azithromycin/younghyopark_pick_tblock_mp_centered_debug6

Pick TBlock MP Centered Debug6 TsFile

This is an Apache TsFile conversion of younghyopark/pick_tblock_mp_centered_debug6, a LeRobot v2.1 Dual-Panda manipulation dataset for a centered T-block pick task.

Source and Attribution

• Original dataset: younghyopark/pick_tblock_mp_centered_debug6
• Original repository creator and uploader: Younghyo Park (younghyopark)
• License: Apache-2.0
• Robot type: DualPanda
• Task metadata: pick something up
• Split: train
• Sampling rate: 50 fps
• Scale: 10 episodes, 3,759 frame rows, 1 task, 10 source Parquet files
• Source frame layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet

The source card provides no paper or completed citation. Cite the original Hugging Face dataset and Younghyo Park when using this conversion.

The 10 train episode shards are merged into one table-model TsFile. Filter the episode_index and task_index TAG columns to select a device/episode.

Conversion Details

• Time = round(timestamp * 1000) in milliseconds. Time starts at zero in each episode.
• Source timestamp is dropped because it is exactly represented by Time / 1000 seconds.
• Source index is renamed to sample_index; frame_index is preserved.
• Vector and matrix columns are flattened in row-major order, preserving their source prefix with . replaced by _.
• No numeric rows or numeric dimensions are intentionally dropped.
• No source Parquet files are copied under meta/; meta/ contains only source JSON/JSONL metadata updated for TsFile.

Encoding and Compression

• FLOAT/DOUBLE FIELD values: GORILLA + LZ4
• INT32/INT64 FIELD values: TS_2DIFF + LZ4
• BOOLEAN FIELD values: RLE + LZ4
• Time: TS_2DIFF + LZ4
• TAG values: TsFile table device/TAG storage (STRING, PLAIN + LZ4 physically)

These are verified from the generated TsFile schema and Time chunk metadata, not inferred from configuration text.

Videos

There are no source videos to include or link. The pinned Hugging Face file tree contains no video files or videos/ directory, and source meta/info.json declares total_videos: 0 and video_path: null. Accordingly, this converted dataset contains no video copies and no fabricated video path. Numeric rows are self-contained and retain episode_index plus frame_index.

Minimal Read Example

from tsfile import TsFileReader

reader = TsFileReader("data/younghyopark_pick_tblock_mp_centered_debug6_train.tsfile")
table_name = "younghyopark_pick_tblock_mp_centered_debug6_train"
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_ctrl_0",
    "observation_environment_tblock_center_x",
]

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