What started as a simple mobile game in 2016 is now helping machines navigate cities with precision. The millions of Pokémon Go players roaming cities and other places unknowingly created a massive database of images, which is today being used for training robots.
About ten years back, Niantic introduced Pokémon Go through augmented reality technology, which displayed digital Pokémon characters throughout actual locations. Players visited places using their phones to scan multiple locations, including gyms and monuments and public spaces, being unaware that the company was training robots via scanned images.
Reportedly, these Pokémon Go scans generated a dataset of more than 30 billion images. Each scan captured not only visuals but also metadata, including GPS coordinates, camera angles, and device movement. Aggregated across millions of players, this information forms a detailed 3D map of streets, buildings, and public areas.
Niantic’s AI spinoff, Niantic Spatial, is now utilising this data set to create a "Visual Positioning System". Unlike GPS, which is not always accurate in densely populated cities, this system works by matching the live feed of the cameras with the data set of the images.
This is now being utilised by Coco Robotics. Their robots, which roam the sidewalks of cities across the US and Europe, utilise multiple cameras to compare the environment with the data set of Niantic Spatial.
Pokémon Go dataset has now enabled the robots to go exactly to the doorstep, as opposed to just the general location, allowing them to improve the accuracy of the delivery of groceries as well as food orders, utilising the same data that was initially used to guide people to find their Pokémon in the real world.