Waymo's co-CEO Dmitri Dolgov laid out the clearest technical case yet for why cameras alone cannot take a self-driving system to full autonomy, arguing that "weak sensing" hits a safety ceiling long before it reaches superhuman performance. Speaking at Y Combinator's Startup School, Dolgov explained that while cameras can match human driving ability and power driver-assist products, they fundamentally cannot reach the reliability required for cars with nobody behind the wheel.

What Makes Waymo's Multi-Sensor Approach Different?

Dolgov spent considerable time explaining why Waymo uses three distinct sensing types instead of relying on cameras alone. Each sensor type addresses different real-world challenges that cameras cannot overcome on their own.

• Cameras: Provide high resolution and color information but are passive sensors that degrade significantly in darkness, glare, and dust storms, making them unreliable in adverse conditions.
• Lidar: Directly measures the three-dimensional structure of the world using active sensing, allowing it to see clearly in pitch darkness and straight into blinding sunlight where cameras fail.
• Radar: Punches through fog, rain, and snow while reading velocity directly using Doppler technology, providing critical information in weather conditions where cameras become nearly useless.

Dolgov emphasized that these sensors are not backups to each other. Instead, each one runs its own encoder, and the data fuses into a single view of the world that is "vastly superior to what you get with any one sensor," he explained. He illustrated this with specific scenarios where cameras fail completely: a dust storm in Phoenix where the camera sees almost nothing while lidar cleanly picks out a pedestrian at the roadside, or children chasing dogs across a pitch-black street with no headlights or streetlights.

Dolgov

How Does the "Exponential Ladder of Nines" Explain the Safety Problem?

Dolgov introduced a critical mathematical concept that separates demonstration projects from production autonomous vehicles. Reliability exists on what he calls an "exponential ladder of nines," where each additional nine of reliability requires roughly ten times more engineering effort than the previous one.

A demonstration might need one nine of reliability. A driver-assist product needs a few. But a car with children in the back and nobody driving needs a whole stack of them. The trap, Dolgov warned, is picking the technology with the fastest early ramp, the one that makes the best demo, and then projecting that steep slope forward, only to hit a plateau "way before the performance that is required by your product".

Dolgov

The real-world data supports this argument. Tesla's own robotaxi data shows a crash rate about three times worse than human drivers, even with a human safety monitor in the front seat. In contrast, Waymo's latest safety report, based on 220 million rider-only miles, shows 94% fewer serious-injury crashes than human drivers would cause over the same distance, roughly 17 times better.

"If the goal were to just approximately match human performance or to build an assist product, that's a very reasonable way to go. But if you're targeting full autonomy and strongly superhuman performance, you find that weak sensing just leads to a safety curve that flattens out way too early," said Dmitri Dolgov.

Dmitri Dolgov, Co-CEO at Waymo

What About the Cost Argument for Camera-Only Systems?

Dolgov also pushed back on the one knock that camera-only advocates always reach for: cost. Waymo is on its sixth generation of hardware, and each generation has drastically cut the price. Betting against sensors like lidar on today's prices means betting on "a number that has a fairly short shelf life," he warned.

This matters because the cost argument is the last technical leg the camera-only case has to stand on. As lidar and radar technology mature, prices continue to fall, similar to how advanced driver assistance systems like ABS and airbags went from expensive options to standard equipment.

How Is Waymo Scaling Its Autonomous Service?

While the sensor debate continues, Waymo is rapidly expanding its real-world operations. The company opened its doors to all public riders in Dallas on Tuesday, allowing anyone in the area to hail a fully self-driving ride in Jaguar I-Pace vehicles. The service won't yet operate at Dallas Love Field Airport or on freeways, though Waymo says it plans to open those routes after finalizing testing.

With the Dallas launch, Waymo's autonomous ride-hailing service is now available to the public in nearly a dozen cities, including the San Francisco Bay Area, Phoenix, Los Angeles, Miami, Orlando, Nashville, Atlanta, and Austin. The company also has waitlists for Houston and San Antonio and is expanding to more than two dozen other cities across the country, including locations with more variable climates like Chicago, Detroit, and Minneapolis.

To handle more variable climates, Waymo is rolling out its sixth-generation self-driving technology aboard its new Ojai vehicle, which can detect more details and gauge objects in a variety of lighting and weather conditions. Waymo is now serving around half a million paid trips a week across its operating cities and is scaling toward a 1-million-weekly-rides target.

The contrast in scale is striking. Tesla recently confirmed it has accumulated 380,000 driverless miles over the past year since launching its robotaxi, while Waymo's driverless service does that in a single day. This gap reflects the fundamental difference Dolgov described: the difference between a demonstration in a geo-fenced area with limited speed and low availability, and a production service running millions of miles weekly across diverse conditions and cities.