vials_recap_v1.1 TsFile

This dataset is converted from Sichang0621/vials_recap_v1.1 to Apache TsFile format.

Modalities: Time-series. The converted repository contains numeric robot state/action trajectories, scalar labels, and LeRobot metadata. Camera videos are not included in this TsFile conversion; they remain in the original Hugging Face dataset under its videos/ tree.

Source Dataset

• Original dataset: Sichang0621/vials_recap_v1.1
• License: apache-2.0
• Robot type from source metadata: yam
• Source codebase version from metadata: v3.0
• Collection setup: teleoperation data collected with limb on yam arms.
• Task: pick up all vials and place them in the stand
• Source split: train, episodes 0:93
• Sampling rate: 30 fps
• Source scale: 93 episodes, 105,592 frames, 1 task.
• Source data 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
• Source cameras: head_camera, left_wrist_camera, right_wrist_camera
• Source video codec: hevc

Converted Data

• TsFile path: data/vials_recap_v1_1.tsfile
• Table name: vials_recap_v1_1
• Converted rows: 105,592
• Episodes converted: 93
• Tasks: 1
• Source frame Parquet files converted: 93
• Output file size: 9,250,447 bytes
• Time precision: milliseconds

Schema

• Time: integer timestamp in milliseconds, computed as round(timestamp * 1000).
• TAG columns: episode_index, task_index.
• Integer FIELD columns: frame_index, sample_index, correction_index, intervention.
• FLOAT FIELD columns: reward, reward_label, value_label.
• STRING FIELD columns: phase, adv_ind.
• FLOAT FIELD columns: observation_state_0 through observation_state_13.
• FLOAT FIELD columns: action_0 through action_13.

The 14 state and action dimensions follow the source metadata order: left_joint_0, left_joint_1, left_joint_2, left_joint_3, left_joint_4, left_joint_5, left_gripper, right_joint_0, right_joint_1, right_joint_2, right_joint_3, right_joint_4, right_joint_5, right_gripper.

Conversion Notes

• Vector columns are flattened while preserving source names: observation.state becomes observation_state_0..observation_state_13, and action becomes action_0..action_13.
• The source timestamp column is dropped because it is represented by Time / 1000 seconds after conversion.
• The source index column is renamed to sample_index.
• Video features observation.images.head_camera, observation.images.left_wrist_camera, and observation.images.right_wrist_camera are omitted from this repository. Use the original dataset videos at Sichang0621/vials_recap_v1.1/videos.
• Source meta/ files are mirrored, with meta/info.json updated to describe the converted TsFile path and schema.

Read Example

from pathlib import Path

tsfile_path = Path("data/vials_recap_v1_1.tsfile")
print(tsfile_path)

# Read this file with the Apache TsFile Java/Python SDK or TsFile tools.