Niantic, the developer of Pokémon Go, has been using the vast amounts of data its players generate to train an AI model that a variety of systems will use to navigate the physical world.
As reported by 404 Media, an outlet you should be reading every day in addition to us, Niantic took some inspiration from the Large Language Models that fuel the predictive text algorithms of AI chatbots like ChatGPT and is applying it to real-world navigation in ways that could one day make Google Maps obsolete.
Videos by VICE
The developer is creating an AI model of their own called the Large Geospatial Model and its eventual purpose would be to let computers better understand the physical world so, say, a delivery robot or an autonomous vehicle, for example, can better navigate a neighborhood.
Just like how a Large Language Model can predict the next word in a sentence based on context, the Large Geospatial Model will do the same with physical environments using geo-located images. Sure, Pokémon GO players around the world willingly submitted these images but likely didn’t know they were contributing to an AI model that may one day help a robot navigate their neighborhood.
All of this will work in tandem with another system Niantic developed called Lightship Visual Positioning System. If you’ve played Pokémon GO recently, you’re probably familiar with the Pokémon Playgrounds that lets you pin a Pokémon on a specific location on a map that correlates with a spot in the real world with, according to Niantic, “centimeter-level accuracy.”
So far Niantic says they scanned 10 million locations with another million or so added every week. And it’s not just streets like Google Maps but pedestrian-only areas, too.
For now, Niantic just wants to use the data to fuel its augmented reality gaming features, but who knows what this data will be used for in the future.
