Niantic Repurposes Pokemon Go AR for Robot Maps

🔑 Enhanced Key Takeaways

• •Niantic Spatial, a spin-off company announced in March 2025, developed the Visual Positioning System (VPS) using data from Pokémon Go and Ingress.[1][2][4]
• •The VPS model was trained on 30 billion images captured by players, with dense coverage around over one million game hotspots worldwide for high-accuracy positioning.[1][2]
• •Coco Robotics' delivery robots use four hip-height cameras to feed images into the VPS, enabling precise navigation despite perspective differences from player data.[2]
• •The system supports direction-aware positioning and extends accuracy beyond hotspots through model predictions on sparse data.[2]

🛠️ Technical Deep Dive

• •VPS uses visual landmark recognition from smartphone images to achieve centimeter-level positioning, trained via photogrammetry on player-submitted AR scans from Pokémon Go's AR Mapping feature introduced in fall 2020.[4]
• •Model processes snapshots of buildings to predict location and facing direction, with over 30 billion urban images, densest at one million+ game hotspots.[1][2]
• •Integrates with robot hardware like Coco's four omnidirectional cameras at hip height; data adaptation handles viewpoint shifts from player eye-level.[2]

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline