Pokémon GO’s Hidden Legacy: 3D AI Maps
Have you ever seen a robot delivering pizza in the middle of a busy city? I recently watched a video of a delivery robot navigating a forest of skyscrapers, and it made me wonder: how well can these navigation systems actually perform?
As I looked into it, I was surprised to find an unexpected savior helping these robots find their way: the augmented reality game Pokémon GO, which we spent years playing while wandering the streets with our smartphones. I took a closer look at how the street photos I once took without much thought were transformed into 3D maps for robots and AI, and how my own gameplay played a pivotal role in the process.
Robots Lost in the City?
One of the physical hurdles facing autonomous driving and delivery robots is localizing themselves in urban environments. In areas densely packed with high-rise buildings, GPS signals bounce off structures, causing significant interference. If a robot has a positioning error of several meters or more, it ends up stopping at the wrong place instead of the customer’s doorstep.
How Was Pokémon GO Data Used?
The Niantic logo that appears when launching Pokémon GO
Coco Robotics, which operates over 1,000 delivery robots in the U.S. and Europe, partnered with Niantic Spatial to solve this problem. Niantic Spatial is a spatial AI company recently spun off from the developer of Pokémon GO. Their core asset is a crowd-sourced library of 30 billion images, captured by users at over a million locations worldwide over the past decade.
My Pokémon, sorted by CP after logging in
If you recall, around 2016, the streets were filled with people walking around holding their smartphones. I vividly remember wandering through my neighborhood trying to catch rare monsters. At the time, key locations like gyms and PokéStops were placed on real-world landmarks, shop signs, and intersections within the game.
To complete in-game missions and earn rewards, users were more than happy to scan their surroundings with their cameras. Our habit of pointing our lenses at buildings, even on rainy days or at night, was actually the process of collecting high-quality spatial data.
Over the past decade, that has accumulated into a staggering 30 billion field photos from around the world. It is truly remarkable that the passion of users wanting to catch virtual characters has created the largest and most dense dataset of alleyways on the planet.
How Does Pokémon GO’s Visual Positioning Work?
An Elite Raid gym near my house
To overcome the blind spots of satellite signals, “Visual Positioning” was introduced. The lens on a device captures the exterior of nearby buildings or road structures in real-time and compares them against the 30 billion images we helped accumulate.
- VPS (Visual Positioning System): A technology where a robot’s camera recognizes the contours of surrounding buildings, signs, and structures, and matches them against pre-learned 3D image data.
- Minimizing Error: It reduces GPS errors to the centimeter level, allowing the robot to accurately determine its exact location and orientation.
The vast amount of data—which includes various environmental variables like time of day and weather—has boosted the accuracy of VPS. Thanks to countless people repeatedly taking photos under different conditions, robots can now correct their location to the centimeter even in highly unfamiliar environments. The act of users scanning the streets with their smartphones for battles and rewards effectively built precise 3D maps for robots.
3D Maps Created by Pokémon GO Users?
Having accumulated this massive data, Niantic has even set up a dedicated spatial modeling team to provide a technical infrastructure for the AI industry. They have created a 3D map for machines, built not from the perspective of car-centric flat maps, but specifically from the viewpoint of pedestrians and the sidewalk level.
Now, new visual information collected by autonomous devices roaming the streets is reflected in the central system in real-time. Seeing a feedback loop where real-world road conditions and building changes are updated in the digital world without delay makes the pace of technological advancement feel truly intense.
It is fascinating that the time we spent pointing our lenses to catch Pokémon has evolved into a massive infrastructure for autonomous driving. The steps of millions of people roaming alleyways in search of rare monsters have, in effect, become a guide that helps machines perceive physical reality. Seeing those monster balls I tossed for fun turn into an asset that powers the robotics industry, I am very excited to see what other everyday activities will become the foundation for future innovation.
🔗 Original Post :
포켓몬고가 쏘아올린 작은 볼, 길 잃은 배달 로봇 구한 AI 3D 지도로!
https://blog.naver.com/PostView.naver?blogId=k5kun&logNo=224300739216