THULab/pick_tblock_mp_pdscale_10

Pick TBlock MP PDScale 10 TsFile

This dataset is a TsFile conversion of younghyopark/pick_tblock_mp_pdscale_10, a LeRobot v2.1 dataset created with the DualPanda robot setup.

Modalities: Time-series.

Source Dataset

• Original dataset: younghyopark/pick_tblock_mp_pdscale_10
• Created with: LeRobot
• License: apache-2.0
• Codebase version: v2.1
• Robot type: DualPanda
• Source task: pick something up (the source tasks.jsonl string contains one trailing space)
• Source split: train 0:1000
• Source scale: 1000 episodes, 173,000 frames, 1 task, 1 data chunk, 50 fps
• Source videos: 0 videos; source video_path is null

Converted Data

• TsFile path: data/pick_tblock_mp_pdscale_10.tsfile
• Table name: pick_tblock_mp_pdscale_10
• Rows: 173,000
• Episodes: 1000
• Time precision: milliseconds
• Layout: all train episodes are stored in one TsFile, using TsFile table TAG columns for episode and task identity

Schema

Time is computed as round(timestamp * 1000) in milliseconds and restarts per episode. The source timestamp column is not retained as a FIELD because it is represented by Time / 1000 seconds.

TAG columns:

• episode_index
• task_index

Scalar FIELD columns:

• frame_index
• sample_index converted from the source index column

Flattened vector FIELD groups:

• observation.joint -> observation_joint_0 through observation_joint_15
• observation.state -> observation_state_0 through observation_state_29
• observation.environment_state -> observation_environment_state_0 through observation_environment_state_6
• observation.qvel -> observation_qvel_0 through observation_qvel_29
• observation.env_qvel -> observation_env_qvel_0 through observation_env_qvel_5
• observation.ee_pose -> observation_ee_pose_0 through observation_ee_pose_15
• action.ctrl -> action_ctrl_0 through action_ctrl_15
• action.joint -> action_joint_0 through action_joint_15
• action.ee_pose -> action_ee_pose_0 through action_ee_pose_15

All flattened vector values are stored as scalar FLOAT fields. The flattening preserves the source column name prefix, replaces . with _, and appends the element index.

Conversion Notes

• Source timestamp was used to create Time and was dropped as a redundant field.
• Source index was renamed to sample_index.
• Source columns episode_index and task_index are TAG columns, not regular fields.
• Source metadata is mirrored under meta/, with meta/info.json updated to describe the TsFile artifact.
• No videos are uploaded here because the source metadata reports total_videos: 0 and video_path: null.

Read Example

The converted file is data/pick_tblock_mp_pdscale_10.tsfile. After loading it with Apache TsFile or IoTDB tooling, query one episode by TAG:

SELECT frame_index, observation_state_0, action_ctrl_0
FROM pick_tblock_mp_pdscale_10
WHERE episode_index = 0 AND task_index = 0
ORDER BY Time;