Niantic Spatial has announced a partnership with Coco Robotics to deploy its Visual Positioning System (VPS) in last-mile delivery robots navigating city sidewalks. The VPS — trained on more than 30 billion images crowdsourced from Pokémon Go players over nearly a decade — enables centimeter-accurate localization by visually recognizing nearby buildings and landmarks, rather than relying on GPS signals that degrade in dense urban environments. Coco's robots, which ferry food and grocery orders over short distances, will use four mounted cameras alongside VPS to orient themselves in real time. The partnership was reported by Popular Science on March 13, 2026, drawing on an interview with Niantic Spatial CEO John Hanke in MIT Technology Review.
What makes the system work starts with an unusual origin story. Pokémon Go players were by design required to travel to specific locations and point their phones at various angles to interact with the game, generating rich, multi-perspective spatial data as a byproduct. A 2020 in-game feature called Field Research explicitly rewarded players with in-game items for scanning real-world statues and landmarks, accelerating the data collection effort. Because many different users captured the same locations under different lighting, weather, and viewing conditions, the resulting 3D models handle real-world variability better than data collected by a single dedicated mapping pass. Hanke summarized the technical overlap concisely: anchoring a virtual Pikachu to a real-world surface and guiding a physical robot through a sidewalk are, at the navigation layer, the same localization problem.
Consumer platforms quietly converting user behavior into commercial AI infrastructure is a pattern regulators are starting to examine directly. Niantic's terms of service permitted broad data use, but the scale of the repurposing — from a mobile game to a robotics navigation product serving enterprise clients — raises substantive questions about informed consent. Similar scrutiny has followed Google's CAPTCHA image-labeling system, long suspected of training computer vision models, and Waze user data reportedly accessed by law enforcement. The EU AI Act's provisions on data governance and the FTC's ongoing review of commercial data practices both provide concrete frameworks under which this kind of repurposing could face challenge.
Niantic Spatial frames the Coco partnership as an entry point into a larger ambition: a continuously updated "living map" of the world. Once deployed, VPS-equipped robots will feed new environmental data back into the model, creating a compounding data flywheel that improves localization accuracy over time — a strategy already central to how Waymo and Tesla operate. With Pokémon Go still maintaining an estimated 50 million active users in 2026, the crowdsourced data pipeline remains live. Niantic Spatial's next move is whether it can sell that infrastructure to robotics clients beyond Coco before well-funded <a href="/news/2026-03-14-ionrouter-yc-w26-launches-high-throughput-llm-inference-platform-powered-by-ionattention-engine">AI robotics platforms</a> build their own.