THULab/fast_pick_and_place_30

Fast Pick and Place 30 TsFile

This dataset is a TsFile conversion of bartek-niedzielski/fast_pick_and_place_30, a LeRobot v2.1 robotics dataset created with LeRobot. The original task is: place the green cube in the yellow area.

Modalities: Time-series. The converted repository contains numeric robot state, action, timing, and task metadata in Apache TsFile format. The original videos are not duplicated here.

Source Dataset

• Original dataset: bartek-niedzielski/fast_pick_and_place_30
• Source author: bartek-niedzielski
• Source license: apache-2.0
• Source format: LeRobot v2.1 Parquet episodes plus MP4 videos
• Split: train: 0:30
• Scale from source meta/info.json: 30 episodes, 7,200 frames, 1 task, 1 chunk, 60 videos
• Sampling rate from source meta/info.json: 15 fps
• Source video streams: exterior_image_1_left and wrist_image_left, both 480x640x3, AV1, 15 fps

Converted Files

• TsFile path: data/fast_pick_and_place_30.tsfile
• Table name: fast_pick_and_place_30
• Rows: 7,200
• Time precision: milliseconds
• Metadata: source meta/ files are mirrored under meta/; meta/info.json is rewritten so data_path points to the converted TsFile and includes a tsfile_conversion object.

Schema

Time is the TsFile time column. The source episode_index and task_index columns are declared as TAG columns so each robot episode/task can be filtered as device metadata.

Column	Role	Type	Notes
Time	TIME	INT64	round(timestamp * 1000) milliseconds
episode_index	TAG	INT64	Source episode index
task_index	TAG	INT64	Source task index
frame_index	FIELD	INT64	Source frame index, retained
sample_index	FIELD	INT64	Source index renamed
gripper_position	FIELD	FLOAT	Scalar gripper position
joint_position_0..joint_position_6	FIELD	FLOAT	Flattened from joint_position[7]
actions_0..actions_7	FIELD	FLOAT	Flattened from actions[8]
state_0..state_7	FIELD	FLOAT	Flattened from state[8]

Conversion Notes

• Converted with the generic LeRobot converter in merged granularity: all 30 episodes are stored in one TsFile table.
• Time = round(timestamp * 1000) in milliseconds. The source timestamp column is dropped because it is redundant with Time / 1000 seconds.
• The source index column is renamed to sample_index.
• Vector columns are flattened by preserving the source column name and appending the element index.
• Source videos are not uploaded to this repository. They remain available in the original dataset under videos/.
• Aside from the redundant timestamp column and omitted video payloads, no source numeric time-series rows are intentionally dropped.

Validation

The converted file was validated locally with the project pipeline:

• pipeline.py validate: non-empty TsFile found at data/fast_pick_and_place_30.tsfile
• TsFile readback check: 7,200 rows from TsFile metadata and query path
• Staged Parquet comparison: 7,200 rows

Minimal Read Example

from tsfile import TsFileReader

path = "data/fast_pick_and_place_30.tsfile"
with TsFileReader(path) as reader:
    schemas = reader.get_all_table_schemas()
    print(schemas.keys())