Your Pokemon Go data is being used to train AI delivery robots

Niantic Spatial announced last week a new partnership with CoCo Robotics, a firm focused on small-scale self-driving delivery vehicles (think little robot boxes on wheels). How those robots go about seeing and understanding the world around them stands to be improved, due to them having access to a decade of user data, gleaned via sources such as Pokemon Go.

Niantic Spatial is an independent tech firm spun-off from Scopely’s Niantic, and the company is home to many who worked on both Pokemon GO and Niantic’s augmented reality platform. The two companies share data.

Although the news from Niantic Spatial (which you can see here) doesn’t mention Pokemon Go by name, the company’s Visual Positioning System (VPS) relies on data from the long-running popular AR game among other sources. The Niantic Spatial site notes their VPS is built using “a large-scale 3D map created from real-world scans” adding that “these scans come from a variety of sources, including people capturing interesting locations through Niantic’s games“.

Those scans will include imagery captured via the game’s optional AR Mapping feature — where players are rewarded for scanning real-world PokéStop locations with their camera. The game’s broader location data (Pokemon GO users have logged over 30 billion miles of exploration since launching in 2016) isn’t understood to directly feed the VPS, though Niantic does collect and retain it..

Either way, it’s a huge crowdsourced data set being used to train an AI vision model — I imagine not many of the still ~20 million active weekly users of the Pokemon Go app realise some of their anonymised data is being used in this way.

Will Douglas Heaven, writing for MIT Technology Review, has more on this announcement.