are not included in this TsFile repository. The original dataset has 200 frame-aligned AV1 MP4 files totaling 475,092,161 bytes (about 453.1 MiB), 640x480 at 30 fps with no audio, in two streams:

• observation.images.top
• observation.images.wrist.left

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

Validation

The source Parquet values and transformed scalar fields were compared exactly. Source, staged Parquet, and complete Java TsFile readback all contain 59,693 rows. TAG values, 100 episode indexes, one task, vector widths, source-derived Time, per-episode monotonicity, physical codecs, file size, and SHA-256 were checked locally. Conversion scripts and validation reports are intentionally not included in this upload-ready directory.

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/seonixx_so100_test_v3.tsfile")
with reader.query_table(
    "seonixx_so100_test_v3",
    ["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()