Niantic said it is developing a “Large Geospatial Model” (LGM) that combines millions of scans collected from the smartphones of Pokémon Go players and other company products. The model will help computers and robots understand and interact with the world in new ways.
Image source: nianticlabs.com
«LGM Spatial Intelligence is based on neural networks developed within Niantic’s Visual Positioning System (VPS). The company has devoted the last five years to working on VPS. It allows you to determine the location and orientation of the author of the photo based on one photo from your phone, comparing the data with images taken by other people. Today, Niantic’s database contains more than 10 million scanned locations around the world. “This data is unique because it comes from a pedestrian’s perspective and includes areas that are inaccessible to cars,” the company said in a blog post.
Niantic Chief Scientist Victor Prisacariu mentioned the work on the project back in 2022: “Using data uploaded by our users in games like Ingress and Pokémon Go, we created high-quality 3D world maps that include both 3D geometry ( or the shape of objects), and semantic understanding (what is on the map, for example, earth, sky, trees, etc.).” Although it’s unlikely that Pokémon Go players in 2016 realized that their data would be used to train AI.