The 20-year journey to fully autonomous cars with Dmitri Dolgov of Waymo

Dmitri Dolgov, co-CEO of Waymo, discusses the technical architecture and future of self-driving technology. He details Waymo's sensor suite—lidar, radar, and cameras—and AI-driven system that builds a world model to make driving decisions in real-time. Dolgov addresses the end-to-end versus modular debate, explaining Waymo's approach of using a foundation model specialized into off-board teachers (driver, simulator, critic) then distilled into smaller models. He shares that Waymo currently operates 3,000 cars, providing half a million weekly rides across 11 U.S. cities. The conversation touches on the operational infrastructure behind Waymo, hardware evolution with the upcoming sixth-generation custom vehicle, and the potential for autonomous driving to reshape urban landscapes by reducing the need for parking.

Outlines

Part 1: Background and Technical Architecture

Part 2: AI Foundations and Training Models

Part 3: Safety and Real-World Navigation

Part 4: Hardware Evolution and Vehicle Design

Part 5: Scaling, Metrics, and Emergent Intelligence

Part 6: Future Outlook and Operational Infrastructure

Part 7: Corporate Culture and Vision

Sign in to continue reading, translating and more.

Continue

Cheeky Pint

Part 1: Background and Technical Architecture

Dmitri Dolgov's Journey: From Soviet Russia to Waymo's Co-CEO

From Google Interview to Waymo's Autonomous Driving Architecture

Waymo's Real-Time Inference and Cloud-Based Post-Trip Analysis

Part 2: AI Foundations and Training Models

Waymo's AI Foundation: Foundation Models, Simulators, and Critics

The Evolution of Self-Driving Tech: AI Breakthroughs and Iterative Learning

Encoding the World: End-to-End Models and the Role of VLMs

Beyond Imitation: Reinforcement Learning and the Need for Simulation

Intermediate Representations: Augmenting End-to-End Models for Scalability

Part 3: Safety and Real-World Navigation

Optimizing for Safety, Smoothness, and Socially Aware Driving

Drop-Off Nuances and the Long Tail of Freeway Complications

Waymo's Core Tech: Scaling Globally with AI and Specialization

Generalizing AI: Leveraging VLMs and Defining the Operating Domain

Part 4: Hardware Evolution and Vehicle Design

Waymo's Generational Leap: From Chandler to San Francisco

AI as the Backbone: The Shift from Version 4 to Version 5

Waymo's Sixth Generation: Custom Vehicle Design and Passenger Experience

De-Risking the Fundamentals: Prioritizing Software and Customer Adoption

Gen 6: New Vehicle, New Sensors, Similar Software

Hardware Cost Reduction: Unification, Simplification, and Mature Supply Chains

LiDAR vs. Radar: Complementary Sensing Modalities in Adverse Conditions

Part 5: Scaling, Metrics, and Emergent Intelligence

Global Expansion and the Thrill of AI Progress

Emergent Behavior: The Magic of Foundation Models and Distillation

Seeing Through the Bus: Sensor Fusion and Intermediate Representations

Waymo's Business Metrics: Cars, Rides, and Autonomous Miles

Part 6: Future Outlook and Operational Infrastructure

The Convergence of Autonomy: Level 4/5 vs. Driver-Assist Systems

The Qualitative Jump: From Driver-Assist to Full Self-Driving

The Operational Infrastructure Behind Waymo's Autonomous Fleet

Automated Charging, Rider Behavior, and the Future of Waymo's Reach

Second-Order Effects: Traffic Flow, Parking Lots, and Urban Landscapes

Part 7: Corporate Culture and Vision

Google's Faith in Waymo: A Long Journey and the Power of Iteration

Google's Talent: Culture, Vision, and Investment in Technical Talent

The 20-year journey to fully autonomous cars with Dmitri Dolgov of Waymo

Cheeky Pint

Part 1: Background and Technical Architecture

00:01Dmitri Dolgov's Journey: From Soviet Russia to Waymo's Co-CEO

Dmitri Dolgov's Journey: From Soviet Russia to Waymo's Co-CEO

01:50From Google Interview to Waymo's Autonomous Driving Architecture

From Google Interview to Waymo's Autonomous Driving Architecture

03:39Waymo's Real-Time Inference and Cloud-Based Post-Trip Analysis

Waymo's Real-Time Inference and Cloud-Based Post-Trip Analysis

Part 2: AI Foundations and Training Models

06:26Waymo's AI Foundation: Foundation Models, Simulators, and Critics

Waymo's AI Foundation: Foundation Models, Simulators, and Critics

09:58The Evolution of Self-Driving Tech: AI Breakthroughs and Iterative Learning

The Evolution of Self-Driving Tech: AI Breakthroughs and Iterative Learning

11:24Encoding the World: End-to-End Models and the Role of VLMs

Encoding the World: End-to-End Models and the Role of VLMs

14:47Beyond Imitation: Reinforcement Learning and the Need for Simulation

Beyond Imitation: Reinforcement Learning and the Need for Simulation

17:36Intermediate Representations: Augmenting End-to-End Models for Scalability

Intermediate Representations: Augmenting End-to-End Models for Scalability

Part 3: Safety and Real-World Navigation

19:47Optimizing for Safety, Smoothness, and Socially Aware Driving

Optimizing for Safety, Smoothness, and Socially Aware Driving

21:11Drop-Off Nuances and the Long Tail of Freeway Complications

Drop-Off Nuances and the Long Tail of Freeway Complications

23:44Waymo's Core Tech: Scaling Globally with AI and Specialization

Waymo's Core Tech: Scaling Globally with AI and Specialization

25:32Generalizing AI: Leveraging VLMs and Defining the Operating Domain

Generalizing AI: Leveraging VLMs and Defining the Operating Domain

Part 4: Hardware Evolution and Vehicle Design

27:13Waymo's Generational Leap: From Chandler to San Francisco

Waymo's Generational Leap: From Chandler to San Francisco

28:50AI as the Backbone: The Shift from Version 4 to Version 5

AI as the Backbone: The Shift from Version 4 to Version 5

31:33Waymo's Sixth Generation: Custom Vehicle Design and Passenger Experience

Waymo's Sixth Generation: Custom Vehicle Design and Passenger Experience

33:07De-Risking the Fundamentals: Prioritizing Software and Customer Adoption

De-Risking the Fundamentals: Prioritizing Software and Customer Adoption

35:05Gen 6: New Vehicle, New Sensors, Similar Software

Gen 6: New Vehicle, New Sensors, Similar Software

36:19Hardware Cost Reduction: Unification, Simplification, and Mature Supply Chains

Hardware Cost Reduction: Unification, Simplification, and Mature Supply Chains

37:45LiDAR vs. Radar: Complementary Sensing Modalities in Adverse Conditions

LiDAR vs. Radar: Complementary Sensing Modalities in Adverse Conditions

Part 5: Scaling, Metrics, and Emergent Intelligence

40:19Global Expansion and the Thrill of AI Progress

Global Expansion and the Thrill of AI Progress

41:56Emergent Behavior: The Magic of Foundation Models and Distillation

Emergent Behavior: The Magic of Foundation Models and Distillation

43:26Seeing Through the Bus: Sensor Fusion and Intermediate Representations

Seeing Through the Bus: Sensor Fusion and Intermediate Representations

46:02Waymo's Business Metrics: Cars, Rides, and Autonomous Miles

Waymo's Business Metrics: Cars, Rides, and Autonomous Miles

Part 6: Future Outlook and Operational Infrastructure

47:49The Convergence of Autonomy: Level 4/5 vs. Driver-Assist Systems

The Convergence of Autonomy: Level 4/5 vs. Driver-Assist Systems

49:31The Qualitative Jump: From Driver-Assist to Full Self-Driving

The Qualitative Jump: From Driver-Assist to Full Self-Driving

50:41The Operational Infrastructure Behind Waymo's Autonomous Fleet

The Operational Infrastructure Behind Waymo's Autonomous Fleet

53:20Automated Charging, Rider Behavior, and the Future of Waymo's Reach

Automated Charging, Rider Behavior, and the Future of Waymo's Reach

55:07Second-Order Effects: Traffic Flow, Parking Lots, and Urban Landscapes

Second-Order Effects: Traffic Flow, Parking Lots, and Urban Landscapes

Part 7: Corporate Culture and Vision

57:42Google's Faith in Waymo: A Long Journey and the Power of Iteration

Google's Faith in Waymo: A Long Journey and the Power of Iteration

1:01:01Google's Talent: Culture, Vision, and Investment in Technical Talent

Google's Talent: Culture, Vision, and Investment in Technical Talent