qnoens Clean Table TsFile
Apache TsFile representation of the LeRobot robot dataset
qnoens/clean_table.
It contains numeric robot state, pose, joint, gripper, action, reward and
success signals for the tabletop clean-up task.
Source
- Original dataset:
qnoens/clean_table - Author/uploader:
qnoens - License: Apache-2.0
- LeRobot version: v2.1
- Task: clean up the table
- Paper and citation: not provided in the source dataset card
- Source split:
train, episodes 0 through 149 - Source data layout:
data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
Data Layout
- Split:
train - Table:
qnoens_clean_table - Rows: 102,885
- Episodes: 150
- Tasks: 1
- Sampling rate: 10 Hz
- Time precision: milliseconds
- Data file:
data/qnoens_clean_table.tsfile
Schema
Time is round(timestamp * 1000) in milliseconds and restarts from zero for
each episode. The source timestamp column is not retained because it is
represented by Time / 1000 seconds. Rows are ordered by episode_index,
task_index, and Time.
| Role | Columns |
|---|---|
| Time | Time (INT64, milliseconds) |
| TAG | episode_index, task_index |
| FIELD | frame_index, sample_index (source index), next_reward, next_success, seed, gripper_state |
| Flattened FLOAT FIELD | state_0 through state_6; robot_pose_0 through robot_pose_5; joints_0 through joints_5; action_0 through action_6 |
Vector values are expanded into scalar fields while preserving their source
names. Dots in field names are replaced with underscores, so next.reward
becomes next_reward and next.success becomes next_success.
Video Source
The four video streams remain only in the original Hugging Face repository:
wrist_image_original, scene_image_original, wrist_image, and
scene_image. Their source layout is
videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4.
Browse them at
qnoens/clean_table/videos.
Frame alignment uses the original episode_index and frame_index metadata.
Usage
from tsfile import TsFileReader
path = "data/qnoens_clean_table.tsfile"
reader = TsFileReader(path)
schema = reader.get_all_table_schemas()["qnoens_clean_table"]
columns = [column.get_column_name() for column in schema.get_columns()]
with reader.query_table("qnoens_clean_table", columns, batch_size=65536) as result:
batch = result.read_arrow_batch()
- Downloads last month
- 43