Datasets:
SVLA SO100 Task4 V3 Clean TsFile
This dataset is converted from danaaubakirova/svla_so100_task4_v3_clean to Apache TsFile format.
Modalities: Time-series. The converted repository contains numeric SO100 robot state/action trajectories and LeRobot metadata. Camera videos are not included in this TsFile conversion; they remain in the original Hugging Face dataset under its videos/ tree.
Source Dataset
- Original dataset:
danaaubakirova/svla_so100_task4_v3_clean - License:
apache-2.0 - Robot type from source metadata:
so100 - Source codebase version from metadata:
v2.1 - Task:
Put the red cube on top of the blue cube. - Sampling rate: 30 fps
- Source data 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 - Source card note: created using LeRobot.
At conversion time, the repository exposed 50 frame Parquet files, totaling 21,897 rows. The source meta/info.json declares 56 episodes and 22,956 frames, while the dataset card embeds an older 40-episode snapshot; this conversion follows the files actually present in the repository.
Converted Data
- TsFile path:
data/svla_so100_task4_v3_clean.tsfile - Table name:
svla_so100_task4_v3_clean - Converted rows: 21,897
- Episodes converted: 50
- Episode indexes converted:
0-19,21-36, and40-53 - Tasks: 1
- Source frame Parquet files converted: 50
- Output file size: 412,485 bytes
- Time precision: milliseconds
Schema
Time: integer timestamp in milliseconds, computed asround(timestamp * 1000).- TAG columns:
episode_index,task_index. - Integer FIELD columns:
frame_index,sample_index. - FLOAT FIELD columns:
action_0throughaction_5. - FLOAT FIELD columns:
observation_state_0throughobservation_state_5.
The six action and observation-state dimensions follow the SO100 joint names from the source metadata:
main_shoulder_panmain_shoulder_liftmain_elbow_flexmain_wrist_flexmain_wrist_rollmain_gripper
Conversion Notes
- Vector columns are flattened while preserving source names:
actionbecomesaction_0..action_5, andobservation.statebecomesobservation_state_0..observation_state_5. - The source
timestampcolumn is dropped because it is represented byTime / 1000seconds after conversion. - The source
indexcolumn is renamed tosample_index. - Video features
observation.images.topandobservation.images.wristare omitted from this repository. Use the original dataset videos atdanaaubakirova/svla_so100_task4_v3_clean/videos. - Source
meta/files are mirrored, withmeta/info.jsonupdated to describe the converted TsFile path, converted schema, row count, and video policy.
Read Example
from pathlib import Path
tsfile_path = Path("data/svla_so100_task4_v3_clean.tsfile")
print(tsfile_path)
# Read this file with the Apache TsFile Java/Python SDK or TsFile tools.
- Downloads last month
- 56