You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

UR5 teleoperation demonstrations (ManiUniCon)

12 manipulation tasks recorded on a UR5 with a parallel gripper, teleoperated through the ManiUniCon operator dashboard and converted to LeRobot v3.0 format.

420 episodes · 149,907 frames · 30 fps

Each task lives in its own folder at the repo root; every folder is a complete, self-contained LeRobot v3.0 dataset (meta/, data/, videos/).

folder episodes frames task instruction
red2blue_cup/ 50 21,767 move the red cup to the blue one
blue2red_cup/ 50 20,527 move the blue cup to the red one
cup_bowl_2_plate/ 51 19,919 move the cup and the bowl to the plate
cup_plate_2_bowl/ 50 18,630 move the cup and the plate to the bowl
red_block_to_blue_plate/ 50 16,915 put the red block on the blue plate
blue_block_to_red_plate/ 50 16,085 put the blue block on the red plate
dice_brr/ 20 6,682 arrange the dice in blue, red, red order
dice_rbr/ 19 6,030 arrange the dice in red, blue, red order
dice_rbb/ 20 5,978 arrange the dice in red, blue, blue order
dice_brb/ 20 5,946 arrange the dice in blue, red, blue order
dice_bbr/ 20 5,854 arrange the dice in blue, blue, red order
dice_rrb/ 20 5,574 arrange the dice in red, red, blue order

Recording setup

robot UR5, 6 joints + parallel gripper
control mode joint
observation.state 7-dim — 6 joint positions + gripper state
action 7-dim — 6 target joint positions + gripper command
cameras observation.images.camera_0 (Intel RealSense D405, wrist), observation.images.camera_1 (Azure Kinect, scene)
resolution 640×480 RGB, AV1
fps 30

Loading

The folders sit one level below the repo root, so point LeRobotDataset at a downloaded folder rather than at the repo id:

from huggingface_hub import snapshot_download
from lerobot.datasets.lerobot_dataset import LeRobotDataset

root = snapshot_download("Ngseo/ur5-maniunicon", repo_type="dataset", allow_patterns="dice_bbr/*")
ds = LeRobotDataset("Ngseo/ur5-maniunicon", root=f"{root}/dice_bbr")

print(ds.num_episodes, ds.num_frames, list(ds.meta.tasks.index))

Notes

  • dice_rbr has 19 episodes rather than 20: one recording was truncated mid-write and was excluded rather than shipped corrupt.
  • The dice_* folder names encode the target colour order, e.g. dice_bbr = blue, blue, red.
Downloads last month
35

Models trained or fine-tuned on Ngseo/ur5-maniunicon