Skip to content

Dataset · genrobot2025

Robot Data: 13k Hours of Manipulation

genrobot2025/10Kh-RealOmin-OpenData

Robot manipulation video corpus: 13k hours, 5M clips, and everyday scenarios for embodied AI training.

What's inside

This is a corpus of bimanual robot manipulation videos and trajectories: featuring over 13,000 hours, 5 million+ clips, and frame alignment with under 1 ms latency. Collected from 10,000+ real everyday scenarios with contributions from 3,000+ participants; the card lists en and zh languages. It is an open corpus for embodied AI, positioned as the industry's largest.

What you can do with it

  • Train policies for robotics RL
  • Evaluate embodied agents on everyday scenarios
  • Fine-tune LoRA for manipulation understanding
Tasks
Robotics, Reinforcement learning
Languages
English, Chinese
Size
over 1T rows
Downloads this month
196K
Likes
281
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("genrobot2025/10Kh-RealOmin-OpenData")