Skip to content

Dataset · InternRobotics

Multimodal dataset for 4D world modeling

InternRobotics/OmniWorld

Large multi-domain image-text corpus for 4D world modeling and robotics.

What's inside

OmniWorld is a multi-domain and multimodal dataset for 4D world modeling accepted at ICLR 2026. The card mentions OmniWorld-Game with metric scale, RH20T-Robot, RH20T-Human, Ego-Exo4D, EgoDex, and Epic-Kitchens. Modalities are image and text, language is English, and metadata size is 1B-10B.

What you can do with it

  • Training text-to-video and image-to-video models
  • 3D/4D scene reconstruction and robotic policy training
  • Evaluating models like Pi3X across different domains
Tasks
Text → video, Image → video, Image → 3d, Robotics, Other
Languages
English
Size
1B–10B rows
Downloads this month
35.4K
Likes
103

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("InternRobotics/OmniWorld")