Niantic is building a ‘geospatial’ AI model based on Pokémon Go player data

Niantic has announced that it’s gathering a caller “Large Geospatial Model” (LGM) that combines millions of scans taken from the smartphones of players of Pokémon Go and different Niantic products. This AI exemplary could let computers and robots to recognize and interact with the satellite successful caller ways, the institution said successful a blog station spotted by Garbage Day.

The LGM’s “spatial intelligence” is built connected the neural networks developed arsenic portion of Niantic’s Visual Positioning System. The blog station explains that “Over the past 5 years, Niantic has focused connected gathering our Visual Positioning System (VPS), which uses a azygous representation from a telephone to find its presumption and predisposition utilizing a 3D representation built from radical scanning absorbing locations successful our games and Scaniverse,” and “This information is unsocial due to the fact that it is taken from a pedestrian position and includes places inaccessible to cars.”

Niantic Chief Scientist Victor Prisacariu was much explicit successful a 2022 Q&A, saying, “Using the information our users upload erstwhile playing games similar Ingress and Pokémon Go, we built high-fidelity 3D maps of the world, which see some 3D geometry (or the signifier of things) and semantic knowing (what worldly successful the representation is, specified arsenic the ground, sky, trees, etc).”

As 404 Media points out, cipher who downloaded Pokémon Go successful 2016 could person predicted their information would “one time substance this benignant of AI product.”