azithromycin/puneetpanwar_so101_green_cube_pickup

SO101 Green Cube Pickup TsFile

This dataset provides the numeric robot trajectories from puneetpanwar/so101_green_cube_pickup as an Apache TsFile table. The demonstrations use a LeRobot v2.1 SO101 follower robot for the task "Put the green cube into pen holder."

Source Dataset

• Author and repository owner: Puneet (puneetpanwar)
• License: Apache-2.0
• Split: train
• Scale: 49 episodes, 14,787 frames, 1 task, and 49 source Parquet shards
• Sampling frequency: 30 fps
• Robot type: so101_follower
• LeRobot codebase version: v2.1
• Source data layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• Paper and citation: the source dataset card does not provide either one

The source card contains an older embedded example showing 29 episodes. The current source meta/info.json, repository files, and data rows consistently describe 49 episodes and 14,787 frames.

Data Layout

• TsFile: data/puneetpanwar_so101_green_cube_pickup.tsfile
• Table: puneetpanwar_so101_green_cube_pickup
• Rows: 14,787
• Devices: 49, identified by the two TAG columns
• Time precision: milliseconds
• Time range within an episode: 0 to 10,100 ms

All 49 episode shards are represented in one table. Filter by episode_index and task_index to select a device trajectory.

Schema

Column	TsFile type	Role	Meaning
Time	TIMESTAMP	TIME	round(timestamp * 1000) in milliseconds
episode_index	STRING	TAG	Source episode index stored by the TsFile device/tag mechanism
task_index	STRING	TAG	Source task index stored by the TsFile device/tag mechanism
frame_index	INT64	FIELD	Frame position within the episode
sample_index	INT64	FIELD	Source global index value
action_0 ... action_5	FLOAT	FIELD	Six SO101 action components
observation_state_0 ... observation_state_5	FLOAT	FIELD	Six SO101 joint-state components

The component order for both six-element vectors is shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, and gripper.pos.

Transform Details

• Time is derived from the source timestamp and restarts at zero for each episode. The source timestamp is omitted because it is equivalent to Time / 1000 seconds.
• index is renamed to sample_index; frame_index is preserved.
• action[6] is flattened to action_0 through action_5.
• observation.state[6] is flattened to observation_state_0 through observation_state_5.
• Rows are ordered by episode_index, task_index, and Time.
• FLOAT and DOUBLE fields use GORILLA + LZ4. INT32 and INT64 fields use TS_2DIFF + LZ4. Time uses TS_2DIFF + LZ4. BOOLEAN fields, when present, use RLE + LZ4. TAG values use the TsFile table-model device/tag mechanism.

Videos

The original dataset has 98 AV1 MP4 files, with 49 files for each camera:

• videos/chunk-000/observation.images.front/episode_XXXXXX.mp4
• videos/chunk-000/observation.images.wrist/episode_XXXXXX.mp4

They remain available in the source repository's videos/ directory and are not included here. Use episode_index and frame_index to align a numeric row with the corresponding video frame.

Usage

from tsfile import ColumnCategory, TsFileReader

path = "data/puneetpanwar_so101_green_cube_pickup.tsfile"
reader = TsFileReader(path)
table_name = "puneetpanwar_so101_green_cube_pickup"
schema = reader.get_all_table_schemas()[table_name]
columns = [
    column.get_column_name()
    for column in schema.get_columns()
    if column.get_category() in (ColumnCategory.TAG, ColumnCategory.FIELD)
]

with reader.query_table(table_name, columns, batch_size=65536) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())

Attribution

The demonstrations were published by Puneet under the Apache-2.0 license and were created with LeRobot. When using this data, cite the original Hugging Face dataset URL above.