World Models Set to Transform Robotics and Autonomous Cars
Niantic, the creator of Pokémon Go, is now focusing on world models—visual AI systems that reconstruct real environments and predict future visual states. Chief technology officer Brian McClendon says the company aims to use its massive collection of scanned surroundings to generate 3D replicas that robots can practice in without moving physically.
World models differ from text-based large language models by taking an image as input and forecasting the next visual state. Oliver Cameron, CEO of Odyssey, describes these simulations as looking and feeling real, providing a digital training ground for both humanoid robots and driverless cars.
McClendon explains that a robot could be photographed in its intended workspace, have a 3D model built, and then run millions of simulated scenarios to learn the space’s parameters. Wayve’s chief scientist Jamie Shotton adds that for autonomous vehicles, a world model must understand physics, cause and effect, and how objects and people react to the vehicle’s actions. Wayve uses dash-cam footage to recreate hazardous road conditions that are unsafe to test in reality.
The technology is also attracting interest from the gaming sector, with companies like Tencent investing in world models that align with video-game physics and 3D rendering. Large AI firms such as Google continue to publicize improvements, while investors back notable figures: Yann LeCun raised over a billion dollars for his Advanced Machines Intelligence startup, and Fei-Fei Li’s World Labs was acquired by AMD for more than $8 billion.
Building these models requires extensive real-world video data. Gig workers are already recording daily activities—folding laundry, washing dishes, pouring drinks—and selling the footage through apps for modest compensation. As demand for such data grows, more gig labor will be needed to feed the expanding world-model ecosystem.
Why it matters
World models will shape the next wave of home robots and autonomous vehicles, directly affecting industries and consumers alike.
FAQ
What is a world model?
A visual AI system that takes an image of the current state and predicts the next visual state, enabling realistic simulations.
How are world models used for robots?
Robots can be trained in digital replicas of real spaces, running millions of simulated scenarios without physical movement.
Which companies are investing in world models?
Niantic, Odyssey, Wayve, Tencent, Google, and startups backed by Yann LeCun and Fei-Fei Li are all developing or funding world-model technology.
What labor is needed to build world models?
Gig workers record everyday activities on video, which is sold to AI labs to provide the real-world data needed for training.
Source: WIRED AI. See the original source for full details.
Scan with your phone
Scan with your phone