How is Waymo training its autonomous vehicles?

Waymo is training its autonomous vehicles using a hyper-realistic virtual world created with Google's Genie 3 AI model.

What kind of scenarios can Waymo simulate?

Waymo can simulate extreme and rare scenarios such as tornadoes, snow-covered bridges, flooded neighborhoods, and fires.

What AI model does Waymo use for its virtual simulations?

Waymo uses Google's Genie 3 AI model to generate interactive 3D environments for its virtual testing.

Home / Technology / Waymo Simulates Tornadoes for AI Driver Training

Waymo Simulates Tornadoes for AI Driver Training

6 Feb

•

Summary

Waymo created a hyper-realistic virtual world for AI driver training.
The system uses Google's AI model, Genie 3, to generate 3D environments.
Simulations include extreme scenarios like tornadoes and fires.

Waymo Simulates Tornadoes for AI Driver Training

Waymo is leveraging Google's AI to create a hyper-realistic virtual world for its autonomous vehicle development. This advanced simulation environment, built with Google's Genie 3 AI model, allows Waymo to test its robotaxis in a vast array of scenarios. The system can generate interactive 3D environments from simple text or image prompts, specifically adapted for the demands of autonomous driving.

Simulation is crucial for autonomous vehicle development, enabling tests in rare and dangerous situations without physical risk. Waymo can now simulate events like tornadoes, snow-covered bridges, flooded neighborhoods, or even fires. These simulations help the Waymo Driver encounter and learn from 'edge cases,' which are unusual circumstances that might occur infrequently in real-world driving.

Genie 3 offers unique features for Waymo's simulations, including driving action control for 'what if' scenarios, scene layout control for customizing road conditions and traffic, and language control for adjusting time-of-day and weather. This flexibility allows for testing in challenging conditions like low light or high glare. Waymo can also convert real-world dashcam footage into virtual environments for highly realistic testing.