# Niantic’s Expansive Geospatial Model: How Pokémon Go Users Are Influencing AI Navigation
In an innovative step, Niantic, the creator of the widely played mobile game *Pokémon Go*, has unveiled plans for a new AI framework aimed at navigating the physical realm. This framework, known as the “Large Geospatial Model” (LGM), is being developed with input from data gathered from *Pokémon Go* players and users of Niantic’s *Scaniverse* application. The company’s ambitious initiative aspires to transform how AI engages with real-world settings, utilizing millions of visual scans provided by its community.
## The Importance of Visual Scans in Developing the LGM
Niantic’s venture into the LGM commenced with its Visual Positioning System (VPS), a technology that employs images from smartphones to ascertain position and orientation in the real world. This system depends on a 3D map created from scans of actual locations, collected by participants of Niantic’s games such as *Pokémon Go* and users of *Scaniverse*. These scans are largely conducted from a walking perspective, offering an exclusive viewpoint of sites that are frequently unreachable to conventional mapping technologies like street-view cameras mounted on vehicles.
In a recent blog entry, Niantic disclosed that its LGM is founded on over 10 million scanned sites globally, with users adding around 1 million new scans weekly. These scans are refined by neural networks that condense thousands of images into digital representations of physical environments. The outcome is a highly intricate, AI-oriented map of the planet that can be utilized for navigation, augmented reality (AR), and beyond.
## A Different Type of AI Model
The LGM represents a unique kind of AI model akin to large language models (LLMs) like OpenAI’s ChatGPT. While LLMs are trained on extensive amounts of text data to comprehend and formulate human language, Niantic’s LGM processes visual information to interpret physical environments. The model is intended to identify and analyze locations from geolocated images, enabling it to navigate the surroundings with remarkable accuracy.
Niantic has trained more than 50 million neural networks, each symbolizing a specific location or viewpoint. These networks encompass over 150 trillion parameters—modifiable values that aid the AI in understanding and identifying various locations. Multiple networks can be amalgamated to map a single site, constructing a thorough model that can interpret even unfamiliar angles of a given area.
## First-Person Scans: An Exclusive Outlook
One of the most notable benefits of Niantic’s data gathering strategy is the first-person viewpoint offered by its users. Unlike conventional mapping systems, which depend on images taken by vehicles or drones, Niantic’s scans are captured by individuals traversing parks, streets, and other public venues. This pedestrian perspective provides a more personal and intricate view of the world, documenting spaces often overlooked by alternative mapping technologies.
Niantic’s LGM is crafted to utilize this distinctive perspective. For instance, if a user finds themselves behind a church, the model can draw on its understanding of thousands of other churches globally to recognize common architectural traits and establish the user’s position. This collective intelligence empowers the LGM to comprehend physical environments in ways that traditional GPS or mapping systems cannot achieve.
## Uses Beyond Gaming
Although Niantic’s LGM is being constructed with data from *Pokémon Go* and *Scaniverse*, its potential uses reach well beyond mobile gaming. The company envisions this technology being applied across numerous sectors, including augmented reality, robotics, and autonomous systems. The LGM may also be valuable for spatial planning, logistics, and remote collaboration, providing a fresh approach for AI to engage with the physical environment.
One illustration of this technology in practice is *Pokémon Go*’s Pokémon Playgrounds feature, enabling players to place virtual Pokémon at specific real-world coordinates for others to discover. This feature highlights the accuracy of Niantic’s VPS, which can position virtual items in the real world with centimeter-level precision.
## Ethical Issues and Player Feedback
The extent of Niantic’s data gathering prompts crucial inquiries regarding privacy and consent. Were the millions of *Pokémon Go* players who contributed scans to the LGM aware that their data would be utilized to train an AI framework? While the company’s data collection policies are probably outlined in its terms of service, many players may not have been fully cognizant of how their scans were being employed.
A recent piece by *404 Media* spotlighted this concern, igniting discussions on platforms like Reddit. Some players expressed astonishment, while others found it unsurprising, pointing out that Niantic’s business model has consistently been reliant on data gathering. One Reddit user remarked, “Definitely wasn’t unwittingly. Most of us knew their business model didn’t revolve around supporting the actual players.”
As the LGM progresses, it’s probable that these ethical issues will become an increasingly prominent aspect of the dialogue.
