Hugging Face's logo

Datasets:
THULab
/
coop_multitask_dataset_10all

Tasks:
Robotics
Modalities:
Time-series
Size:
10K<n<100K
Tags:
LeRobot
tsfile
timeseries
format:tsfile
Dataset card Files
xet
 Community
Episodes Preview Franka Panda Visualizer
140 episodes · 10 fps

Coop Multitask Dataset 10-All (TsFile)

Converted from shivakanthsujit/coop_multitask_dataset_10all at pinned revision 6cb422f666597a34d84c689a3433d55e6a8a483a. Modalities: Time-series.

Dataset description

This pinned LeRobot v2.1 Panda dataset contains 140 manipulation episodes spanning ten task labels involving a pasta pan, pot, mugs, wine glasses, and a blue target area.

The ten source tasks cover placing the pasta pan or pot in the blue area; picking up the pasta pan or pot; picking up the leftmost/rightmost mug; placing the mug in the blue area; picking up the leftmost/“righmost” wine glass (source spelling); and placing the wine glass in the blue area.

  • Source repository owner/publisher: shivakanthsujit
  • License: not declared by the pinned source; no license is asserted here.
  • Paper/homepage/citation: no completed paper, homepage, or citation is documented in the pinned source card unless linked above.

Source metadata note: The pinned source revision has no README.md; source facts come from its meta/info.json, task metadata, data Parquets, and config. No source license, paper, homepage, citation, or author list was declared.

Dataset Scale

Split Episodes Tasks Source trajectory Parquets TsFile rows Sampling rate TsFile files/shards
train 140 10 140 15,286 10 Hz 1

The staged Parquet has 15,286 rows and 254 columns including Time; TsFile chunk metadata independently reports the same 15,286 rows.

TsFile schema

Column Role TsFile type Observed/source range
Time TIME INT64 0–19,000 ms; restarts per episode
episode_index TAG STRING source integer 0–139
task_index TAG STRING source integer 0–9
frame_index FIELD INT64 0–190 within an episode
sample_index FIELD INT64 0–15,285 globally

Exact remaining FIELD names/ranges and imported types:

  • active_robot (INT64), gripper_states, gripper_commands (FLOAT)
  • state_0–state_6, delta_actions_0–delta_actions_6 (FLOAT)
  • env_state_0–env_state_155 (FLOAT)
  • tcp_states_0–tcp_states_11, joint_states_0–joint_states_7, joint_actions_0–joint_actions_7 (FLOAT)
  • qpos_0–qpos_6, absolute_actions_0–absolute_actions_6, absolute_commands_0–absolute_commands_5 (FLOAT)
  • commanded_actions_0–commanded_actions_11, transformation_matrices_0–transformation_matrices_15 (FLOAT)

Conversion

  • All source episodes in the train split are merged into data/coop_multitask_dataset_10all_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 560 source videos across image, wrist_image, left_ego_image, and right_ego_image. 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/coop_multitask_dataset_10all_train.tsfile")
print(reader.get_all_table_schemas().keys())
reader.close()

Source and provenance

Downloads last month
37
Total file size:
12.7 MB