THULab/lirislab_push_cup_target

Push Cup Target TsFile

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

Source Dataset

• Publisher and source file contributor: lirislab
• License: Apache-2.0
• Task: Push the red cup to the pink target.
• Split: train (0:30)
• Sampling rate: 30 fps
• Episodes: 30; frame rows: 8,528; tasks: 1; source Parquet shards: 30
• 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_push_cup_target.tsfile as table lirislab_push_cup_target. The meta/ directory contains the source JSON/JSONL metadata needed to describe the dataset; source Parquet files are not placed in meta/.

Time is synthesized as round(timestamp * 1000) in milliseconds and restarts from zero for each episode. The source timestamp is omitted after this mapping because it is represented by Time / 1000. frame_index is preserved and index is renamed to sample_index.

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)

No rows or action/state dimensions are intentionally removed. The only omitted source column is the redundant timestamp after the millisecond Time mapping.

Videos

The original repository stores 60 MP4 files 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 remain aligned with the original video frames through episode_index and frame_index.

Read Example

from tsfile import TsFileReader

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