‘Pokémon Go’ players have been unknowingly training delivery robots
Nearly a decade ago, Pokémon Go has turned the real world into a digital treasure hunt, with virtual creatures hidden in plain sight. The first augmented reality app for smartphones inspired hundreds of millions of players to wander parks, parking lots and even dimly lit alleys, peering through their phone cameras for Pikachus and Charizard that the app superimposed on their surroundings. It was a major success. But 10 years after the app’s heyday, it turns out that capturing digital creatures can now help that piping hot pizza you ordered find your way back to you.
This week, Niantic Spatial, part of the team behind Pokémon Goannounced a partnership with Coco Robotics, a company that makes short-distance delivery robots for food and groceries. Soon, these robot couriers will navigate sidewalks using Niantic’s Visual Positioning System (VPS), a navigation tool that would be able to pinpoint a location to within a few centimeters simply by looking at nearby buildings and landmarks. Niantic has trained this VPS model on over 30 billion images captured by Pokémon Go users, and claims this will help robots operate in areas where GPS is insufficient.
In other words, all that time users were spending wandering around playing Pokémon Go will now help you determine how well a courier robot can deliver your takeout. This is a striking example of how crowdsourced data, seemingly collected for one purpose, can be quietly repurposed years later for something entirely different.
“It turns out that having Pikachu run realistically and having Coco’s robot move safely and accurately through the world are actually the same problem,” Niantic Spatial CEO John Hanke said in a recent interview with MIT Technology Review.
How Niantic repackages Pokémon Go data
Instead of helping users navigate like GPS does, VPS determines where a person is based on their surroundings. This makes Pokémon Go particularly useful as a data source, as players had to physically move to specific locations and point their phones at different angles. This mapping effort received a significant boost in 2020, when the app added what it called “Field Research,” a feature inviting players to scan real-world statues and landmarks with their cameras in exchange for in-game rewards. Some of the data would also come from areas known as “Pokémon battle arenas.”
Whether players knew it or not, these scans created 3D models of the real world that would eventually feed into the Niantic model. More data means better accuracy, and because Niantic collected images of the same locations from many different users, it could capture the same locations in different weather conditions, lighting, angles, and heights. There is no shortage of raw materials either. At its peak in 2016, Pokémon Go had approximately 230 million monthly active players. Although less culturally relevant in 2026, the game still has around 50 million active users by some estimates.
Related Stories
How Pokémon Go data could help robots find their way
Niantic and Coco are betting that Pokémon Go the data will help delivery robots understand precisely where they are simply by looking at landmarks around them. Although most autonomous robots currently use some form of GPS for navigation, it is not always reliable. Other delivery robots tested on college campuses get lost or have trouble crossing streets. This confusion can cause delays. As any customer who has waited too long for a hot meal from a delivery app can attest, it’s crucial that those couriers arrive on time. After all, time is money.
“The promise of last-mile robotics is immense, but the reality of navigating chaotic city streets is one of the toughest engineering challenges,” Hanke said in a statement.
And while most people associate spotty GPS with national parks or isolated rural areas, reliability is also often compromised in the tall, densely populated buildings of a concrete jungle. All of these structures can interfere with signals, causing the location point on a map to drift. The idea is that Coco’s robots can use the VPS and four cameras mounted around the machine to get a much more accurate reading of their surroundings. In turn, the well-equipped robot will deliver the food on time.
It would also not be the first time that data freely retrieved by Internet users for a specific purpose ended up feeding something else entirely. Google’s CAPTCHA tests, which ask users to click on images of bicycles or traffic lights to verify that they are human, have come under scrutiny. Computer scientists have long speculated that CAPTCHA tests have been used to help train AI vision models. More recently, law enforcement has allegedly accessed or purchased user-generated content from the consumer mapping tool Waze to aid police investigations. And while Niantic hasn’t suggested providing its VPS data to authorities, it’s not hard to see how a tool that can accurately pinpoint a location based on landmarks in a photograph could seem appealing to law enforcement.
On a broader level, Niantic says its partnership with Coco Robotics is part of a longer-term effort to build a “living map” of the world that updates as new data becomes available. Once VPS-equipped delivery robots are on the streets, they will collect even more information that can be fed back into the model to further boost its accuracy. This kind of continuous collection of real-world data is already at the heart of how autonomous vehicle companies like Waymo and Tesla operate, and is a big part of why the technology has improved so dramatically in recent years.
So the next time you see someone in a park trying to “catch ’em all,” it’s quite possible that the data gleaned from that scavenger hunt will play a key role in determining whether the pizzas of the future will arrive at their destination on time.
The best new PopSci 2025 releases
