Mundo turns a few room photos into an explorable, simulatable 3D world — at 1.5 FPS

mark_k · x · 2026-10-27

Mundo, a research project from MMLab NTU, Nanyang Technological University and ACE, turns a few photos of a room into a world you can explore and simulate: it reconstructs a 3D scene, represents objects and physical properties in code, and combines a simulator with a video model — demos range from opening drawers to navigating a robot through a reconstructed hotel. Far from real-time: reconstruction takes 10–20 minutes and rendering runs at 1.5 FPS on eight GPUs.

Original post →

More from Embodied

Embodied channel →