THULab/pietroom_first_task_short

First Task Short TsFile

This dataset is an Apache TsFile conversion of pietroom/first_task_short, a LeRobot v2.1 SO100 robot-manipulation dataset for passing a marker.

Source Dataset and Attribution

• Original dataset: pietroom/first_task_short
• Original author, repository owner, uploader, and video contributor: pietroom
• License: Apache-2.0
• Task: "Pass the marker."
• Robot: so100; LeRobot version: v2.1
• Split: train; sampling rate: 30 fps
• Scale: 90 episodes, 53,830 frames, 1 task
• Source shards: 90 Parquet files totaling 2,911,374 bytes
• Source layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• The source card provides no paper, external homepage, or completed citation.

Data Summary

• Path: data/pietroom_first_task_short.tsfile
• Table: pietroom_first_task_short
• Rows: 53,830; episodes/devices: 90; tasks: 1
• Time precision: milliseconds
• TsFile/source-Parquet size ratio: 0.366

Schema and Mapping

Time = round(timestamp * 1000) milliseconds. Time starts at zero and is strictly increasing in every episode. The source timestamp is omitted after conversion because it is recoverable as 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, action dimension, or state 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

Videos

Videos are not included in this TsFile dataset. The 180 original AV1 MP4 files remain in the source repository, 640x480 at 30 fps with no audio, in two streams with 90 episode files each:

• observation.images.laptop
• observation.images.phone

The source layout is videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4. Use episode_index and frame_index to align numeric rows with video frames.

Minimal Read Example

from tsfile import TsFileReader

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

Citation

The source card supplies no paper or completed citation. Cite the original Hugging Face dataset and its publisher, pietroom, when using this conversion.