Pouring (TsFile)

This is a time-series conversion of garySue/pouring, a LeRobot v2.0 robotics dataset containing demonstrations of the pouring task with a dual-arm bi_ur5 robot.

• Modalities: Time-series
• Robot: bi_ur5
• Task: pouring
• License: Apache License 2.0

Dataset scale

Split	Episodes	Frames / TsFile rows	Sampling rate	Tasks	Source Parquets	Final TsFiles
train	80	16,818	30 fps	1	80	1

All numeric frame data is merged into exactly one file, data/pouring.tsfile, containing the table pouring. The file is approximately 2.59 MB.

TsFile schema

Column	TsFile role	Type	Description
Time	TIME	INT64	Milliseconds, computed as Time = round(timestamp * 1000); time restarts within each episode.
episode_index	TAG	INT64	Original episode identifier (0 through 79).
task_index	TAG	INT64	Original task identifier (0 for the single pouring task).
frame_index	FIELD	INT64	Original frame position within the episode.
sample_index	FIELD	INT64	Original global index, renamed to avoid ambiguity.
action_0 ... action_13	FIELD	FLOAT (float32)	Complete 14-element source action vector.
observation_state_0 ... observation_state_13	FIELD	FLOAT (float32)	Complete 14-element source observation.state vector.
observation_velocity_0 ... observation_velocity_13	FIELD	FLOAT (float32)	Complete 14-element source observation.velocity vector.
observation_gripper_position_0 ... observation_gripper_position_1	FIELD	FLOAT (float32)	Complete two-element source observation.gripper_position vector.

The source timestamp column is dropped after generating Time because it is the redundant seconds representation of the same time coordinate (Time / 1000). No rows or other numeric source columns are dropped. All four vectors are flattened without truncation, index is renamed, and the episode, task, and frame identifiers remain available.

Videos and metadata

The source contains three 480 x 640 RGB H.264 camera streams: observation.images.top_rgb, observation.images.left_rgb, and observation.images.right_rgb. Their 240 MP4 files are intentionally omitted from this time-series repository and remain in the original dataset's videos/ tree.

Numeric rows retain episode_index, frame_index, and the 30 fps time coordinate, so they remain aligned with the corresponding source video frames. The source meta/ files are mirrored, except that meta/info.json is rewritten to describe the single TsFile path, TIME/TAG/FIELD schema, flattened and renamed features, omitted video features, and source-video alignment. No videos/ directory is included here.

Reading the data

Install the Apache TsFile Python package, then query the actual table. query_table returns the time column automatically in addition to the requested TAG and FIELD columns.

from tsfile import TsFileReader

reader = TsFileReader("data/pouring.tsfile")
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "observation_state_0",
    "observation_velocity_0",
    "action_0",
]

with reader.query_table("pouring", columns, batch_size=65536) as result:
    batch = result.read_arrow_batch()
    if batch is not None:
        print(batch.to_pandas().head())

Source and citation

The source dataset was created with LeRobot and is published at garySue/pouring. Its dataset card does not provide a paper or BibTeX citation; consult the source repository for any future citation updates.