Jul 7, 2026Industry

Your phone splats are now robot training data

A humanoid robot walks through a gaussian-splat scan of a city street, with a depth-map overlay in the corner

Image: Niantic Spatial · Industry

Scaniverse can now export a USDZ file pairing a gaussian splat with an automatically generated, aligned mesh — so a single capture produces both the look and the physics of a place.

Why it matters: This ties consumer-grade splat capture directly into the physical AI and robotics simulation economy — one of the strongest commercial pulls for 3D reconstruction right now. Robotics teams call the problem "sim-to-real": robots trained in simulators that don't match reality fail in the field. Splats of real places narrow that gap.

Zoom in:

The exported digital twins import into NVIDIA Isaac Sim for robot training and testing.
The workflow runs on a roughly $500 360° camera — positioned against RGB-LiDAR mapping rigs that can cost tens of thousands of dollars.
The capability is live in Scaniverse now, building on Niantic Spatial's April launch of the integrated Scaniverse platform and VPS 2.0.

Go deeper:

⟵ Back to the brief