Mapping the World For Machines with Scaniverse

World models are advancing rapidly – but most are trained on text and images. Operating in the physical world requires something different: models with precise coordinates and geometry to make environments navigable and machine-readable. That matters for the 80% of the economy that happens outside of digital screens.

Niantic Spatial is building that foundation: a living model of the world that people and machines can talk to. Today we're launching Scaniverse for businesses as the front door to our spatial intelligence services and Large Geospatial Model.

Capturing a space and knowing exactly where you are within it are two different problems. Most companies solve one. Niantic Spatial creates both geometrically accurate and spatially grounded models that allow machines to understand and interact with the physical world.

Here’s what we’re launching:

• Scaniverse: An integrated web and mobile platform that captures 3D spaces – small and large – supporting multiple devices, to generate visual positioning maps, meshes, and Gaussian splats.

• VPS 2.0: Precise visual positioning that now works at global scale – no prior scanning required. In places mapped with Scaniverse, VPS delivers near centimeter-accurate 6DoF localization – full position and orientation. Everywhere else, it corrects GPS errors and dropout to provide improved, reliable positioning and heading, especially in GPS degraded environments.

And coming this month, Niantic Spatial Development Kit (NSDK) 4.0 is a unified SDK for Unity, Swift, Android, and ROS 2, connecting directly to Scaniverse and VPS 2.0.

Scaniverse: The Entry Point to Niantic Spatial Services

Scaniverse is our self-serve platform for delivering spatial services: it powers data capture and upload, generation of spatial assets, inspection of meshes and splats, and localization with VPS.

Scaniverse supports versatile capture of physical space ranging from individual rooms to areas of thousands of square meters.

Unlike traditional scanning platforms, Scaniverse works with regular consumer phones – and now with data from off-the-shelf 360° cameras, no expensive proprietary equipment or training required. Support for new types of data capture and formats is coming later this year.

Mobile

The updated Scaniverse app builds on a platform already used to scan millions of objects and places in the real world - one of the world's largest collections of 3D Gaussian splats.

• Single capture, multiple outputs: From a single scan – or many – users can generate precise visual positioning maps and high‑fidelity meshes and splats directly in the Scaniverse viewer.

• Collaboration: Multiple users can contribute scans to a shared project at different times on different devices. Uploads are stored and managed in the cloud, fused into a single unified model that updates as new scans are added.

• On-device previews: On‑device VPS map previews let teams validate coverage and quality immediately, even in low‑connectivity environments, so it’s easy to catch issues on site and test localization directly in the app.

The experience for existing Scaniverse users remains unchanged – you can continue scanning and processing as usual. For more info on business and personal accounts, see our FAQs.

Web

At scaniverse.nianticspatial.com teams can upload, manage and process data from the Scaniverse mobile app and 360° cameras and visualize outputs directly in the viewer.

• Large, complex areas: Upload 360° camera footage to reconstruct large areas – up to thousands of square meters – as Gaussian splats to visualize construction sites, industrial facilities, or city streets.

• Process and review assets in the Scaniverse viewer: Generate and preview meshes, splats, and location maps to use with Niantic Spatial VPS. VPS support for 360° camera footage is coming soon.

• Collaborate across organizations, with shared access to sites, scans, and processing status so field teams, engineers, and operators are always working from a current shared model of the site.

Assets can also be exported in standard 3D formats – meshes in FBX and splats in PLY and SPZ, our open-source Gaussian splat format being adopted across the industry. Files can be converted and imported into robotics simulators.

Locate with global VPS

GPS, now over 50 years old, was built for navigation and approximate positioning. These systems are often fragile – in dense urban canyons, GPS signals bounce off buildings, causing positioning errors to compound to the point where a delivery robot can't determine which side of the street it's on. And in increasingly common security scenarios, signals can be jammed or spoofed.

Our Visual Positioning System (VPS) delivers precise, real-time position and orientation using camera input and advanced computer vision. This is essential for robots navigating complex sidewalks, field technicians finding a specific valve in an oil refinery, AI glasses understanding their surroundings, or aerial drones coordinating with the ground.

Niantic Spatial's VPS delivers six degrees of freedom (6DoF) localization – full position and orientation – accurate to centimeters, compared with the 3-5m accuracy of GPS in favorable conditions, and far more precise than what GPS can provide indoors, underground, or in GPS-degraded environments.

Until now, our VPS approach has required pre-scanning locations. VPS 2.0 removes that restriction – extending coverage to an increasingly global scale, augmenting GPS, and providing three degrees of freedom (3DoF) positioning. It draws on visual context and multiple data sources to correct for GPS drift and signal dropout, delivering stable, consistent positioning where GPS alone would falter.

The result is a unified system: localize reliably anywhere with global VPS, then transition seamlessly into 6DoF localization accurate to centimeters where it matters most.

Deploy spatial applications with NSDK

With our Niantic Spatial Development Kit, developers can build spatial applications that are integrated with Scaniverse. This month, we’ll be making NSDK 4.0 generally available with expanded support for Swift, Unity, Native Android, and early support for ROS 2 (Robot Operating System 2), available on request.

This update will enable developers to leverage their existing workflows – whether building high-performance mobile apps or complex robotic systems – without being locked into a single programming environment.

Who this is for

Scaniverse and Niantic Spatial VPS are designed for use across:

• Robotics OEMs and operators – robots can lose position indoors or in GPS-degraded environments. VPS 2.0 provides continuous, accurate localization without depending solely on GPS.

• Energy, construction and logistics – complex sites are hard to navigate, slowing inspection and collaboration. Scaniverse maps them precisely, so teams and machines can work from a shared spatial model.

• Public sector – GPS-degraded or denied environments put missions at risk. VPS 2.0 delivers reliable position and heading where GPS fails.

• Large venues – Scaniverse and VPS enable persistent, location-aware spatial applications.

Today’s updates are focused on the Capture, Reconstruct, and Localize services in our Large Geospatial Model. Later this year, we’ll share more on our spatial foundation models for semantic understanding of scenes, which will enable AI to not only navigate the world but to reason about the objects and environments within it.