Skip to content

Dataset · lerobot

Bimanual robot trajectories with subtask labels

lerobot/FineART

Large teleoperation video dataset of a bimanual robot with detailed manual step labels for training robotics models.

What's inside

It contains 40,543 video episodes (1,718 hours) of bimanual robot operation using a Trossen Stationary AI system (similar to ALOHA), covering 151 tasks. The data was collected with human operators at Scale AI's Physical AI division in collaboration with Hugging Face LeRobot. Each episode is equipped with detailed manual subtask annotations, totaling 533,913 labels in the set.

What you can do with it

  • Training bimanual manipulator control policies and VLA models
  • Training hierarchical planning and task decomposition algorithms for robots
Tasks
Robotics
Size
10K–100K rows
Downloads this month
1.2M
Likes
46

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("lerobot/FineART")