The Download: Pokémon Go to train world models, and the US-China race to find aliens

Niantic Spatial, an AI company spun out from Pokémon Go creator Niantic last year, is repurposing one of the largest augmented-reality datasets ever created. The company's CTO Brian McClendon revealed that the 2016 megahit, installed by 500 million people in just 60 days, generated a "vast trove of crowdsourced data" about real-world locations and spatial relationships. This data is now being used to train world models—a cutting-edge AI technology that connects large language models (LLMs) to physical environments.

World models represent a significant advancement beyond traditional navigation systems, providing AI with contextual understanding of spaces rather than just mapping coordinates. Niantic Spatial's approach could enable delivery robots to navigate complex urban environments with centimeter-level precision, recognizing landmarks, obstacles, and optimal pathways learned from millions of human interactions. The technology essentially creates a shared digital twin of the physical world, continuously updated through crowd-sourced inputs.

The move positions Niantic Spatial at the intersection of two rapidly evolving fields: augmented reality and autonomous systems. While most robotics companies rely on proprietary sensor data, Niantic's approach leverages what McClendon describes as "the world's first augmented-reality megahit" to create more robust and scalable navigation solutions. This could significantly reduce deployment costs for robotic delivery services while improving reliability in dynamic environments where traditional GPS and mapping often fail.