azithromycin/RasmusP_so100_Orange2Green

SO100 Orange2Green TsFile

This Apache TsFile dataset is derived from RasmusP/so100_Orange2Green, a LeRobot v2.1 SO100 robot-manipulation dataset. It contains numeric trajectories, timing, and episode/task tags. The camera videos remain in the original repository.

Source Dataset and Attribution

• Original dataset: RasmusP/so100_Orange2Green
• Repository owner, publisher, and commit author: RasmusP
• License: Apache-2.0
• Task: "Grasp the orange block and drop it in the box."
• Robot: so100; LeRobot version: v2.1
• Split: train; sampling rate: 30 fps
• Scale: 50 episodes, 29,712 frames, 1 task
• Source shards: 50 Parquet files totaling 1,357,643 bytes
• Source frame layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• Paper, external homepage, separate real-name author, and completed citation: not provided by the source card.

All three source repository commits are attributed to RasmusP.

TsFile Data

• File: data/rasmusp_so100_orange2green.tsfile (400,808 bytes)
• Table: rasmusp_so100_orange2green
• Rows: 29,712; episodes/devices: 50; tasks: 1
• Time precision: milliseconds
• TsFile/source-Parquet size ratio: 0.295
• Source meta/ is retained, with meta/info.json updated for the TsFile schema and source-video alignment.

Schema and Mapping

Time = round(timestamp * 1000) milliseconds. Time starts at zero and is strictly increasing inside every episode. The source timestamp is dropped afterward because 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 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 contains no BOOLEAN field)
• TAG: TsFile table/device TAG storage

The physical table schema, every field codec, TAG roles, and all 29,712 rows were read back with the Apache TsFile Java API.

Videos

Videos are not included here. The source contains 100 frame-aligned AV1 MP4 files (513,790,985 bytes), 640x480 at 30 fps with no audio, in two streams with 50 episode files each:

• observation.images.webcam
• observation.images.phone

The source layout is videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4. Use episode_index, frame_index, and meta/episodes.jsonl to align numeric rows with both original video streams.

Data Integrity

Source Parquet, staged Parquet, and complete Java TsFile readback contain the same 29,712 rows. Scalar values, flattened vector values, TAGs, episode indexes, Time mapping, monotonicity, physical codecs, file size, and SHA-256 were checked locally. The conversion script and local reports are excluded from the upload-ready directory.

Minimal Read Example

from tsfile import TsFileReader

reader = TsFileReader("data/rasmusp_so100_orange2green.tsfile")
with reader.query_table(
    "rasmusp_so100_orange2green",
    ["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 provides no paper or completed citation. Cite the original Hugging Face dataset and its publisher RasmusP when using this dataset.