3D World Simulator predicts robot touch at 50+ FPS in unified 3D space

YunzhuLiYZ · x · 2026-11-03

Researchers unveiled the 3D World Simulator, an interactive world model running at 50+ FPS on an RTX 5090 that predicts scene evolution in a unified 3D representation space, generating novel-view appearance, multi-view-consistent RGB-D, 3D geometry and spatially aligned tactile signals. Unlike video world models, it explicitly models 3D scenes, so renders stay consistent from any camera and support background swaps, different robot embodiments and tactile data. Learning from predicted visual-tactile observations (FlexiTac data) enables real-world contact-rich manipulation. Website, code and data are all open-sourced.

Related event: 3D World Simulator Debuts: Interactive World Model Predicting Appearance, Geometry and Tactile at 50+ FPS(5 posts)→

Original post →

More from Embodied

Embodied channel →