Mapping real-world routes.
Everyone in tech seems to be into AI these days, developer Niantic doesn’t want to be left behind. With the help of images captured through the Pokémon Go app, the company is helping its spinout Niantic Spatial endeavour, which focuses on “geospatial AI” and digital maps of the physical world. Niantic Spatial’s long-term goal is to enable machines, robots, and AR glasses to understand, navigate, and interact with the physical world with centimeter-level precision.
From an interview with MIT Technology Review focusing on Niantic Spatial’s new partnership with Coco Robotics, a delivery robot start-up (via GamesRadar), we’ve now learned a little more about how exactly Niantic is using images collected through the Pokémon Go app to perfect and reshape digital maps of our surroundings. As always, the payment for free-to-play games and apps isn’t immediately obvious, though before you start complaining about being misled, Niantic’s term and conditions do state images “are banked as mapping data.” Likewise, the specific photo-scanning process is opt-in only in the game – users must voluntarily agree to upload their scans directly to Niantic Spatial.
MIT Technology Review’s article on Niantic’s AI spinout, which includes chatter with CTO Brian McClendon, is quite revealing, noting that the partnership started when “everybody thought that AR was the future.” While that didn’t pan out, clearly it doesn’t mean AR is useless.
Niantic Spatial has trained its model on “30 billion images captured in urban environments,” it’s claimed. Usually the images themselves, together with precise metadata, can analyse recognisable environments, landmarks, iconic buildings, and more to better pinpoint when and where exactly the mobile phones took them. “Maps are not only becoming more detailed; they are being used more and more by machines,” says McClendon.
Whereas other companies are fighting over LLMs and the (re)creation of information from scratch, Niantic is moving in a different direction. “I’m very focused on trying to re-create the real world,” concluded McClendon. It’s long been the aim of the company – though perhaps the breadth of uses, into robotics and beyond, may come as a suprise to the casual player.
Update: 19th March A previous version of this story implied that users may be inadvertently participating in the scanning and uploading of images. This is not correct: photo-scanning in the game is “entirely optional and clearly marked”, per Niantic, and the data used by Niantic Spatial is not from normal gameplay, but rather operates as an opt-in system; Niantic Spatial will only receive new Pokémon Go Scans when users voluntarily agree to upload the scans directly to Niantic Spatial, using a third-party service. It is not the case that the company trains the model via standard AR photos taken in-game. We apologise for the error and have updated the article and headline to reflect this.
Mentioned in this article
