Niantic is using Pokémon Go player scans to train spatial AI models

Back in 2016, when millions of people walked around neighborhoods pointing their phones at sidewalks to catch a virtual Charizard, nobody imagined they were training a future robotics foundation model.

Niantic has announced that it is building a Large Geospatial Model, or LGM. The model combines millions of real-world scans taken directly from the smartphones of players playing Pokémon Go, Ingress, and Scaniverse.

According to Niantic, the ultimate goal is to let computers and robots understand and interact with the physical world in ways never seen before, basically helping machines navigate physical terrain the way ChatGPT spits out text.

Mapping what street view cars cannot reach

The spatial intelligence powering this new model builds directly on Niantic's Visual Positioning System, an effort the company spent the past five years developing.

Niantic VPS works by using a single smartphone snapshot to calculate precise location and orientation against a crowdsourced 3D map.

Unlike traditional mapping data gathered by car-mounted cameras, this dataset is taken entirely from a pedestrian perspective. It covers alleys, footpaths, plazas, and hidden spots completely inaccessible to vehicles.

Back in 2022, Niantic Chief Scientist Victor Prisacariu explained that the scans do not just map 3D geometry. They also provide semantic understanding, helping the system identify what is the ground, sky, or trees.

As outlet 404 Media noted, not a single person downloading Pokémon Go back in the day could have guessed their casual gameplay would fuel an AI training pipeline down the road.

Why it matters

For gamers and mobile users, it is a glaring reminder of how free-to-play apps quietly monetize your labor. You thought you were scanning a neighborhood playground for in-game perks, but you were actually building a geospatial training set for spatial computing.

For the robotics and embodied AI sector, however, this pedestrian dataset is pure gold. While tech giants have mapped millions of miles of roadways, finding high-fidelity, foot-level 3D scans of global public walkways is nearly impossible without an army of millions of people.

We think Niantic's strategy is undeniably clever, but it walks a fine line on data ethics. Repurposing casual consumer gameplay into foundational spatial AI shows incredible foresight, yet it will likely spark fresh blowback around whether early players ever genuinely consented to this kind of downstream use.