Your PokéWalks Helped Robots Learn the Streets

Welcome back to SavvyMonk, your go-to source for AI and tech news that actually matters.

Remember when the whole world lost its mind over Pokémon Go in 2016? Turns out all those hours wandering around scanning landmarks weren't just for bragging rights. Niantic Spatial just revealed it used billions of player-captured images to build a positioning system that guides delivery robots through city streets.

Let's get into it.

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

When Pokémon Go launched in the summer of 2016, 500 million people installed it in 60 days. Players walked through parks, down sidewalks, and around landmarks, phones pointed at the world.

Niantic, the company behind the game, wasn't just running an AR experience. It was collecting data.

In the fall of 2020, Niantic introduced AR Mapping. The feature let players scan real-world locations by walking around them while the app captured images from their phone cameras. It was framed as “field research.”

Complete a scan, get a reward. Simple enough. But every scan uploaded detailed visual data along with metadata like the phone's exact position, orientation, direction, and speed.

Over the years, across Pokémon Go and Niantic's other AR games like Ingress and Pikmin Bloom, the company accumulated well over 30 billion images of urban environments. Each image is tagged with precise location data. The same spots captured over and over, in different lighting, weather, and seasons. That's not a photo album. That's a training dataset.

The Spin-Off

In March 2025, Niantic announced it was selling its games division to Scopely for $3.5 billion. Scopely, for context, is owned by the Savvy Games Group, part of Saudi Arabia's Public Investment Fund.

But Niantic didn't sell everything. It spun off a separate company called Niantic Spatial, which kept the mapping data, the geospatial technology, and the AR platform. The deal closed in May 2025.

Niantic Spatial launched with $250 million in capital. Its mission: build a “Large Geospatial Model” that helps machines understand the physical world.

How the Visual Positioning System Works

The core product is a Visual Positioning System (VPS). It works differently from GPS. Instead of relying on satellite signals, which can drift 50 meters or more in dense cities, VPS figures out where a device is by analyzing what its camera sees and matching it against Niantic's massive database of real-world images.

McClendon told MIT Technology Review that the system covers well over one million locations worldwide and can pinpoint a device's position within a few centimeters. That level of precision matters when you're a small robot trying to find the right entrance on a crowded city block.

For each location in the dataset, Niantic Spatial has thousands of images taken from slightly different angles, at different times of day, in different weather. That density is what makes the system reliable even when a street looks entirely different from one visit to the next.

Enter the Robots

On March 10, 2026, Niantic Spatial announced a partnership with Coco Robotics, a startup that builds autonomous sidewalk delivery robots. Coco was founded in 2020 and has completed over 500,000 zero-emission deliveries across cities like Los Angeles, Chicago, Miami, and Helsinki. It delivers through Uber Eats, DoorDash, and Wolt, serving more than 3,000 merchants.

Coco's robots need precise positioning to do their jobs. They have to navigate around construction, find specific building entrances, and handle curbs and bike lanes. GPS alone can't do that in a dense city. Niantic Spatial's VPS fills the gap.

Coco CEO Zach Rash said the system gives robots reliable localization that improves navigation. Hanke put it more bluntly: getting Pikachu to realistically run around a park and getting a delivery robot to safely move through a city are fundamentally the same problem.

Coco also recently launched Coco 2, its next-generation robot that can operate fully autonomously and move onto streets and bike lanes at up to 13 mph (20.92 km/h). The company plans to scale to thousands of robots globally by the end of 2026.

The Consent Question

This is where things get uncomfortable. Niantic says the AR scanning feature was always opt-in. Players had to visit a specific location and tap to scan. The game made it clear they were collecting AR mapping data.

But “opt-in” doesn't mean “fully informed.” Most players probably understood they were helping improve the game's AR features. Very few would have guessed their scans would end up training navigation systems for delivery robots years later. One viral post on X summed up the feeling: 143 million people thought they were catching Pokémon. They were actually building one of the largest real-world visual datasets in AI history.

Niantic Spatial says the scans are not connected to player accounts and that future data will be collected through an opt-in third-party service. But the comparison to Google's CAPTCHA system is difficult to ignore. For years, people clicked on traffic lights and crosswalks to prove they were human. That data helped train computer vision models for self-driving cars. The pattern is the same: make a task feel like a small, harmless interaction. Use the output for something much bigger.

The Bigger Picture

Niantic Spatial's ambitions go beyond delivery robots. McClendon has said the company wants to build a “living map” of the real world that updates in real time as new data comes in. Robots using the VPS would also feed images back into the system, creating a loop where the map gets more accurate with every delivery.

The company positions this as a “world model,” a term that's getting a lot of attention in AI right now. Google DeepMind and World Labs are building virtual world models for training AI agents. Niantic Spatial is coming at it from the opposite direction, starting with real-world data and layering intelligence on top.

If they pull it off, the implications stretch far beyond pizza delivery. Autonomous vehicles, AR glasses, city planning, logistics. Any system that needs to understand physical space could plug into it.

The Bottom Line

Pokémon Go was always more than a game. It was one of the largest crowdsourced data collection efforts ever built, and most participants had no idea what they were really contributing to. The technology is impressive. But the playbook of turning user activity into AI training data without clear communication about the end use is a pattern the tech industry keeps repeating. The question isn't whether this data is useful. It's whether the people who created it ever had a real say in how it would be used.

Category: Data Privacy Audit

“Review the privacy policies and data collection practices of [App Name]. Identify what types of data the app collects, how it describes the use of that data, and whether any language is vague enough to allow the data to be repurposed beyond its stated purpose. Summarize the key risks a typical user should be aware of.”

Every app on your phone is collecting something. Usually, the value of that data only becomes clear years later, when someone figures out a new way to use it. Pokémon Go is just the most visible example. The real question for all of us is simple: do we actually read what we agree to, or do we just tap “accept” and keep playing?

Hit reply; I read every response.

See you in the next one.

— Vivek

P.S. Know someone who cares about AI, privacy, or how their data gets used? Forward this their way. They can subscribe at https://savvymonk.beehiiv.com/