XR AI Spotlight Geospatial AI Explained by Niantic Spatial
Feb 4, 2026
Asim Ahmed, Head of Product Marketing at Niantic Spatial and multi-award-winning developer behind Pokémon GO and Peridot, discusses building large geospatial models. He breaks down reconstruction, localization and semantic understanding. He covers applications from entertainment to robotics, partnerships with major platforms, and embodied AI companions like Project Jade.
AI Snips
Chapters
Transcript
Episode notes
Large Geospatial Model Gives Machines Street Smarts
- Niantic Spatial builds a "large geospatial model" to give machines "street smarts" about the physical world.
- The model aims to let AI understand, interact, and navigate real-world spaces with human-like awareness.
Three Foundations: Reconstruct, Localize, Understand
- The model has three core capabilities: reconstruction, localization, and semantic understanding.
- Together they enable high-fidelity digital twins, centimeter-level positioning, and per-pixel semantic queries of the world.
VPS Is Part Of A Bigger Predictive System
- Niantic's VPS is the localization layer within the larger geospatial model, not a separate product.
- The geospatial model can regularize noisy or limited scans using massive prior scan data to predict accurate real-world structure.