A new AI system was trained on 30 billion Pokémon GO images

Niantic, the developer behind Pokémon GO, has revealed it used nearly a decade of player data to train an AI navigation system.

A new AI system was trained on 30 billion Pokémon GO images

Niantic, the developer behind Pokémon GO, has revealed it used nearly a decade of player data to train an AI navigation system capable of pinpointing locations within centimetres.

Last week, the company announced the technology will be used to help delivery robots navigate city streets.

The announcement has raised questions about the ethics of how the data was collected, including the nature of player consent.

Here’s what you need to know.

Pokémon GO

Pokémon GO is an ‘augmented reality’ smartphone game developed by Niantic in partnership with Nintendo.

Augmented reality games use a device’s camera and screen to layer graphics over the real world.

The game asks users to grant access to their precise location so they can find Pokémon nearby.

Real world locations such as statues and churches are PokéStops, where users can access in-game items.

In the first 60 days after its 2016 launch, 500 million people downloaded Pokémon GO.

The game had 100 million active users across 2024.

In 2025, Niantic sold off the video-game arm of its company, but retained the mapping data and technology to use for its AI spinoff Niantic Spatial.

According to Bond University Associate Professor Dr. James Birt, “Pokémon GO was really just a gamification layer for allowing Niantic... to essentially gather data of real world locations”.

Every time a user scanned a landmark for in-game rewards, their phone was capturing photos, GPS coordinates, camera angles, and movement data.

You have read 0 articles this year.

Your contribution ensures The Daily Aus can continue doing the work you love.

This resulted in 30 billion images collected across thousands of cities in every weather condition, time of day, and lighting situation – all feeding into a growing visual map of the world.

Application

Niantic Spatial’s first commercial application is a partnership with a delivery robot startup called Coco Robotics.

Niantic’s tech solves a core problem with GPS in cities, where tall buildings can interfere with satellite location data.

The delivery robots will be able to identify locations by matching camera footage against the database of real-world images, much of which was generated by Pokémon GO players.

Ethics

All Pokémon GO users approved the collection and third party use of their data by agreeing to the game’s terms of service.

Niantic says it has used “player-contributed scans” to build a model of the world from its database of images.

“This scanning feature is completely optional – people have to visit a specific publicly-accessible location,” it says.

However, Dr. Birt questioned whether players actually knew how their data would be used.

Other apps

The practice of using user data for third party applications is common in the tech industry.

For example, Tesla collects data from drivers in its cars to train its autonomous systems.

“All of these systems that were given away for ‘free’,” Dr. Birt says, “really were a data aggregate source.”

Get Australia's free morning news brief.

Trusted by 400,000 Australians. Free, every weekday.

Already subscribed? Just enter your email above. Privacy Policy.