Pokemon Go player data is being used to train AI
Niantic, the maker of popular augmented reality (AR) games such as Pokemon Go and Ingress, is using player data to build an AI model that can navigate the physical world. The company announced the effort in a recent blog post, introducing the idea of a "Large Geospatial Model" or LGM. It is aimed at helping computers perceive and understand physical spaces in new ways.
A new approach to understanding physical spaces
The LGM is trained on millions of geolocated images from across the world, enabling it to predict its immediate surroundings. This is similar to how Large Language Models (LLMs) such as OpenAI's ChatGPT, generate coherent sentences by statistically determining the likelihood of one word following another. Niantic believes that "Large Geospatial Models will help computers perceive, comprehend, and navigate the physical world in a way that will seem equally advanced."
LGM's potential applications in AR and beyond
Niantic imagines the LGM as a key component for AR glasses and other domains such as robotics, content creation, as well as autonomous systems. The company has proposed that as we move from phones to wearables connected to the real world, spatial intelligence will be the world's future operating system. This vision highlights Niantic's dedication to pushing AI capabilities for real-world use cases.
LGM enhances Niantic's visual positioning system
The LGM builds on Niantic's Lightship Visual Positioning System (VPS), allowing players to pin virtual items to particular locations in the real world with incredible precision. A case of this tech in action is the experimental Pokemon Playgrounds feature in Pokemon Go. In this, players can place Pokemon at a particular spot that will remain there for others to view and engage with.
Niantic's unique data collection approach
Niantic's method of data collection is different as it collects information from a pedestrian point of view, from places that cannot be reached by vehicles. The company already has 10 million scanned locations across the globe, over a million of which have been activated and made available for use with its VPS service. This data is collected via Niantic's games and Scaniverse, an app created by the company for 3D scanning objects and locations.
