Waymo says it is using DeepMind's Genie 3 to create realistic digital worlds for its autonomous driving technology to train on edge-case scenarios
Alphabet Inc.'s Waymo said it is using DeepMind's Genie 3 AI model to create realistic, digital worlds for its autonomous driving technology to train on edge-case scenarios.
The new element is the use of Genie 3, which DeepMind introduced as a model able to generate interactive 3D worlds from prompts. It puts a newer generative-world model into an established autonomous-vehicle simulation workflow.
First-order effects
Waymo can use Genie 3-generated environments to expand training and evaluation around unusual driving situations, rather than relying only on its existing simulated-world tooling and road experience.
DeepMind gains an internal Alphabet deployment for Genie 3 in a safety-critical operational setting, linking its world-model research directly to Waymo’s vehicle-development pipeline.
Second-order effects
Waymo’s simulation operation will need to assess whether generated scenarios are sufficiently realistic and useful for validating driver behavior; scenario quality becomes as important as the volume of worlds produced.
Other autonomous-driving developers face added pressure to improve simulation capabilities, either by developing generative-world tooling or securing comparable model access.
Third-order effects
If generative world models reliably produce useful edge cases, autonomous-vehicle development could shift further toward software-defined simulation as a core testing and iteration layer, alongside real-world driving.
The strategic advantage may increasingly sit with companies that combine frontier models, vehicle data, and deployment feedback within one organization—an outcome that is not guaranteed by generation capability alone.
The trend: Generative world models are moving from general AI demonstrations into industrial simulation systems for training and testing embodied AI.
WOW this is so incredibly cool. World models are helping self-driving cars simulate long-tail scenarios, which is crucial for robust autonomous systems. As powerful models like Genie 3 continue to advance, I'd expect post-training and adaptation to different hardware stacks
We've been working with our friends at Waymo for quite some time to bring world models to this critical domain. Reliably simulating rare, safety-critical OOD scenarios is one of the first viable applications of world models in the real world. We've demonstrated this just
We're excited to introduce the Waymo World Model—a frontier generative mode for large-scale, hyper-realistic autonomous driving simulation built on @GoogleDeepMind's Genie 3. By simulating the “impossible”, we proactively prepare the Waymo Driver for some of the most rare and [im…
I thought Genie 3 was cool, but not yet useful. Just a week later and now we get to see it's real world potential! @Waymo released their first world model built on top of @GoogleDeepMind 's Genie 3! The model can take real dashcam footage and expand it into interactive [video]
Genie 3 🤝 @Waymo The Waymo World Model generates photorealistic, interactive environments to train autonomous vehicles. This helps the cars navigate rare, unpredictable events before encountering them in reality. 🧵 [image]
Simulation has always been a critical component of Waymo's AI ecosystem and one of the 3 key pillars of our approach to demonstrably safe AI in the physical world. The Waymo World Model is a high-fidelity, controllable environment helping the Waymo Driver master the long tail.