THULab/KuphDev_place_green_duck_pruned_3rd

Place Green Duck Pruned 3rd TsFile

This dataset is an Apache TsFile conversion of KuphDev/place_green_duck_pruned_3rd, a LeRobot v3.0 SO follower manipulation dataset for the task "Place green duck on target".

Modalities: Time-series. The converted repository contains numeric robot observations, actions, frame timing, and episode/task tags. The original camera videos remain in the source Hugging Face dataset.

Source Dataset and Author

• Source dataset: KuphDev/place_green_duck_pruned_3rd
• Repository owner and uploader: KuphDev
• Author note: the source card provides no separate author list, so this card attributes the verifiable repository owner/uploader and does not invent additional authors.
• License: Apache-2.0
• Robot type: so_follower
• LeRobot codebase version: v3.0
• Split: train
• Scale: 182 episodes, 145,378 frames, 1 task, 30 fps
• Source frame files: 39 Parquet shards under data/chunk-000/
• Source frame layout: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
• Source video layout: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
• Task: Place green duck on target. Paper/citation: not supplied by the source dataset card

The source meta/tasks.parquet maps task_index = 0 to Place green duck on target.

TsFile Schema

Time is synthesized as round(timestamp * 1000) milliseconds and restarts from 0 for each episode.

TAG columns:

• episode_index
• task_index

Scalar FIELD columns:

• frame_index
• sample_index, renamed from source column index

Flattened FLOAT FIELD groups:

• action[6] -> action_0 ... action_5
• observation.state[6] -> observation_state_0 ... observation_state_5

For both vectors, the six source dimensions are, in order: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, and gripper.pos.

Conversion Notes

• The shared config-driven lerobot converter is used; the dataset-specific convert_kuphdev_place_green_duck_pruned_3rd.py script is a local orchestration and validation entry point.
• The train split is merged into one table-model TsFile. Filter by episode_index and task_index to select an episode or task.
• Vector columns are flattened to scalar TsFile fields. Full source prefixes are preserved, with . replaced by _.
• The source timestamp column is dropped after Time synthesis because it is redundant with Time / 1000 seconds.
• Source index is renamed to sample_index; frame_index is retained.
• All 39 source Parquet shards share the same schema. No numeric row, episode, task index, action dimension, or observation-state dimension is dropped.

Videos

Videos are not duplicated in this converted repository. The single frame-aligned top-camera stream remains in the original dataset:

• videos/observation.images.top/chunk-000/

The source path template is videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4. Source metadata describes 1280x720 H.264 video at 30 fps without audio. The remote repository contains 39 MP4 files (288,387,551 bytes), from file-000.mp4 through file-038.mp4. Numeric rows remain aligned with the original videos through episode_index, frame_index, and the source episode metadata under meta/episodes/.

Usage

from tsfile import TsFileReader

path = "data/kuphdev_place_green_duck_pruned_3rd_train.tsfile"
reader = TsFileReader(path)
table_name = "kuphdev_place_green_duck_pruned_3rd_train"
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_0",
    "observation_state_0",
]

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