Waymo Catches World Mannequin Fever, and the Solely Prescription Is Extra World Fashions

Waymo autos have reportedly racked up greater than 200 million miles of autonomous driving on public roads. However it’s but to run right into a twister or an elephant, and odds are that it’d reply poorly if it did. To attempt to assist with these once-in-a-billion-miles eventualities, Waymo introduced Friday that it’s introducing Waymo World Model, a generative AI mannequin that it’ll use to run near-endless conditions to strive to verify its vehicles are ready for the unpredictable, which additionally simply occurs to suit into the most recent pattern within the AI area.

To be clear, Waymo’s world mannequin makes about as a lot sense as any use case for the expertise. The corporate has a ton of high-definition knowledge that it has collected from its time on the highway that it may well use to generate real looking re-creations of roads. However, the corporate mentioned, as a substitute of constructing a mannequin based mostly solely on that data, it’s going to make use of Google’s Genie 3 mannequin to place its vehicles in simulated conditions that stretch past what’s already in its knowledge set collected from cameras and lidar sensors.

We’re excited to introduce the Waymo World Mannequin—a frontier generative mode for large-scale, hyper-realistic autonomous driving simulation constructed on @GoogleDeepMind’s Genie 3.

By simulating the “unimaginable”, we proactively put together the Waymo Driver for among the most uncommon and… pic.twitter.com/Pl80OMDqLC

— Waymo (@Waymo) February 6, 2026

Google made a splash final month when it released a beta version of Genie 3 to the public, permitting a subset of paid subscribers to generate 3D worlds with real looking physics. Not like a big language mannequin (LLM)—the underlying expertise that powers most AI instruments together with Google’s personal Gemini—which use the huge quantity of coaching knowledge they’re given to foretell the more than likely subsequent a part of a sequence, world fashions are skilled on the dynamics of the true world, together with physics and spatial properties, to create a simulation of how bodily environments function.

Genie 3 🤝 @Waymo

The Waymo World Mannequin generates photorealistic, interactive environments to coach autonomous autos.

This helps the vehicles navigate uncommon, unpredictable occasions earlier than encountering them in actuality. 🧵 pic.twitter.com/m6rlmkMFJH

— Google DeepMind (@GoogleDeepMind) February 6, 2026

Waymo plans to faucet into that to place its vehicles via a gauntlet of eventualities that they probably wouldn’t discover themselves in till it’s too late. That features excessive climate situations and pure disasters, so the vehicles can work out the way to navigate a twister or flood waters; sudden security emergencies like falling tree branches or an accident with numerous particles; and run-ins with the surprising, like an elephant on the highway. “By simulating the ‘unimaginable,’ we proactively put together the Waymo Driver for among the most uncommon and sophisticated eventualities,” the corporate mentioned.

The speculation is definitely sound, although world fashions aren’t with out their drawbacks. The early suggestions on the patron model of Genie 3 was a bit spotty, and world fashions are nonetheless vulnerable to hallucinations. We’re nonetheless within the earliest phases of seeing these fashions deployed, they usually have numerous room to iterate.

And Waymos have positively had their points in edge-case eventualities in the true world. Late final yr, a Waymo ran over a beloved bogeda cat named Kit Kat, and final month, one ran into a kid in a school zone. These interactions aren’t even notably uncommon for a driver to search out themselves in, so hopefully Waymo can refine its responses in these eventualities on prime of prepping for probably the most unlikely conditions.