CLVR Jaco Play Dataset Overview

Basic Information

• Name: CLVR Jaco Play Dataset
• Number of Segments: 1,085 remote‑operated robot segments
• Content: Language annotations
• Data Collection Frequency: 10 Hz

Data Download

• Download Method:
  • Use the provided script: bash download_jaco_data.sh ./data
  • Manual download link: https://drive.google.com/file/d/1tVSilmXhyQd8gxZAEhvKMnynw0qzRFSZ/view?usp=share_link
• Data Size: Approximately 16 GB

Data Structure

• Observations:
  • front_cam_ob: Observation from a third‑person camera
  • mount_cam_ob: Observation from a fixed camera
  • ee_cartesian_pos_ob: End‑effector Cartesian position (first 3 elements are position, last 4 are orientation as a quaternion)
  • ee_cartesian_vel_ob: End‑effector Cartesian velocity (first 3 elements are positional velocity, last 3 are angular velocity in roll‑pitch‑yaw format)
  • joint_pos_ob: Joint positions of the Jaco arm (only the last 2 elements correspond to the gripper joint)
• Action Data:
  • First 3 elements are Cartesian increments
  • 4th element is a label: 0 (open gripper), 1 (no gripper movement), 2 (close gripper)
• Termination Flag: 1 at the end of each skill
• Prompt: Natural language description of the goal
• Reward: 1 at the end of each skill

Citation

bibtex @software{dass2023jacoplay, author = {Dass, Shivin and Yapeter, Jullian and Zhang, Jesse and Zhang, Jiahui and Pertsch, Karl and Nikolaidis, Stefanos and Lim, Joseph J.}, title = {CLVR Jaco Play Dataset}, url = {https://github.com/clvrai/clvr_jaco_play_dataset}, version = {1.0.0}, year = {2023} }

License

• License Type: Creative Commons Attribution 4.0 International License
• License Link: https://creativecommons.org/licenses/by/4.0/