Pokémon Go Players Helped Map the World—Now That Data Is Training Delivery Robots

The millions of players who spent years scanning landmarks while playing Pokémon Go helped build the mapping technology now guiding delivery robots through city streets, according to a recent report by MIT Technology Review.

The news comes as AI and robotics developers work to give robots a more accurate sense of their surroundings so they can move through cities without relying solely on GPS.

In February, Niantic Spatial partnered with Santa Monica, CA-based Coco Robotics to provide navigation technology for the company’s autonomous delivery machines.

Pokémon Go, released in 2016, sent players into real-world locations to catch digital creatures and interact with landmarks through their phone cameras.

Players could also submit optional scans of public landmarks such as statues or buildings to improve the system’s spatial mapping.

That mapping technology now powers San Francisco-based Niantic Spatial’s Visual Positioning System, which determines location by analyzing nearby physical landmarks. In May 2025, Niantic Spatial spun off from Niantic Inc to become its own company.

“Our initial VPS was built using scans that users choose to take in games—but no single source defines the model,” a Niantic Spatial spokesperson told Decrypt. “What makes our approach distinctive is the combination of scale and ground-level detail–and increasingly, the data our customers are generating is what drives accuracy in the environments that matter most to them.”

Founded in 2020, Coco Robotics operates small autonomous robots that deliver food and retail orders across city neighborhoods in Los Angeles, Chicago, Miami, and Helsinki.

Due to increased traffic, construction, and other hazards, robotics companies are increasingly exploring vision‑based positioning systems to complement GPS navigation.

With GPS, signals can bounce off buildings or disappear entirely in narrow streets, making precise navigation difficult for autonomous machines. VPS tackles this by comparing camera images with detailed visual maps of the environment and can provide much more reliable location data in GPS‑challenged conditions.

Critics, however, argue that the dataset behind Niantic’s spatial AI was built by players who may not have realized how their scans could be used.

“143 million people thought they were catching Pokémon,” one user wrote on X. “They were actually building one of the largest real-world visual datasets in AI history.”

“The killer move wasn’t the map, it was the incentive design,” another wrote. “Pokémon Go turned millions of players into unpaid edge-case hunters and made the data exhaust feel like play.”

Despite these concerns, Niantic reiterated that participation in scanning was voluntary.

“Players could choose to submit anonymized scans of public places to help improve VPS,” the spokesperson said. “This scanning was and remains entirely optional,” they said, adding that scans are not connected to player accounts.