For nearly a decade, millions of Pokémon GO players have wandered through parks and city streets, unknowingly participating in the world’s largest crowdsourced mapping project. On March 16, 2026, Niantic’s AI spin-off, Niantic Spatial, revealed that this decade of “unpaid labor” has culminated in a Large Geospatial Model (LGM) trained on over 30 billion images, now serving as the primary brain for autonomous delivery robots.
The “Pikachu” to “Pizza” Connection
In a partnership with Coco Robotics, Niantic is deploying its Visual Positioning System (VPS) to help small, sidewalk-roaming robots navigate dense urban environments.
“It turns out that getting Pikachu to realistically run around a fountain and getting Coco’s robot to safely navigate a sidewalk is actually the same problem,” said John Hanke, CEO of Niantic Spatial.
Why GPS Isn’t Enough
Standard GPS can be off by several meters, especially in “urban canyons” where skyscrapers block satellite signals. For a delivery robot, being off by two meters means the difference between the sidewalk and the street.
- Centimeter Precision: Niantic’s VPS uses the robot’s cameras to “look” at landmarks—statues, storefronts, and unique architecture—and compares them to the 30-billion image database.
- Accuracy: The system can pinpoint a robot’s location within centimeters, allowing it to stop precisely at a customer’s door or navigate complex intersections.
- Environmental Versatility: Because players scanned the same spots at different times, the AI understands how a location looks during rain, snow, night, or golden hour.
How the Data Was Gathered
The dataset is the result of years of clever in-game incentives that turned players into mobile mapping units:
- PokéStops & Gyms: Players naturally pointed their cameras at landmarks to interact with the game.
- “Field Research” (2020): Niantic introduced explicit tasks asking players to “scan” local monuments in exchange for rare items like Poffins or Rare Candies.
- Diversity of Perspective: Unlike Google Street View cars that stay on roads, Pokémon GO players walked down alleyways, through parks, and up pedestrian stairs, providing “ground-level” data cars can’t reach.
Privacy and “The Product” Debate
The revelation has sparked a firestorm on social media, with the hashtag #TheProduct trending as users realize their hobby has been monetized for industrial robotics.
| Player Action | Data Collected | Industrial Utility |
| Catching a Pokémon | Camera angle & depth data | Training robot “depth” perception |
| Scanning a PokéStop | 360-degree visual video | Creating 3D “Digital Twins” of cities |
| Walking to a Gym | Pedestrian-only pathing | Mapping robot-friendly sidewalks |
While Niantic maintains that scanning features were voluntary and opt-in, critics argue that the “terms and conditions” did not clearly state that the data would be sold to third-party robotics firms.
What’s Next: The “Living Map”
Niantic Spatial isn’t stopping at delivery robots. The company aims to build a “Living Map”—a real-time simulation of the world that updates as robots move through it. As Coco’s 1,000+ robots roam cities like Los Angeles, Helsinki, and Miami, they will feed fresh images back into the system, ensuring the map stays current even as storefronts change or construction occurs.
