World models: an investment boom around simulations for training robots and autonomous vehicles
World models – technology generating 3D simulations of the physical world for training robots and autonomous vehicles – are attracting major investment. Niantic has reoriented itself toward this field after selling Pokémon Go, AMD acquired World Labs for $8+ billion, and Yann LeCun raised a billion dollars for his own startup.
Niantic, the creator of Pokémon Go, has reoriented itself toward developing world models after selling the game last year – a technology that, unlike language models, works with visual inputs and generates 3D digital environments simulating physics. According to Brian McClendon, CTO of Niantic Spatial, the company wants to use its real-world data to create replicas of it. According to Oliver Cameron, CEO of the startup Odyssey, a world model, unlike a language model, does not predict the next word but the next visual state of the world based on an image – the resulting simulations can be used as a training environment.
According to the source, these environments are meant to train robots without requiring physical movement – a 3D model can be created from photographs of a space and millions of simulations can be run. World models work similarly for autonomous vehicles: the startup Wayve, according to its chief scientist Jamie Shotton, builds a generative world model partly on dashcam footage, which can reconstruct dangerous road situations unsuitable for testing in reality. The technology is also expected to find use in video games, where according to the article companies such as Tencent are investing, and Google regularly presents its own progress in this area.
Large investments are flowing into the field: Yann LeCun, former chief AI scientist at Meta, raised over a billion dollars for his startup Advanced Machines Intelligence. The startup World Labs, founded in 2024 by computer scientist Fei-Fei Li, was acquired this year by AMD for more than $8 billion. The article's author expects that 2027 could be a key year for the expansion of home robotics and autonomous vehicles powered by world models.
The article further describes that training data for world models require real human recordings – the author themself recorded videos of household activities (doing laundry, washing dishes) using an iPhone strapped to their forehead and sold them through an app for, in their own words, a low fee. According to the author, as the field develops, similar occasional work opportunities may increase.
Why it matters
World models are changing how robots and autonomous vehicles are trained – instead of costly physical tests, according to the source, they allow millions of simulations to be run in a digital environment, which could make the development of robotics and autonomous systems planned for 2027 cheaper and faster. For individuals, this means growing demand for real-life data, including a new form of occasional work involving recording videos of household activities.
Two audiences, two different impacts
What this means
For individuals
Demand is growing for occasional work involving recording videos of everyday household activities for training world models, which according to the article's author brings rather low earnings but could expand in the coming years.
For a business
According to the source, companies developing robots or autonomous vehicles can use world models for simulated training instead of costly physical tests, which opens up a new market for data and simulation tool providers as well as new investment opportunities (see AMD's acquisition of World Labs for more than $8 billion and the investment of over a billion dollars in the startup Advanced…
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.