Hugging Face's logo

Datasets:
THULab
/
so100_basketball

Tasks:
Robotics
Modalities:
Time-series
Size:
10K<n<100K
Tags:
tsfile
timeseries
time-series
LeRobot
robotics
format:tsfile
License:
Dataset card Files
xet
 Community
so100_basketball / README.md
wangdx25's picture
wangdx25
Add TsFile (converted from hoon-shin/so100_basketball)
2ce712f verified
preview code
|
Raw
History Blame Contribute Delete
3.31 kB
metadata 
license: apache-2.0
task_categories:
  - robotics
tags:
  - tsfile
  - timeseries
  - time-series
  - LeRobot
  - robotics
  - format:tsfile
pretty_name: so100_basketball
size_categories:
  - 10K<n<100K

so100_basketball (TsFile)

Apache TsFile version of hoon-shin/so100_basketball.

Overview

A LeRobot robot dataset recorded on a so100 arm. Task(s): Grasp basketball and place into cup.. Each frame holds the commanded action and observed observation.state joint positions, plus camera views stored as videos in the original dataset.

  • Episodes: 50
  • Frames: 37,289
  • Sampling rate: 30 fps
  • Tasks: 1 — "Grasp basketball and place into cup."

Schema (TsFile structure)

All episodes share one TsFile with episode_index and task_index as TAG columns; query a single episode with WHERE episode_index = N.

  • Time (INT64, milliseconds) — round(timestamp * 1000); the source timestamp column is dropped (it equals Time / 1000).
  • episode_index (TAG) — device dimension.
  • task_index (TAG) — device dimension.
  • episode_index (INT64) — measurement.
  • task_index (INT64) — measurement.
  • frame_index (INT64) — measurement.
  • sample_index (INT64) — measurement.
  • action_0 (FLOAT) — measurement.
  • action_1 (FLOAT) — measurement.
  • action_2 (FLOAT) — measurement.
  • action_3 (FLOAT) — measurement.
  • action_4 (FLOAT) — measurement.
  • action_5 (FLOAT) — measurement.
  • observation_state_0 (FLOAT) — measurement.
  • observation_state_1 (FLOAT) — measurement.
  • observation_state_2 (FLOAT) — measurement.
  • observation_state_3 (FLOAT) — measurement.
  • observation_state_4 (FLOAT) — measurement.
  • observation_state_5 (FLOAT) — measurement.

The vector columns are flattened per joint:

  • action_* — commanded joints: main_shoulder_pan, main_shoulder_lift, main_elbow_flex, main_wrist_flex, main_wrist_roll, main_gripper.
  • observation_state_* — observed joints: main_shoulder_pan, main_shoulder_lift, main_elbow_flex, main_wrist_flex, main_wrist_roll, main_gripper.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("data/so100_basketball.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license