Cheeky Pint The 20-year journey to fully autonomous cars with Dmitri Dolgov of Waymo
178 snips
Mar 24, 2026 Dmitri Dolgov, Waymo co-CEO and an early Google self-driving engineer, charts the 20-year road from lab experiment to 500,000 rides a week. He gets into why lidar still matters, how Teacher and Critic models train the driving AI, why supervised cars will not simply become robotaxis, and what it takes to scale autonomy across cities.
AI Snips
Chapters
Transcript
Episode notes
Great Robotaxis Must Be Safe Smooth And Social
- Waymo optimizes for more than safety by also targeting smoothness, predictability, and socially compatible behavior for riders and everyone else on the road.
- Drop offs are especially nuanced because the car must infer rider intent while avoiding driveways, bad stops, and unnecessary friction.
Waymo Shifted From Research To Deployment
- Dolgov says Waymo has moved from scientific research into accelerated global deployment, with no core technology gap blocking major new driving capabilities.
- Expansion now depends on specialization and validation for each domain, especially cold weather hardware and slippery surface control.
Waymo Expands By Operating Domain Not By City
- Waymo thinks in operating domains like freeways, weather, fog, and density, not cities, and maps those capabilities onto each market.
- The jump from gen four to gen five came from a bigger AI backbone trained on data from many US states, enabling harder launches like San Francisco.