Ground Truth: Visual Positioning With VPS 2.0 When GPS Fails
For decades, the question "where am I?" has had one answer: GPS. It is a feat of engineering, but built for a different era, one of open skies and approximate positioning. Today's autonomous systems often operate in environments where satellite signals are obstructed, degraded, or simply absent.
GPS is fragile. In urban canyons signals bounce and positions drift. Indoors they can vanish. Underground and in rugged terrain they often disappear entirely. And in an increasingly common set of scenarios – from conflict zones to critical infrastructure – signals can be jammed or spoofed entirely.
Niantic Spatial’s Visual Positioning System (VPS), a key service powered by the Large Geospatial Model we are building, is designed to solve these challenges.
Today, we’re introducing VPS 2.0.
The gap we set out to close
While GPS triangulates location from satellite signals, VPS analyzes what a camera sees. It recognizes surfaces, structures, and spatial features and matches them to detailed visual maps to determine precisely where a device is and which way it's facing.
Our original VPS delivered – and still delivers – six degrees of freedom (6DoF) localization with near centimeter-level precision. It works indoors and across a range of scanned outdoor locations.
The limitation of this technology was scale. To achieve this precise positioning, a space had to be pre-scanned and mapped. That's acceptable for known and controlled environments like factories, warehouses, industrial facilities or small outdoor environments.
But what about everywhere else? What about when an autonomous system ventures beyond the boundary of its mapped zone? What about operations in unfamiliar terrain, or rapidly evolving environments where pre-mapping just isn't possible?
This is the gap that VPS 2.0 aims to close.
Localization at global scale
VPS 2.0 introduces two complementary modes that work together seamlessly:
- Existing 6DoF precision with prior scanning: Near centimeter-level, full position and orientation – in environments that have been pre-scanned and mapped. This is what our original VPS delivered, and VPS 2.0 makes it faster and easier to deploy through integration with Scaniverse, our new self-service capture platform. The scope of environments covered with full 6DoF precision is set to expand over the course of this year as we add new capture sources.
-
Expanded 3DoF coverage without prior scanning: Reliable three-degrees-of-freedom positioning at a global scale, drawing on visual context and multiple data sources to correct for GPS drift and signal dropout without requiring site-specific scanning. The system draws on multiple real-world data sources:
- Basemap: Traditional 2D and 3D geospatial datasets, including 2D maps and 3D terrain and surface models.
- Scanned: Scans from mobile and aerial platforms.
- Device-level inputs: Accelerometer, magnetometer, real-time video, and LIDAR (when available).
Our solutions for 6DoF precision and broad 3DoF coverage work as a unified system. An autonomous platform – whether a ground robot, drone, or mixed reality headset – localizes reliably at an increasingly global scale, and transitions seamlessly into more precise localization when it enters a mapped zone.
VPS 2.0 operates across a spectrum of localization modes, each unlocking different capabilities. Where no spatial map or pre-scanned data exists, the system provides reliable position and heading - enough for navigation and coordination. In environments that have been pre-scanned, VPS 2.0 delivers the centimeter-scale precision required for precise navigation in complex environments. VPS 2.0 enables devices to move between these modes automatically, without user intervention, as map coverage changes.
Resilient, robust and redundant localization services
Traditional approaches to fixing GPS gaps relied on expensive, dedicated hardware or methods that fail under changing lighting, weather, or environmental conditions. Niantic Spatial VPS takes a different approach: redundancy and robustness through a flexible deep learning-based architecture complemented by real-time computer vision and on-device sensor fusion.
Increasingly, GPS is unreliable precisely where it’s needed most: dense urban environments, inside complex structures, and contested terrain. VPS 2.0 provides a resilient layer of visual positioning that complements GPS when satellite signals are degraded, obstructed, or unavailable. This creates a far more resilient positioning layer for autonomous systems operating in the real world.
| Feature | Niantic Spatial VPS | Others |
|---|---|---|
| Data Sources |
Deep learning based architecture enables localization from a variety of 2D and 3D data sources. Flexible inputs |
Typically require pre-mapping using proprietary software or use of closed proprietary datasets. Locked ecosystem |
| Robustness |
Industry-leading research underpins a redundant deep-learning based system that ensures robust performance. All conditions |
Often rely on brittle CV approaches that may not be reliable in varying lighting and weather conditions. Environment-sensitive |
| Pipeline |
Pipeline produces localization maps and 3D digital twins with semantic content. Multi-purpose output |
Pipelines are typically single purpose for localization and produce only limited mesh outputs. Limited output |
VPS 2.0 closes the gap between precision and scale in geolocation. By combining near centimeter-level positioning with global visual coverage, Niantic Spatial delivers a resilient visual positioning layer that complements GPS when signals degrade or disappear. It’s a new positioning layer for the autonomous systems and spatial applications that will define the next generation of real-world computing.
Who VPS 2.0 is built for
VPS 2.0 delivers a resilient positioning layer for systems that can’t afford to rely on GPS alone. That includes public sector and public safety teams operating in GPS-contested or GPS-denied environments; autonomous platforms navigating dense urban infrastructure; and industrial operators running precision workflows in facilities that satellite signals can’t reach.