Niantic Uses Pokemon Go Player Data To Train Raya Geospatial Models



Niantic announced they are creating a Large Geospatial Model (LGM) using Pokemon Go and Ingress player data. Scans of surrounding images taken by millions of players using smartphones will train an artificial intelligence (AI) that understands visuals and locations in a 3D map.



LGM is trained using 50 neural networks, 150 trillion parameters, and scans at 10 million locations worldwide. It forms the basis of the Visual Positioning System (VPS) system developed by Niantic over the past five years with it now using accuracy on the centimeter scale


VPS can know the location by just using a piece of 2D image. All player data collected over the years contributes to a complete 3D mapping system as it also captures areas inaccessible to vehicles.


LGM will allow computers to see and understand physical space, but also to interact through AR glasses, robotic systems and autonomous systems. With this an operating system with spatial intelligence can be used in the near future.

Previous Post Next Post

Contact Form