The Sneaky Way Pokémon Go Became a Robot Training Program

Niantic Spatial, the company behind Pokémon Go, has partnered with delivery robot maker Coco Robotics to steer sidewalk robots using a Visual Positioning System trained on more than 30 billion images captured by Pokémon Go players.

For years, players unknowingly built this crowdsourced map, especially through a 2020 Field Research feature that rewarded them for scanning real world landmarks and statues, along with data gathered at the game's battle arenas.

The system pinpoints a location within centimeters using visual landmarks rather than GPS, which often fails on city streets hemmed in by tall buildings.

Coco's robots rely on four mounted cameras to read those surroundings and navigate dense urban environments where autonomous delivery machines have long struggled to cross streets.

Niantic chief executive John Hanke framed the crossover neatly, noting that getting Pikachu to run around realistically and getting a robot to move safely through the world turn out to be the same problem.

The deal shows how data collected for entertainment can be repurposed years later for commercial use, echoing earlier examples like Google's CAPTCHA systems and law enforcement's use of Waze.

The pizza robots rolling through Los Angeles and Helsinki learned their streets from game players who never thought of it as mapping. Niantic Spatial trained its Visual Positioning System on some 30 billion urban images and now guides roughly 1,000 Coco Robotics sidewalk couriers through city canyons where GPS can drift 50 meters MIT Technology Review. The consent story is less sneaky than the headline: scans were opt in behind a level 20 gate with third party sharing disclosed, yet few players read a notice delivered as a game chore a Euronews fact check. Watch whether European regulators accept that disclosure as enough.