Dmitri Dolgov on Waymo, the DARPA Urban Challenge, and the Evolution of Self-Driving AI

This episode features Dmitri Dolgov, CTO of Waymo, tracing his journey from early computer science fascination to leading the autonomous driving revolution. Dolgov recounts his childhood introduction to computers in the late 80s, his early programming attempts, and his unexpected childhood dream of being a traffic control cop, which he humorously connects to his current work in transportation. He details his academic path through the Moscow Institute of Physics and Technology and a postdoc at Stanford, where he first delved into robotics through the DARPA Urban Challenge.

The conversation extensively covers the DARPA Urban Challenge, a pivotal event that ignited the self-driving car industry. Dolgov explains the challenge's setup—autonomous vehicles navigating a simulated city alongside human-driven cars—and his role in motion planning. He shares vivid memories, including the magical experience of seeing his code control a car for the first time and the nerve-wracking competition day. He also humorously recounts a specific bug, dubbed the "Stanford victory lap," where their car would perform an extra lap around an oval before proceeding, highlighting the intricate challenges and unexpected behaviors in early autonomous systems.

Dolgov then transitions to the birth of the Google Self-Driving Car Project in 2009, initiated by Larry Page and Sergey Brin. He outlines the ambitious early milestones: driving 100,000 autonomous miles and completing 10 distinct 100-mile routes with zero human intervention, which they achieved in under two years. This period was crucial for understanding the problem's immense complexity and building confidence in the technology's feasibility. He emphasizes the distinction between prototyping for learning and engineering a production-ready system.

A significant pivot occurred around 2013 when the project shifted from a driver-assist (L3) focus to the vision of building a fully driverless vehicle without a human safety driver. This decision, Dolgov states, was fundamental and led to Waymo's current path. He highlights the world's first fully driverless ride on public roads in 2015 using their custom-built "Firefly" vehicle as a key milestone, underscoring Waymo's leadership in deploying publicly accessible autonomous vehicles at scale, a testament to overcoming one of the 21st century's most difficult AI challenges.

Waymo is currently leading in the fully autonomous vehicle space in that they actually have an at-scale deployment of publicly accessible autonomous vehicles driving passengers around with no safety driver with nobody in the driver's seat.

This to me is an incredible accomplishment of engineering on one of the most difficult and exciting artificial intelligence challenges of the 21st century.

I remember going back and you know later at night trying to fall asleep and just being unable to fall asleep for you know the rest of the night just my mind was blown and yeah that that's what I've been you know doing ever since for more than a decade.

Our car in the hand which hit the checkpoint and then it would do an extra lap around the awful and only then you know leave and go on its merry way so over the course of you know the full day it accumulated some extra time and the problem was that we had a bug where it wouldn't you know start reasoning about the next waypoint and plan around to get to that next point until it hit the previous one.

The most important one is probably that we believe that it's doable and we've gotten far enough into the problem that you know we had a I think only a glimpse of the true complexity of the the domain.

The two milestones were you know one was to drive a hundred thousand miles in autonomous mode which was at that time you know orders of magnitude that more than anybody has ever done and the second milestone was to drive 10 routes each one was 100 miles long they were specifically chosen to become extra spicy.

It's a little bit like you know climbing a mountain where you kind of see the next peak and you think that's kind of the summit but then you get to that and you kind of see that that this is just the start of the journey.

It also became very clear and that it's even the way you go building a driver assist system is you know fundamentally different from how you go building a fully driverless vehicle.