AI Labs · Spatial intelligence

Training data for spatial intelligence.

VLGE turns how worlds are built, played, and explored into a scalable dataset: the human behavioral data layer for systems learning to operate in 3D.

world / —
samples / 0

What a session looks like.

Every session people build and play becomes a continuous event feed of pose, velocity, state changes and dwell, time-aligned at 30 Hz. Below are four real captures, replayed and interleaved as they're stored.

Two people, one space,
two completely different paths.

Most datasets capture what a space is. VLGE captures how it's used: one person wanders and lingers while the next walks straight through. The same world produces opposite behavior, and every number here is read straight from the two real captures below.
behavioral ground tracks ● 2 sessions · real capture
session Asession B straightness dwell clusters radius of gyration

What data is available

One session. Every data stream, aligned.

The console plays one real captured session end to end, with every panel reading the same playhead. As you scroll the streams, it re-reads that session through each one.

01 / 05

3D Environments

Production-grade interactive 3D scenes, from interiors and streets to towers and stores, authored to be walked and played inside online.

watertight geometry · materials · navmesh · floor plans
02 / 05

3DGS

Gaussian-splat captures of the same worlds: photoreal radiance fields for training perception and novel-view synthesis at the fidelity the physical world actually has.

gaussian splats · radiance fields · multi-view
03 / 05

Spatial Metadata

The structured layer beneath the geometry: labelled zones, object semantics, adjacencies, sightlines and coordinates, the grammar a model needs to reason about a space. You can read it in the fingerprint.

coordinates · semantics · zones · sightlines · fingerprint
04 / 05

Continuous Event Streams

Time-aligned behavior at 30 Hz: pose, gaze, velocity, dwell, build and interaction events. This is the record that turns a static scene into how it's actually used, and it's the JSON streaming on the right.

pose · gaze · dwell · interactions · build-log · JSON
05 / 05

Custom Data Collection & Delivery

Need a specific environment, task, population or modality? We build the world, run the collection, and deliver the dataset to your spec, ready to scrub, sample and export.

bespoke worlds · task design · audio · scheduled delivery

Example applications

Built for the systems learning to operate in 3D.

01
build logs · geometry

World Models

Ground generative world models in how real spaces are built and changed.

02
human trajectories

Simulation

Populate simulators with real human trajectories, not scripted agents.

03
30 Hz pose · navmesh

Robotics

Navigation & manipulation priors from human behavior in 3D.

04
9 behavioral channels

Embodied AI

Agents that learn to perceive, plan and act in physical space.

05
benchmarks · fingerprints

Spatial Intelligence

Benchmarks & datasets for the research defining the field.

A growing library of spatial datasets for today's research, flexible for tomorrow's.

Request data access Talk to the data team →