From Models to Mobility: Building Waymo with Dmitri Dolgov

Waymo co-CEO Dmitri Dolgov details the transition of autonomous driving from long-term scientific research to a phase of accelerated global scaling and commercial deployment. The architecture behind the Waymo driver relies on a foundation model specialized into three off-board "teacher" models—the driver, the simulator, and the critic—which are then distilled into efficient, real-time inference models. This system integrates data from cameras, LiDAR, and radar to create a robust, 360-degree understanding of the physical world, enabling the vehicle to navigate complex scenarios like detecting pedestrians obscured by buses. With over 500,000 weekly autonomous rides across 11 cities, the focus has shifted toward operational efficiency and hardware optimization. Future progress centers on leveraging general world knowledge from vision-language models to improve zero-shot capabilities and expanding service to new international markets like London and Tokyo.