Niantic has announced plans to create an artificial intelligence (AI) model for navigating the physical world using data collected from its mobile games and apps. The company will utilize scans from Pokémon Go players and Scaniverse app users, providing a unique source of training data for the AI.
The Visual Positioning System (VPS), developed over five years, relies on a single image to determine position and orientation using a 3D map built from user-scanned locations in games and apps. Niantic’s new creation is called a “large geospatial model” (LGM), which processes physical spaces using geolocated images.
The scale of data collection reveals Niantic’s significant presence in the augmented reality space. The LGM draws from over 10 million scanned locations worldwide, with users capturing approximately 1 million new scans weekly through Pokémon Go and Scaniverse. These pedestrian-sourced scans capture areas inaccessible to cars and street-view cameras.
Niantic has trained over 50 million neural networks, each representing a specific location or viewing angle. These networks compress mapping images into digital representations of physical spaces, containing over 150 trillion parameters that help recognize and understand locations. The goal is to create a comprehensive model that can grasp any location from unfamiliar angles.
Source: https://arstechnica.com/ai/2024/11/niantic-uses-pokemon-go-player-data-to-build-ai-navigation-system/