Introducing the Places Library | Niantic Spatial, Inc.

An embodied AI policy is only as good as the environments it learns from. A policy that performs well in one clean, familiar setting can still fail the moment floor plans, lighting, or operating conditions change because it never learned to handle that variation.

We’re introducing the Places Library, starting with a curated catalog of 100 high-fidelity, real-world 3D environments, delivered as simulation-ready assets, to give teams a far broader range of places to train and evaluate their policies. The initial catalog spans four common deployment environments – industrial environments, last-mile and logistics settings, commercial spaces, and residential environments.

We are just getting started. The library will keep growing, with new environments added in response to what the community asks for.

Grounded in real-world capture, the library advances visual-spatial intelligence by exposing policies to the messy, variable conditions they'll encounter in deployment. Access comes through our Embodied AI plans, with the number of places you can access scaling by plan.

The bridge between visual fidelity and accurate geometry

A place in simulation needs to look right, but it also needs to behave consistently. What a policy sees should match the surfaces it can collide with.

That's why each Places Library environment ships as a single USDZ package pairing a photorealistic Gaussian splat with an aligned collision mesh, both derived from the same reconstruction. The splat carries the visual richness; the mesh provides the physical structure for interaction. The asset arrives gravity-aligned, metric-scale, and collider-ready for robot simulation and policy training.

Depth is where our reconstruction stands apart, and it's what customers tell us they value most. Photometric reconstruction alone struggles with the surfaces that matter for robotics – white walls, glass, and other low-texture areas where a policy can least afford phantom obstacles. Our MVSAnywhere zero-shot multi-view depth fills exactly those gaps, producing cleaner, more consistent geometry at true metric scale, from an ordinary 360 camera with no LiDAR. The result is a collision mesh a policy can trust.

Because both layers come from one source, the visual scene and the physics layer stay in register. There's no manual cross-registration between an appearance model and a separate collision environment, and no drift between what a policy sees and the surfaces it can hit. Policies never perceive mesh artifacts as real obstacles, so nothing phantom carries into the real world.

And it fits the tools teams already use: the workflow runs in NVIDIA Isaac Sim, NVIDIA Isaac Lab, and compatible OpenUSD simulators.

How you can get started

Start with a free sample scene, and contact our team to sign up for a plan to access more. Once you're on a plan, you can select any place based on the environment and use case you want to evaluate. Then:

Every plan pairs Places Library access with capture minutes you can spend reconstructing your own sites into custom, simulation-ready scenes, so you can extend your training data with the environments that matter most to your team.

We can also relight selected scenes through our Gaussian Splat Relighting beta, generating new lighting and weather variants from the same capture.

Real places as a foundation for domain randomization

The Places Library is an early step toward a larger idea: using real places as the foundation for domain randomization in embodied AI. We're building this for both Places Library environments and the custom scenes teams capture at their own sites.

A single capture shouldn't have to represent a single training condition. A growing catalog of places, together with the environments you capture yourself, gives teams variation across geometry, materials, lighting, and layouts without building every environment from scratch. We're exploring how both catalog and custom-scene environments can support the full real-to-sim-to-real loop: reconstructing deployment sites, training and evaluating policies, identifying failure modes, and using what teams learn to guide the environments and capabilities we build next.

That next phase depends on testing the library where teams actually work, so we're inviting embodied AI teams with high-value training and evaluation use cases to explore the catalog and help shape what comes next.