Hugging Face's logo

Datasets:
jzpeng
/
handshake_data

Tasks:
Robotics
Modalities:
Time-series
Tags:
LeRobot
so100
tutorial
tsfile
timeseries
License:
Dataset card Files
xet
 Community
handshake_data / README.md
jzpeng's picture
jzpeng
Add TsFile (converted from advpatel/handshake_data)
7572136 verified
preview code
|
Raw
History Blame Contribute Delete
3.9 kB
metadata 
license: apache-2.0
task_categories:
  - robotics
tags:
  - LeRobot
  - so100
  - tutorial
  - tsfile
  - timeseries
modality:
  - timeseries
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/handshake_data_train.tsfile

Handshake Data (TsFile)

Converted from advpatel/handshake_data at pinned revision c64b79ec9dd64531a42715d921d1b736d2fa35ff. Modalities: Time-series.

Dataset description

This LeRobot v2.1 SO100 dataset records demonstrations of a robot shaking a person's hand.

The pinned task table contains one instruction: “Shake the person's hand.”

  • Source repository owner/publisher: advpatel
  • License: apache-2.0 (declared by the source card).
  • Paper/homepage/citation: no completed paper, homepage, or citation is documented in the pinned source card unless linked above.

Source metadata note: The source README embeds an older one-episode/353-frame snapshot with a different camera entry. Counts and camera names here use the pinned meta/info.json, 50 source episode Parquets, staged rows, and TsFile metadata.

Dataset Scale

Split Episodes Tasks Source trajectory Parquets TsFile rows Sampling rate TsFile files/shards
train 50 1 50 15,779 30 Hz 1

The staged Parquet has 15,779 rows and 17 columns including Time; TsFile chunk metadata independently reports the same 15,779 rows.

TsFile schema

Column Role TsFile type Observed/source range
Time TIME INT64 0–14,333 ms; restarts per episode
episode_index TAG STRING source integer 0–49
task_index TAG STRING source integer 0–0
frame_index FIELD INT64 0–430 within an episode
sample_index FIELD INT64 0–15,778 globally

Exact remaining FIELD names/ranges and imported types:

  • action_0–action_5 (FLOAT): shoulder pan/lift, elbow flex, wrist flex/roll, and gripper action
  • observation_state_0–observation_state_5 (FLOAT): the corresponding SO100 state

Conversion

  • All source episodes in the train split are merged into data/handshake_data_train.tsfile; episode_index and task_index are TAG dimensions.
  • Time = round(timestamp * 1000) in milliseconds. The source timestamp column is omitted because it is redundant with Time / 1000 seconds.
  • frame_index is retained. Source index is retained as sample_index.
  • Every vector is fully flattened: the complete source name is kept, . becomes _, and element indices are appended. Float vectors are imported as single-precision FLOAT fields.
  • Other scalar source columns shown above are retained; no trajectory rows are intentionally dropped.
  • Source metadata is mirrored for publication, with copied meta/info.json rewritten to describe the converted data path and conversion semantics.

Video policy

The pinned metadata declares 100 source videos across tripod and arm cameras. MP4 files were not downloaded or uploaded; they remain in the pinned source videos/ tree.

Minimal read example 

from tsfile import TsFileReader

reader = TsFileReader("data/handshake_data_train.tsfile")
print(reader.get_all_table_schemas().keys())
reader.close()

Source and provenance