What began as a global augmented-reality game is quietly becoming infrastructure for autonomous machines. Over the past decade, millions of players of Pokémon Go have taken photos of streets, buildings and public landmarks while hunting for virtual creatures. Today, that same visual data is helping delivery robots navigate cities with remarkable precision.
The technology comes from Niantic, the developer behind Pokémon Go. By analysing billions of AR images captured by players in Pokémon Go and its earlier game Ingress, the company has built a Visual Positioning System (VPS) — a form of geospatial AI capable of locating devices within a few centimetres. Unlike GPS, which can struggle in dense urban environments, VPS relies on visual cues from the real world. Buildings, signs, street corners and other landmarks are matched against a massive 3D map of the environment.
That system is now being used to guide autonomous delivery robots. In partnership with Coco Robotics, Niantic’s mapping infrastructure is helping small sidewalk robots deliver food and groceries in several US cities, including Los Angeles, Chicago, Jersey City and Miami.
The robots navigate using a simple but powerful principle. Cameras on the robot capture live images of the surroundings and compare them with Niantic’s visual map. When the system recognises landmarks — a shopfront, a traffic sign, the entrance to a building — it can determine the robot’s exact position and orientation. This allows the vehicle to stop precisely at a restaurant entrance or a customer’s door rather than somewhere nearby.
In effect, a global community of players has helped create one of the largest pedestrian-level maps ever assembled. Traditional autonomous navigation systems often depend on expensive, highly detailed mapping performed in advance by specialised vehicles. Niantic’s approach is different: the map is continuously built and updated through crowdsourced images from players and additional data collected by robots themselves.
To turn this technology into a business platform, Niantic has spun off its geospatial AI activities into a separate company called Niantic Spatial. The goal is to transform the massive dataset generated by its games into infrastructure for robotics, augmented reality and future spatial computing systems.
Coco Robotics is the first major commercial partner. The company already operates roughly a thousand small delivery robots across several US cities. Equipped with cameras, lidar sensors and Niantic’s VPS technology, these machines can move through complex urban environments and reach precise drop-off points.
The development also raises questions about data usage. Many of the images used to build the system were originally captured by players simply interacting with the game, often without realising that they were contributing to a large-scale dataset with potential applications far beyond entertainment. Critics argue that this highlights the growing importance of transparency in how digital platforms reuse user-generated data.
Niantic frames the story differently. In its narrative, a playful global game has inadvertently created the foundation for a new kind of infrastructure. The same activity that once helped players find Pikachu may now help a robot deliver dinner on time.
In that sense, Pokémon Go did more than bring augmented reality to the masses. It quietly mapped the world — and that map may soon guide the machines moving through it.
