Pokémon Go AR data now trains AI to guide delivery robots through city streets with precision

Pokémon Go's AR data has been turned into centimeter-accurate navigation for delivery robots

Serving tech enthusiasts for over 25 years. TechSpot means tech analysis and advice you can trust. Connecting the dots: Pokémon Go's global AR craze is now steering something far more prosaic than virtual Pikachu: real delivery robots trying to find the right doorway on a crowded city block. The same location data and street-level imagery that once anchored monsters to sidewalks and plazas have been repurposed by Niantic. Coco Robotics is now using that technology to guide its sidewalk bots through dense urban areas where GPS alone is too unreliable to keep them on course. Niantic Spatial, an AI spinout formed in 2025, has turned years of mobile gaming data into what it describes as a high-precision world model of the physical environment. The company is now commercializing that work through a visual positioning system that can locate devices to within a few centimeters using only camera input and map context. Its first large-scale deployment is with Coco Robotics, a last-mile delivery startup operating roughly a thousand sidewalk robots across US and European cities, where satellite signals are often too noisy to support reliable autonomy. The technical problem Niantic Spatial is tackling is straightforward to describe but difficult to solve. GPS degrades badly in dense cities, with position estimates drifting by tens of meters as signals bounce off glass and concrete. That level of error can place a delivery bot on the wrong block or even the wrong side of the street. Coco's robots, which travel at about five miles per hour and carry loads ranging from multiple extra-large pizzas to several grocery bags, must hit promised arrival times and precise pickup and drop-off points if they are to match or exceed human couriers. Niantic Spatial's alternative is a visual positioning system (VPS) that localizes a device based on what it sees rather than relying on radio signals alone. Over the past several years, the company has aggregated data from Pokémon Go and its earlier augmented-reality title, Ingress. Both games encouraged players to visit specific real-world locations such as gyms, battle arenas, and other points of interest. Those gameplay loops produced a dense global dataset of images captured in urban settings, each paired with rich metadata from the phone, including latitude and longitude, camera orientation, device pose, motion data, and other sensor readings. Niantic Spatial says it trained its models on roughly 30 billion images, heavily clustered around more than a million "hot spot" locations photographed from many angles, at different times of day, and under varied weather conditions. Because each frame is tied to a centimeter-scale pose estimate, the training set effectively functions as a multi-view 3D sampling of city streets, crosswalks, storefronts, and building facades. The company then trains its model to infer an exact location and orientation from a handful of current images, even in areas that are less thoroughly covered than those original hot spots. Also read: The rise of delivery robots is sparking vandalism, protests, and debate For Coco, that means its robots can fuse GPS with camera-based localization from Niantic Spatial's model. Each unit carries four hip-height cameras that look in all directions, a perspective different from a person holding up a phone but one that Coco says was straightforward to adapt to the existing data. Coco's robots have already logged hundreds of thousands of deliveries and more than a million miles across Los Angeles, Chicago, Miami, Jersey City, and Helsinki, giving the company a baseline against which to measure improvements in reliability from the new system. Visual positioning itself is not new, but it has historically been constrained by the availability and coverage of high-quality imagery. Niantic Spatial's bet is that the sheer volume and diversity of its crowdsourced data gives it an advantage over rivals that build maps primarily using their own sensor fleets. Other delivery-robot vendors, such as Starship Technologies, use their sensors to build local 3D maps of edges, poles, and building outlines as they move through an area, then rely on those maps for subsequent runs. By contrast, Niantic Spatial aims to maintain a global, shared geospatial model and expose it through an API to any robot, phone, or headset that needs to know exactly where it is. The company calls that model a "living map": a virtual representation of the world that is constantly updated as machines move through it. As Coco's robots and other future partners traverse sidewalks and streets, their sensors can contribute fresh observations that refine and extend Niantic Spatial's underlying maps. The aim is not just geometric accuracy but also semantic understanding, with objects tagged and described in ways that make sense to machines. Niantic's leaders describe this effort as a continuation of long-running work in digital mapping rather than a departure from it. As mapping has evolved from 2D to 3D and into dynamic "digital twin" simulations, the core link between map coordinates and physical locations has remained. What is changing is the primary consumer of those maps. Increasingly, it is machines rather than humans. In that view, the same spatial intelligence that once kept virtual Pikachu aligned with the sidewalk is now being repurposed to keep a 100-pound delivery robot on course through traffic and weather.

Turns out all your Pokémon Go data will be used to train robots

When Pokémon Go debuted in 2016, it became an overnight sensation. From London to New York, it felt as though everyone had installed Niantic's augmented reality Pokémon mobile app and had taken to the streets in a frenzied attempt to catch them all. While Niantic no longer owns Pokémon Go, which Scopely acquired in March 2025, the data collected by Niantic during those years is now being used to train robots. This news comes from Niantic Spatial's announcement of a new partnership with Coco Robotics, which has developed an urban robot designed to deliver food through complex urban landscapes. With this collaboration, Coco Robotics will be leaning on Niantic's expertise in "spatial AI and its Visual Positioning System (VPS)" to further improve Coco Robotics' titular delivery robot, Coco: a fleet of around 1,000 flight-case-size robots built to carry up to eight extra-large pizzas or four grocery bags, deployed in Los Angeles, Chicago, Jersey City, Miami, and Helsinki. One of the biggest challenges Coco faces is that the GPS signal can be weak in cities where radio waves bounce off big buildings. "The promise of last-mile robotics is immense, but the reality of navigating chaotic city streets is one of the hardest engineering challenges," said John Hanke, CEO of Niantic Spatial, an AI offshoot that Niantic founded in May 2025, via the company's blog. "We are thrilled to be working with Coco Robotics as our first robotics partner and deploying spatial intelligence to help solve these challenges head-on." Hanke also added that "It turns out that getting Pikachu to realistically run around and getting Coco's robot to safely and accurately move through the world is actually the same problem." All the data collected by people playing Pokémon Go and its previous augmented reality game, Ingress, is now being used to build an accurate model of the cities that Coco has to navigate. With Niantic's VPS system, Pokémon Go can determine a player's location in the world from their surroundings rather than relying on a player's GPS location. By having players use their phones at different angles, Niantic was able to scan real-world locations and landmarks, with players gathering the data they needed to ensure better accuracy across a range of conditions, such as height, angle, and weather. In 2020, Pokémon Go added a feature called "Field Research," which rewarded players for taking photos and scans of their surroundings in exchange for items and rare Pokémon. Whether they understood the implications or not (Niantic has always been open about building datasets), players helped feed Niantic data that was later used to train fast-food delivery robots. While Niantic never hid that it was collecting data, it's not hard to imagine that some Pokémon fans won't be too happy about that data being used by an AI company to train robots. Hopefully, the Coco fleet will at least have a better sense of direction than Leon from the Pokémon franchise.

With over 30 billion images logged, Pokemon Go is using players' activity to train its spinout AI company to help GPS pinpointing: "I'm very focused on trying to re-create the real world"

Pokemon Go players are helping delivery robots reach consumers' doors, as Niantic Spatial says it's using images captured on the app to improve location pinpointing in partnership with companies like Coco Robotics. Although Pokemon Go makes it clear in-game that it's collecting AR mapping data, with a notification prompting players to acknowledge that they'll help add to this pool for any given location, AI company Niantic Spatial - a spinout from developer and publisher Niantic - is apparently tapping into this collection now to... help delivery robots? Speaking with MIT Technology Review, CTO Brian McClendon reveals as much. Niantic Spatial's latest model is one that can reportedly pinpoint somebody's location on a map to the dot - we're talking a few centimeters - using images of landmarks, like buildings, in the vicinity. The company wants to use this technology to help robots, like those deployed by Coco Robotics, navigate and deliver across Europe and the United States with greater precision, where GPS might be less reliable. "Everybody thought that AR was the future, that AR glasses were coming, and then robots became the audience," as McClendon states. "The urban canyon is the worst place in the world for GPS. If you look at that blue dot on your phone, you'll often see it drift 50 meters, which puts you on a different block going a different direction on the wrong side of the street." That's where Niantic Spatial's new tech would help. How does it work, though? Well, for the last few years, the AI company has been using the aforementioned data from Pokemon Go to build an accurate visual positioning system. As Niantic Spatial CEO John Hanke says, "It turns out that getting Pikachu to realistically run around and getting Coco's robot to safely and accurately move through the world is actually the same problem." Who would've thought? Niantic Spatial has trained this system on some 30 billion images snapped by players in urban environments - the sorts of places that McClendon calls "the worst" for GPS. "We had a million-plus locations around the world where we can locate you precisely," he explains. "We know where you're standing within several centimeters of accuracy and, most importantly, where you're looking." Sounds... ominous, but, hey, it works. The company has even bigger aspirations, too - if the map-making improvements keep on rolling, it'll capture everything. "We're not there yet, but we want to be there," McClendon concludes. "I'm very focused on trying to re-create the real world." That's one big task. It's certainly a unique way to go about things, too, with Charizard and Pikachu captures helping train super-advanced AI mapping tech. It's important to note that this news has sparked fears of Niantic using players who were "unknowingly," as one viral online post puts it, feeding AI tech with personal data, but this doesn't seem to be accurate at all. As mentioned earlier, Pokemon Go does make it clear that images are banked as mapping data, and that's a fact that has been public information for quite some time now. The delivery robot side of things, however, is new. Are you still playing yourself? Be sure to browse through our roundup of the most up-to-date Pokemon Go codes for a little treat in-game.