Key Takeaways
- Niantic builds an artificial intelligence system using Pokémon Go data AI to create detailed geospatial models from player scans.
- Players capture images while exploring, which helps build 3D maps of real environments from a pedestrian perspective.
- The system uses Niantic’s Visual Positioning System to determine smartphone location and orientation from single images.
- This new AI model consolidates multiple neural networks into one, allowing it to recognize locations from partial images.
- Potential uses include support for augmented-reality devices, robotics navigation, and autonomous systems.
Niantic is building a new artificial intelligence system using Pokémon Go data AI collected from players around the world. The company announced that scans captured inside the mobile game are helping train a large geospatial AI model. This system focuses on understanding real-world locations through images and spatial information. Pokémon Go players contribute data while scanning landmarks and points of interest in the game. These scans allow Niantic to construct detailed three-dimensional maps of real environments. The dataset includes millions of images gathered from smartphones.
The project uses real-world visual data collected during gameplay. Players capture the images while walking through cities, parks, and public spaces. These scans show locations from a pedestrian perspective rather than from vehicles. Niantic uses the information to improve spatial mapping technology. The company calls the new system a “Large Geospatial Model.”
How Pokémon Go data AI builds detailed 3D maps
The Pokémon Go data AI system relies on Niantic’s Visual Positioning System. This technology can determine the location and orientation of a smartphone from a single image. The system compares the photo with previously scanned environments stored in a 3D map. The AI then calculates where the device is positioned within that space.
Niantic previously used many separate neural networks to recognize specific places. Each location required its own model. The new approach combines this information into one larger model. The system can learn patterns that apply across many environments. This allows the AI to recognize buildings, monuments, and other structures even when it sees only partial images.
Millions of scans collected from Pokémon Go and other Niantic applications support the training process. The company reports that the mapping system already covers more than a million real-world locations.
Potential uses of Pokémon Go data AI technology
Niantic says the Pokémon Go data AI model may support several technology sectors. The system could assist augmented-reality devices that need to understand physical surroundings. Robotics companies may also use geospatial AI for navigation. Autonomous systems could rely on the technology to recognize objects and landmarks in real environments.
The dataset used to train the AI is aggregated and anonymized. Personal identifiers such as player names or email addresses are not included in the training data. Pokémon Go launched in 2016 and became one of the largest location-based mobile games. Over time, the scanning feature inside the game produced a large collection of spatial images. Niantic is now using that dataset to advance AI systems designed to understand the physical world.
Source: https://www.polygon.com/pokemon-go-data-ai-robots-niantic/