Developer Harnesses Million Weekly Scans for Real-World Mapping
Niantic, the developer behind Pokemon Go, revealed its development of a Large Geospatial Model (LGM) trained on player-submitted location data. The company collects one million new location scans weekly through its Visual Positioning System, building a dataset that enables AI to predict and map real-world locations. On November 22, 2024, Niantic clarified that all data collection remains strictly opt-in, focusing primarily on enhancing player experience.
5 Key Points
- Niantic’s Large Geospatial Model aims to auto-complete real-world location data using AI technology
- The system currently contains 10 million scanned locations globally
- Players submit one million new location scans weekly through Pokemon Go and other Niantic apps
- Visual Positioning System provides centimeter-level accuracy for AR content placement
- All scanning features require explicit player opt-in at public locations
How Niantic’s Geospatial AI Works
Niantic’s Large Geospatial Model represents a departure from traditional mapping technologies by applying artificial intelligence principles to physical world data. While language models like ChatGPT process text to generate human-like responses, Niantic’s system analyzes spatial patterns across thousands of real-world locations to predict characteristics of unmapped areas. The technology draws from a vast database of user-submitted scans encompassing 10 million locations worldwide, with one million actively integrated into the Visual Positioning System. In its technical blog post, Niantic illustrated this capability through a practical example: “Let us assume the closest local model has seen only the front entrance of that church, and thus, it will not be able to tell you where you are. The model has never seen the back of that building. But on a global scale, we have seen a lot of churches, thousands of them, all captured by their respective local models at other places worldwide.” This distributed knowledge approach enables the system to make educated predictions about unseen portions of buildings and landscapes based on common architectural and geographical patterns.
Pokemon Playgrounds Showcases Real-World Applications
The practical implementation of Niantic’s geospatial AI materialized in November 2024 through Pokemon Playgrounds, an experimental feature in Pokemon Go. This new system transforms how players interact with their environment by allowing them to place Pokemon at precise real-world coordinates, creating persistent digital content discoverable by other users. The Visual Positioning System achieves this through centimeter-level accuracy, significantly advancing over previous augmented reality implementations. Players can now anchor digital Pokemon to specific physical locations, where they remain accessible to other users even after the original player departs. This persistent world-building capability demonstrates the practical potential of Niantic’s Large Geospatial Model beyond gaming applications, suggesting possibilities for urban planning, navigation, and architectural documentation.
Data Collection and Privacy Considerations
The scale of Niantic’s data collection operation surpasses traditional mapping efforts, with the company processing hundreds of discrete images from one million new location scans every week. Unlike Google’s Street View, which relies on vehicle-mounted cameras restricted to roadways, Niantic accesses pedestrian areas through its network of Pokemon Go players. On November 22, 2024, Niantic addressed privacy concerns through an official statement to IGN: “This scanning feature is completely optional – people have to visit a specific publicly-accessible location and click to scan.” The company emphasized that casual gameplay does not contribute to AI training, requiring explicit user action to participate in data collection. The distinction between passive gameplay and active scanning participation maintains user privacy while building a comprehensive dataset of public spaces. This approach enables Niantic to document locations inaccessible to traditional mapping vehicles, including pedestrian walkways, parks, and building interiors, while respecting user privacy through opt-in participation.
FAQ
Q: What is Niantic’s Large Geospatial Model (LGM)?
A: The Large Geospatial Model is an AI system that predicts and maps real-world locations using data collected from Pokemon Go players. Similar to how language models process text, LGM analyzes physical world data to understand and predict characteristics of unmapped areas.
Q: How does Niantic collect location data from Pokemon Go players?
A: Niantic uses an opt-in Visual Positioning System where players actively choose to scan public locations using their phones. The company receives one million new scans weekly, each containing hundreds of discrete images. Regular gameplay without scanning does not contribute to the data collection.
Q: What is Pokemon Playgrounds, and how does it use geospatial AI?
A: Pokemon Playgrounds is an experimental feature in Pokemon Go that allows players to place Pokemon at specific real-world locations with centimeter-level accuracy. These digital Pokemon remain in place for other players to discover and interact with, demonstrating the practical application of Niantic’s Visual Positioning System.
Q: Is Niantic’s location scanning system mandatory for Pokemon Go players?
A: No, the scanning feature is completely optional. Players must actively visit specific public locations and choose to initiate a scan. Simply playing Pokemon Go does not automatically contribute to Niantic’s AI training data.
Q: How does Niantic’s mapping differ from Google Street View?
A: While Google Street View relies on car-mounted cameras limited to roadways, Niantic’s system accesses pedestrian areas through voluntary player scans. This allows the company to map locations cars cannot reach, including walkways, parks, and building surroundings.
Q: What are the potential future applications of Niantic’s geospatial AI?
A: According to Niantic’s blog post, the technology could support AR glasses, robotics, and content creation. The system’s ability to accurately predict and map physical spaces suggests applications beyond gaming in fields such as urban planning and navigation.
Citations
Valentine, Rebekah (November 22, 2024). Pokémon Go Players Have Been Training an AI to Auto-Complete the Real World. IGN. https://www.ign.com/articles/pokmon-go-players-have-been-training-an-ai-to-auto-complete-the-real-world
