AI that understands the physical world
Niantic Spatial is building a living model of the world for machines
Mapping the world for machines
Niantic Spatial's Large Geospatial Model is built on real world data from ground and overhead sensors, processed and aligned to train models that deliver three services: Reconstruct, Localize, and Understand.
Reconstruct
High-fidelity, geo-referenced models of real world locations, from buildings to city-scale. We fuse multi-sensor inputs into accurate 3D models, including gaussian splats and textured meshes, enabling simulation, inspection, and planning at any scale.
Localize
Precise global positioning and orientation for humans and machines, even in GPS-degraded environments. Our system uses multiple overlapping methods to ensure robust location and pose, delivering a unified localization solution for humans, robots, and AI agents.
Understand
We're building spatial foundation models that give AI the capacity to reason about the physical world, not just navigate it. Semantic understanding at every 3D point, so humans and machines can ask open-ended questions about the spaces around them.
Research at the frontier of spatial AI
We're building the foundational models and methods for how machines can perceive and reason about the real world.
What's New
AI for the Real World
LLMs and World Models aren't enough, we're focused on the missing piece
Read more→
Scaniverse for Enterprises
Scaniverse is the front door to our geospatial AI services
Read more→
