Niantic’s Large Geospatial Model: Revolutionizing AI with Player Data from Pokémon GO
Niantic has announced a new project: a Large Geospatial Model (LGM). This model will be built using data from PokéStop scans conducted by players. The LGM aims to help computers understand and interact with the physical world more effectively.
### What is a Large Geospatial Model?
A Large Geospatial Model is an AI model that will process large amounts of location data. This includes billions of images and hours of scanned locations. All data points will link to physical locations. The LGM will enhance the understanding of different environments by creating a comprehensive 3D view.
### How It Works
The LGM will use information from local models. Some local models may show only part of a structure, such as a church’s front, but the LGM can combine data from various models to create complete images. For example, even if it lacks a view of a church’s back, it can infer its shape based on many other churches across the globe.
### Features of the Model
The model will think like a human. It will recognize streets and architectural styles and make navigational decisions, even in places it hasn’t been directly scanned. For instance, it could navigate an old European town using its understanding of typical layouts and designs.
How does player-contributed data impact the accuracy of the LGM in real-world scenarios?
Interview with Dr. Emily Carter, Geospatial AI Specialist
November 21, 2024
NewsDirectory3: Thank you for joining us today, Dr. Carter. Niantic has recently announced its ambitious project, the Large Geospatial Model (LGM), which aims to utilize player-contributed data from Pokémon GO to enhance spatial intelligence in AI systems. To start, could you explain what exactly a Large Geospatial Model is?
Dr. Emily Carter: Absolutely. A Large Geospatial Model (LGM) is an advanced AI framework designed to process vast amounts of geolocation data. This includes significant quantities of images and scanned locations that together create a detailed, three-dimensional representation of various environments. It essentially enables computers to comprehend and interact with our physical world in a much more sophisticated way.
NewsDirectory3: Interesting! How does the LGM work in practice?
Dr. Carter: The LGM is built on integrating data from multiple local models. For instance, if a local model captures only one side of a structure, like the facade of a church, the LGM can amalgamate insights from various similar models around the world. This allows it to infer the structure’s complete shape, even if it has not directly scanned all angles. It effectively stitches together fragmented views into a cohesive understanding.
NewsDirectory3: That’s quite remarkable. What features set this model apart from traditional models?
Dr. Carter: One of the most distinctive aspects of the LGM is its potential to emulate human-like thinking. It recognizes different streets, architectural styles, and can make navigational decisions based on prior knowledge, even in areas where data may be sparse. For example, it could expertly navigate the narrow, winding streets of an old European town by leveraging typical urban layout patterns.
NewsDirectory3: What is the current status of the project? How far along is Niantic in developing the LGM?
Dr. Carter: Niantic is making substantial progress but still seeks extensive data. Unlike text-based language models that can learn from internet sources, the LGM relies specifically on input from players using the “Scan a PokéStop” feature. They’ve gathered data from over 10 million scans, with about 1 million already processed for their Visual Positioning System (VPS). However, to reach its full potential, continuous player engagement and data contribution are critical.
NewsDirectory3: Looking to the future, what are Niantic’s aspirations for the LGM?
Dr. Carter: Niantic’s vision is to enhance the augmented reality (AR) experience in games like Pokémon GO and beyond. The ultimate goal is to enable users to accurately place digital objects within real-world environments, creating a persistent spatial interaction. This means that content placed in a location will remain there for others to explore, fostering a rich blend of virtual and physical worlds.
NewsDirectory3: Thank you for sharing your insights, Dr. Carter. It’s fascinating to see how Niantic is pushing the boundaries of gaming and AI technology with the LGM.
Dr. Carter: Thank you for having me. The intersection of augmented reality and spatial intelligence is an exciting frontier, and Niantic’s work has the potential to revolutionize how we engage with our environments.
### Current Progress
Niantic has made significant progress but still requires more data. Unlike language models that can learn from the internet, the LGM needs data from players using the Scan a PokéStop feature. So far, Niantic has collected information from over 10 million scanned locations, with 1 million processed for use in its Visual Positioning System (VPS).
### Future Aspirations
Niantic aims to use this data to improve augmented reality experiences in games like Pokémon GO. Their goal includes allowing users to place digital content in a real-world environment accurately. This content will remain in place for others to interact with later.
Niantic’s Large Geospatial Model represents a significant step in merging virtual and physical spaces, building a deeper understanding of the environments we explore.