Niantic is developing a new AI model called "Large Geospatial Model" (LGM), which uses massive scan data uploaded by players in games such as "Pokémon Go" to build high-precision 3D maps and give them New capabilities for computers and robots to understand and interact with the world. The core of LGM is based on Niantic's Visual Positioning System (VPS), which can determine position and direction through a single mobile phone photo. Its data source is unique and covers pedestrian perspectives that cannot be reached by vehicles, providing valuable data resources for the development of AI technology. This technology not only demonstrates the huge potential of player data, but also provides new directions for future AI applications.
Recently, Niantic (Nintendo) announced that it is developing a new artificial intelligence model called "Large Geospatial Model" (LGM).
The model was built using data from millions of scans uploaded via smartphones by players in Pokémon Go and other Niantic games. Niantic says this AI model will allow computers and robots to understand and interact with the world in entirely new ways.
LGM's "spatial intelligence" is based on the neural network used by the Visual Positioning System (VPS) developed by Niantic. According to a company blog post, the system is able to determine the location and orientation of a single image taken by a mobile phone, all based on a 3D map of interesting locations that the user scans. It is worth noting that these data are unique in that they are obtained from a pedestrian's perspective, covering places that vehicles cannot reach.
Niantic chief scientist Victor Prisacariu pointed out more clearly in a question and answer session in 2022 that the data uploaded by players when playing "Ingress" and "Pokémon Go" helped the company build a high-fidelity 3D map of the world. These maps include not only 3D geometry (i.e., the shape of objects), but also semantic understanding (e.g., what is on the map, such as ground, sky, trees, etc.).
Although players who downloaded "Pokémon Go" for the first time in 2016 may not have thought that the data they uploaded would support such AI products in the future, with the development of technology, the value of these data is gradually being discovered.
Official blog: https://nianticlabs.com/news/largegeospatialmodel?hl=en&_bhlid=4528b8205f253061720f30633407d8060dc88335
Highlight:
Niantic is developing a new AI model called "Large Scale Geospatial Model" that uses player data to build 3D maps.
LGM is based on a vision positioning system that is able to determine position and orientation from mobile phone images, covering pedestrian perspectives and hard-to-reach areas.
The construction of this model demonstrates the potential of player data to drive computers and robots to better understand and interact with the world.
All in all, Niantic's LGM model brings new possibilities for artificial intelligence in spatial understanding and interaction, and its success also relies on players' long-term contributions and data accumulation. This indicates that game data will play an increasingly important role in future AI development.