Pokémon GO Players Unwittingly Trained AI for Delivery Robots

The popular mobile game Pokémon GO has faced scrutiny over the use of images captured within the app to train artificial intelligence systems developed by its creator, Niantic. While concerns have been raised about player consent, the data collection wasn’t entirely without user awareness.

Since its 2016 launch, Pokémon GO transformed cities like Brussels, Paris, and Rome into augmented reality playgrounds. The game remains popular, with over 100 million players as of 2024, according to Scopely, Niantic’s parent company.

Niantic Spatial and the Creation of 3D Maps

MIT Technology Review reported that Niantic Spatial, spun out of Niantic in 2025 after its games division was sold to Scopely, utilized gameplay images to train systems designed to build detailed 3D maps of real-world environments. Posts on X (formerly Twitter) suggested Niantic may be using visual data captured during players’ walks to develop navigation systems for delivery robots.

However, the use of this data wasn’t entirely covert. Niantic introduced dedicated AR mapping features in 2020, allowing players to scan locations and objects by physically moving around them with their phone cameras. This feature is only accessible after reaching level 20 in the game.

Niantic confirmed to Euronews’ fact-checking service, The Cube, that users must actively opt-in to submit anonymous scans and videos of public places to enhance its Niantic Spatial Visual Positioning System (VPS). Participation is voluntary, with users choosing to scan specific landmarks.

When a user points their phone camera at a statue in Brussels’ Parc du Cinquantenaire, a notification appears stating that the user is contributing to augmented reality mapping and that the data is shared with a third-party provider. The notification clarifies that the captured information is used to generate 3D models and support related technologies.

Niantic’s Terms of Service outline this arrangement in the “Rights Granted by You – AR Content” section, where users consent to Niantic’s use of images taken through the AR scan feature to improve its offerings.

30 Billion Images and the Visual Positioning System

Niantic Spatial has trained over 50 million neural nets using approximately 30 billion images, as disclosed by MIT Technology Review. This data has been used to create a Visual Positioning System (VPS) that provides “accurate, vision-based positioning and orientation globally, including areas where GPS signals are weak or unreliable.”

However, Niantic Spatial’s data inputs extend beyond Pokémon GO images, incorporating spatial inputs from robots, drones, and satellites, according to their website.

Partnership with Coco Robotics

In early March 2026, Niantic announced a collaboration with Coco Robotics, a provider of urban robot delivery services, to expand the deployment of its spatial AI and VPS technologies. Euronews reported on this partnership.

Coco Robotics operates autonomous robots delivering groceries, electronics, and food across cities including Los Angeles, Chicago, Jersey City, Miami, and Helsinki. The company has partnered with DashMart since 2018.

Coco Robotics has launched an upgraded series of delivery robots designed for urban environments. Previously reliant on GPS, which can be imprecise in dense urban areas, these robots will now integrate Niantic’s spatial mapping and VPS capabilities to improve navigation.

By utilizing refined 3D maps and vision-based positioning, the robots can navigate city streets with greater accuracy when delivering goods.