Waymo's Dmitri Dolgov: 20 Million Rides and the Road to Full Autonomy
In a Nutshell
Waymo co-founder Dmitri Dolgov recounts his journey from DARPA challenges and early Google self-driving efforts to achieving 20 million fully autonomous rides across 11 cities, powered by a multimodal foundation world model that drives end-to-end from sensors to actions while fueling simulators and critics. The sixth-generation hardware enables rapid scaling—doubling from 100M to 200M miles in six months—with Waymo proving 13x safer than humans for serious injuries, exemplified by superhuman maneuvers like LiDAR-detected pedestrian saves. Future plans target global expansion to cities like London and Tokyo, emphasizing persistent execution for full autonomy and saving lives amid 1.19M annual road deaths.
These notes were generated by AI and may contain inaccuracies.
Host introduces Dmitri Dolgov, Waymo founder in AI for almost 20 years, persistent through great and tough times. Audience poll: many have ridden in Waymo, love the experience. Dolgov confirms he uses it daily.
Dolgov born in Soviet Union, raised in States, attended Moscow Institute of Physics and Technology (same school as parents). Traveled: year in Japan, high school in States, college in Russia for math and physics. Shaped technical foundation and ability to learn independently.
Earned PhD in AI. In 2005, joined DARPA Grand Challenge and Urban Challenge. Light switch moment: technology interesting, mission powerful, hands-on product. Never looked back.
Started 2009 at Stanford Automotive Lab (host worked on solar car side). First years: learning problem space for public roads. Goals: drive 100,000 miles in full autonomy; 10 routes (100 miles each, difficult Bay Area routes) end-to-end without intervention (safety driver present). Small team of ~12, worked 24/7, coding/hardware daytime, testing nights. Achieved in 18 months.
From impossible goals to hundreds of millions of miles today.
Early days most fun professionally: momentum, progress every hour, doing everything (hardware setup, sensor calibration, pose estimation, core software/algorithms, tools/UIs/user experience). Convinced team to pursue fully autonomous product.
AV hype cycle: easy to start, hard to productize full autonomy/superhuman performance. Breakthroughs (convolutional nets, transformers, LLMs) reshape early curve but not long tail. Key: mission importance (crash every 26 seconds worldwide), no easy wins/silver bullets, right stamina.
Core of AI ecosystem: powers driver, simulator, critic.
Multimodal world action language model:
- Understands world physics/dynamics.
- Understands good driving, effects on other agents.
- Multimodal: images/video, LiDAR, radar.
- 3D spatial properties, dynamics, physics, behaviors (cars, pedestrians, cyclists).
- Active participant: controllable, good agent understanding.
- Language-aligned: VLM general knowledge for semantics/social aspects. Productionized for high performance/accuracy/realism.
Waymo foundation model is end-to-end (sensors to actions). Learns rich representations (e.g., encoder-decoder, perception-planning). Beyond vanilla end-to-end: augment with structured materialized intermediate representations for:
- Runtime validation.
- Richer training/evaluation (closed-loop eval/training, RL reward functions).
- Essential for human feedback, full autonomy, superhuman safety, scale (hundreds of millions miles). Not needed for driver assist/prototypes/small scale.
Most advanced sensor suite: performance, simplification, cost reduction, high-volume production. Powers O Hi vehicle platform. Started fully autonomous operations (employees only, public later 2024). Dolgov's first ride: felt like living room (same external size as i-Pace, spacious back, new screens, auto-sliding doors). Designed around rider experience.
16 years to 100M miles, 6 months to 200M. 11 cities, more coming.
Exponential scaling: 8 years for fully autonomous public service in 4 cities; launched 4 cities in one day recently. >20M fully autonomous rides, 10M in last 7 months. Process: data collection/environment characterization, driver validation, community conversations to earn trust. Driver generalizes well, needs rigorous evaluation.
Dolgov uses daily (rode from Palo Alto to SF on freeways). Family (3 kids) uses/loves it; kids annoyed by human-driven cars. Kids' cool spots: doggies, Waymos.
1.19M annual road deaths worldwide. Safety non-negotiable from day one: model architecture, training/eval, mindset. Avoid capability-first to 90%, focus on nines. >4M fully autonomous miles/week. Over 170M miles: Waymo 13x safer than humans for serious injury collisions in operating cities. Prevents serious injury every 8 days.
- Young woman on e-scooter fell in front of Waymo; superhuman swerve/brake, all safe.
- SF intersection: bus blocks view, Waymo at green light detects pedestrian via LiDAR bounce under bus (sparse foot movement return), predicts, reacts defensively, nudges around.
Heads-down execution: from sequential de-risking to rapid parallel global commercialization. Expand in 11 US cities, add new geographies. International: London, Tokyo announced this year. Accelerating deployment.
All in service of our mission. Good news to our team in London. We covered a lot, Dmitri, from the very early days where you could get a lot of distance with not a lot of technology to then persisting through extremely hard times in autonomous vehicles and getting that extra mile. We talked about world models, driver simulator critic architecture. We got into the hardware, the sixth generation hardware, safety and scaling. But most of all, I hope that we learned a little bit more about Dmitri, the man who's brought the magic that is Waymo to so many of us. As I've gotten to know you more and more, I'm constantly struck not only by your brilliance and your persistence and performance, but also your humility. It says a lot for accomplishing this much.
Thank you, Dmitri. Please join me in thanking Dmitri for all he does. [Applause]
Thank you. Thank you. Thank you. And the many lives saved. Thank you.
Keep Sequoia Capital in your library
Save the videos and channels worth coming back to, and find them again in one place.
