THULab/ptizzza_so101_pompom_orange

SO101 Pompom Orange TsFile

This dataset is an Apache TsFile conversion of ptizzza/so101_pompom_orange, a LeRobot v2.1 SO101 robot-manipulation dataset created and uploaded by Olga Perepelkina (ptizzza).

Modalities: Time-series, Tabular.

The source task metadata states: Grasp a yellow pompom and put it in the yellow bowl. The repository name uses orange; this README preserves both the repository name and the authoritative task text.

Source Dataset

• Pinned source revision: 44ff81db8252a189b4f1fbc481dc54f1d6a74661
• Author/uploader: Olga Perepelkina (ptizzza)
• License: Apache-2.0
• Robot type: so101; LeRobot codebase version: v2.1
• Split: train
• Scale: 50 episodes, 22,400 frame rows, 1 task, 50 source Parquet files, 30 fps
• Source frame layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• Source video layout: videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4
• The source card does not provide a paper or completed citation.

Converted Files

• TsFile: data/ptizzza_so101_pompom_orange_train.tsfile
• Table: ptizzza_so101_pompom_orange_train
• Rows: 22,400
• Devices/episodes: 50
• TsFile size: 308,144 bytes
• Metadata: meta/ mirrors the source metadata; meta/info.json records the TsFile path, conversion mapping, source revision, and video policy.

TsFile Schema

Time is an INT64 millisecond timestamp computed as round(timestamp * 1000); it restarts at 0 for each episode. The redundant source timestamp field is dropped.

TAG columns (TsFile table device/tag mechanism):

• episode_index
• task_index

FIELD columns:

• frame_index
• sample_index (renamed from source index)
• action_0 ... action_5 (flattened from action[6])
• observation_state_0 ... observation_state_5 (flattened from observation.state[6]; dot becomes underscore)

No numeric rows or action/state dimensions were dropped. Source camera pixels are omitted from TsFile.

Encoding and Compression

The final file was rewritten with the explicit compact policy from EncodedLeRobotTsFileImporter.java:

• FLOAT/DOUBLE fields: GORILLA + LZ4
• INT32/INT64 fields: TS_2DIFF + LZ4
• Time: TS_2DIFF + LZ4
• BOOLEAN fields: RLE + LZ4
• TAG columns: TsFile table device/tag mechanism

This replaces the importer default of PLAIN/uncompressed encoding and substantially reduces the converted file size.

Videos

Videos are not duplicated in this converted repository. The source contains 50 frame-aligned AV1 MP4 files (640x480, 30 fps, no audio; 163,684,674 bytes) under videos/chunk-000/observation.images.webcam/. Numeric rows align with the original videos through episode_index, frame_index, and meta/episodes.jsonl.

Validation

Local validation passed: 50 source Parquet shards, 22,400 staged rows, 22,400 TsFile metadata rows, 50 devices, 2 TAG columns, 14 FIELD columns, non-empty SDK query readback, and a non-zero TsFile. VALIDATION.md and validation_report.json are kept locally and are not part of the uploaded dataset.

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/ptizzza_so101_pompom_orange_train.tsfile")
with reader.query_table(
    "ptizzza_so101_pompom_orange_train",
    ["episode_index", "task_index", "frame_index", "action_0", "observation_state_0"],
    batch_size=65536,
) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())
reader.close()