zjt24/14_parcelot_50_ep

14-parcelot-50-ep

This dataset is a TsFile conversion of the Hugging Face dataset LeRobot-worldwide-hackathon/14-parcelot-50-ep, which was created using LeRobot.

Modalities: Time-series. The original dataset also contains robot camera videos; those videos are not included in this converted repository and remain available from the original Hugging Face dataset.

Source Dataset

• Original dataset: LeRobot-worldwide-hackathon/14-parcelot-50-ep
• License: apache-2.0
• Robot type: bimanual_parcelot
• Source format: LeRobot v2.1 episode Parquet files plus metadata and videos
• Source scale: 29 episodes, 25,609 frames, 1 task, 1 data chunk, 87 videos
• Sampling rate: 30 fps
• Split: train (0:29)

Task: pick a white parcel with the left hand from the black box, hand it off to the right hand, and place it into the white box on the right.

Converted Files

• TsFile data: data/14_parcelot_50_ep.tsfile
• Mirrored metadata: meta/
• Table name: 14_parcelot_50_ep
• Time precision: milliseconds
• Converted row count: 25,609
• Source episode shard count: 29 Parquet files

The original source data path is data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet. The converted TsFile path is data/14_parcelot_50_ep.tsfile.

Schema

Time is synthesized as round(timestamp * 1000) and restarts within each episode, matching the source LeRobot timestamp semantics.

TAG columns:

• episode_index
• task_index

FIELD columns:

• frame_index
• sample_index (renamed from source index)
• action_0 to action_11
• observation_state_0 to observation_state_11

The source motor names for both action and observation.state are:

• left_shoulder_pan.pos
• left_shoulder_lift.pos
• left_elbow_flex.pos
• left_wrist_flex.pos
• left_wrist_roll.pos
• left_gripper.pos
• right_shoulder_pan.pos
• right_shoulder_lift.pos
• right_elbow_flex.pos
• right_wrist_flex.pos
• right_wrist_roll.pos
• right_gripper.pos

Conversion Notes

The source action[12] and observation.state[12] vector columns are flattened into scalar FLOAT fields. Dots in source column names are replaced with underscores, so observation.state becomes observation_state_0..11.

The source timestamp column is not retained as a separate field because it is redundant with Time / 1000 seconds. The source index column is renamed to sample_index.

The source video features are omitted from this repository:

• observation.images.top
• observation.images.left_wrist
• observation.images.right_wrist

Videos remain in the original dataset at: https://huggingface.co/datasets/LeRobot-worldwide-hackathon/14-parcelot-50-ep/tree/main/videos

The meta/info.json file is mirrored and updated with a tsfile_conversion object that records the source dataset, converted data path, table name, time mapping, TAG columns, flattened features, omitted video features, and row count.

Minimal Read Example

from tsfile import TsFileReader

path = "data/14_parcelot_50_ep.tsfile"

with TsFileReader(path) as reader:
    schemas = reader.get_all_table_schemas()
    print(schemas.keys())