Niantic Utilizes Player Data from Pokemon Go to Train AI for Navigating Robots

A recent blog post from Niantic has caught the eye of many Pokémon GO enthusiasts and tech aficionados. The company detailed an interesting project in which they’ve been working on a machine learning-powered geospatial model, all crafted through the data gleaned from their apps. So, if you’ve been enjoying their games, there’s a good chance you’ve helped in building it!

The post, aptly titled “Building a Large Geospatial Model to Achieve Spatial Intelligence,” reveals how Niantic has been harnessing its Visual Positioning System to amass a wealth of data. According to them, they’ve been putting their geospatial model through extensive training, courtesy of the players’ contributions (big shout-out to 404 Media for highlighting this!).

Interestingly, Niantic shares, “As part of our Visual Positioning System (VPS), we’ve trained over 50 million neural networks with more than 150 trillion parameters, which helps us cover more than a million locations. In our dream of a Large Geospatial Model (LGM), these smaller networks merge into a larger global model. This shared understanding helps recognize locations that are yet unexplored.”

Now, you might wonder: What’s the big deal with this geospatial model? Put simply, it’s a tool that helps computers better understand and navigate real-world environments. As the blog explains, this translates into improvements for any tech involving AR glasses, robots, and autonomous systems. Imagine these devices being able to move around a city, identifying objects like lampposts without a hitch.

And then, there’s the question: What exactly is this Visual Positioning System? Pokémon GO fans might recognize it from the game’s AR mapping tool. This feature has been a part of field research tasks, where players scan real-world landmarks for in-game rewards. Niantic describes VPS as a technology that enhances how AR content maps onto the real world—ensuring, for instance, that a PokéStop is mapped to a museum and not the local pub’s restrooms.

Niantic proudly points out that over 10 million locations have been scanned by its users across their games. They’ve even shared a heat map demonstrating that much of this data comes from Japan, Coastal US, and Central Europe. AI and machine learning, despite their bad reputation for some misapplications, are proving their worth in AR. Just make sure you’re not unintentionally recreating a scene from The Shining with a real-world Metagross in your bathroom while playing Pokémon GO!

So, what’s your take on all of this? Personally, I’m a bit wary about AI tech, especially from an environmental standpoint, considering we haven’t quite tackled the energy crisis yet. But maybe you see things differently! Feel free to share your thoughts in the comments below.