tech
February 7, 2026
Waymo leverages Genie 3 to create a world model for self-driving cars
With Genie 3, Waymo wants to explore rare and even impossible driving conditions.
TL;DR
- Waymo is using Google DeepMind's Genie 3 to create a 'Waymo World Model' for training self-driving car AI.
- The model generates hyper-realistic simulated environments to train AI on rare or impossible driving situations.
- It addresses limitations of relying solely on real-world data by simulating events not frequently encountered.
- Genie 3's long-horizon memory allows simulations to retain context over longer periods.
- The Waymo World Model integrates both 2D video and 3D lidar outputs for comprehensive environmental understanding.
- This simulation technology enables 'driving action control,' allowing engineers to modify real-world video scenarios with simple prompts.
- The model can also generate matching sensor data for existing dashcam videos, showing how the AI would perceive them.
- Waymo plans to use this for 'mutating' real videos, changing conditions like time of day, weather, or adding unusual elements like elephants.
- This is particularly useful for adapting cars to varied conditions found in new markets like Boston and Washington, D.C.
Continue reading
the original article