Pokémon Go’s Mapping Legacy Reaches Delivery Robots—but New Scans Don’t

A March 5, 2026 partnership notice puts Niantic Spatial’s Visual Positioning System into a commercial delivery-robot program with Coco Robotics. The link to Pokémon Go is real but indirect: optional scans from the game helped develop Niantic’s spatial technology; ordinary play did not secretly train individual robots.

The current picture is also narrower than it was when the companies shared a games business. PC Gamer’s June 2026 account quotes Niantic Spatial as saying Pokémon Go AR scanning had been discontinued and game data was no longer shared after Pokémon Go moved to Scopely. Existing scans remain part of the technology’s history, but there is no continuing flow of new Pokémon Go scans into Niantic Spatial.

Players had to choose to scan a location

Catching Pokémon, walking between landmarks and spinning PokéStops were not model-training actions by themselves. The relevant data came from an optional feature that required a player to visit a publicly accessible site, start an AR scan and move around the subject while recording it from street level.

Niantic Spatial’s November 2024 technical account states that merely playing its games did not train the model and describes more than 10 million scanned locations, over one million locations activated for its Visual Positioning System and more than 50 million trained neural networks. Those are company-reported platform figures rather than results from an independent audit, and they cover scans from Niantic games and Scaniverse rather than Pokémon Go alone.

The scans were useful because they captured public places from multiple pedestrian viewpoints, sometimes at different times of day. That perspective can include paths, plazas and areas that vehicle-mounted mapping systems cannot see clearly. Location information attached to the images allows software to connect visual features with positions in the physical world.

What the scans became

The reusable asset is more than a collection of landmark photographs. Sets of overlapping images can be processed into spatial representations that preserve the geometry and recognizable visual features of a place. A localization system can then compare a new camera image with those representations to estimate where the camera is and which way it is facing.

Niantic’s work combined conventional three-dimensional mapping with neural representations trained for individual locations. In practical terms, the system compresses observations of a place into a form that can be queried later. Pokémon Go scans contributed to that broader technical lineage alongside other capture sources; they were not a complete, self-contained map built exclusively by the game’s players.

This distinction also separates localization from robot training. Contributors did not label curbs, demonstrate delivery routes or teach a vehicle how to react to pedestrians. Their scans helped create representations that can answer a more limited but essential question: where is the camera relative to a mapped environment?

VPS fills one layer of the robot’s navigation stack

A sidewalk robot needs more than the approximate position normally associated with consumer GPS. Near tall buildings, satellite signals can be reflected or obstructed, while pickup areas and accessible approaches may require finer positioning than a broad map coordinate provides. A Visual Positioning System, or VPS, uses current camera observations and an existing spatial model to refine location and orientation.

That function does not replace the rest of Coco’s autonomy system. Cameras and other sensors still have to detect obstacles; planning software must select a path; control systems must execute movements; and operational safeguards must handle uncertain situations. Niantic Spatial’s contribution belongs primarily to the localization infrastructure supporting those functions.

The partnership targets difficult urban settings, degraded GPS reception and precise approaches to pickup zones. Its public description establishes a named engineering relationship and a deployment of VPS to Coco’s fleet, so the technology is no longer presented solely as a research possibility. It does not provide an independent accuracy evaluation, a coverage map, a vehicle count using VPS or evidence that every Coco delivery currently relies on it.

That boundary matters when describing the impact. It is reasonable to say that technology developed partly from player scans has reached a commercial delivery-robot program. It is not yet possible to trace a particular food order to Pokémon Go-derived localization or to claim that the system controls every navigational decision made by a robot.

“Accidental robot trainers” is a useful shorthand—with limits

The phrase captures an unexpected change in purpose: data collected for location-based augmented reality later contributed to technology with robotics applications. Yet it blurs the difference between contributing raw observations and directly training a robot’s behavior. It can also imply that every player participated, although the relevant scanning feature required a deliberate opt-in action.

The surprise lies less in hidden gameplay mechanics than in the long life of spatial infrastructure. Niantic spent years developing ways to anchor digital content to real places, which required accurate maps and reliable camera localization. Those same capabilities can help a physical machine orient itself when satellite positioning is weak or insufficiently precise.

Players may not have anticipated that mapping technology derived partly from their submissions would move beyond games. At the same time, the available evidence does not support the stronger claim that millions of people unknowingly trained delivery policies or continuously supply Coco’s robots with data. The contribution was upstream: selected scans supported the mapping and localization technology on which a later robotics partnership now builds.

The corporate split turns a data pipeline into a legacy

Three facts now coexist. Optional Pokémon Go scans contributed to Niantic’s spatial systems; Niantic Spatial is deploying its VPS through a partnership with Coco Robotics; and Pokémon Go no longer sends new scanning data to Niantic Spatial following the game’s transfer to Scopely. The first fact explains the technological lineage, while the last prevents that history from being mistaken for ongoing collection.

Future growth of Niantic Spatial’s maps therefore cannot automatically be credited to current Pokémon Go players. The company can continue developing its models from retained technology and other capture sources, but the game and spatial business now have separate operators. Claims about fresh Pokémon Go scans training today’s delivery systems are outdated unless a new data-sharing arrangement is disclosed.

The accurate version of the ten-year legacy is narrower than the slogan and more consequential than a coincidence. Player-submitted AR scans helped Niantic build high-precision visual localization, and that technology has entered a delivery-robot deployment with Coco. The robots inherit part of Pokémon Go’s mapmaking history, but the people still playing the game are no longer extending that dataset for Niantic Spatial.

Also read:

• BUIDL Reaches $2.71B, but Its UniswapX Market Remains Whitelisted
• Metaverse Gaming Is Already Here—But You Don’t Need a VR Headset
• A WMS Can Deliver Eight Warehouse Gains—but It Cannot Fix Bad Data