THULab/lirislab_clear_square_orig

Clear Square Orig TsFile

Apache TsFile representation of lirislab/clear_square_orig, a LeRobot v2.1 so100 robot dataset.

Source

• Publisher/author: lirislab
• License: Apache-2.0
• Task: Remove everything from the square.
• Split: train (0:50)
• Sampling rate: 30 fps
• Episodes: 50; frame rows: 19,189; tasks: 1; source Parquet shards: 50
• Source frame layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet

The source card does not provide a paper or BibTeX citation.

TsFile layout

The numeric rows are stored in data/lirislab_clear_square_orig.tsfile as table lirislab_clear_square_orig. The meta/ directory contains the JSON/JSONL metadata needed to describe the source; source Parquet files are not copied into meta/.

Time = round(timestamp * 1000) is an INT64 millisecond timeline that restarts at zero for each episode. The redundant source timestamp is dropped. index is renamed to sample_index; frame_index is preserved.

Role	Columns
TIME	Time (INT64, milliseconds)
TAG/device	episode_index, task_index
FIELD	frame_index, sample_index (INT64)
FIELD	action[6] -> action_0 ... action_5 (FLOAT)
FIELD	observation.state[6] -> observation_state_0 ... observation_state_5 (FLOAT)

Storage profile: Time TS_2DIFF + LZ4; FLOAT/DOUBLE GORILLA + LZ4; INT32/INT64 TS_2DIFF + LZ4; BOOLEAN RLE + LZ4 (no BOOLEAN source columns). TAGs use native TsFile table-model device/tag storage.

Dropped source column: timestamp, because it is represented exactly by Time / 1000 seconds. No rows or numeric action/state dimensions are dropped.

Videos

The original repository stores 100 MP4 files (50 episodes x 2 streams, approximately 277 MB) under videos/:

• videos/chunk-000/observation.images.realsense_top/episode_XXXXXX.mp4
• videos/chunk-000/observation.images.realsense_side/episode_XXXXXX.mp4

Videos are not copied into this TsFile dataset. Numeric rows stay frame-aligned with the original videos through episode_index and frame_index.

Read example

from tsfile import TsFileReader

reader = TsFileReader("data/lirislab_clear_square_orig.tsfile")
with reader.query_table("lirislab_clear_square_orig", ["episode_index", "task_index", "Time", "frame_index", "action_0"], batch_size=4096) as result:
    print(result.read_arrow_batch().to_pandas().head())
reader.close()