Waymo Just Took a Not-So-Subtle Dig at Tesla's Self-Driving Strategy

Waymo didn't mention Tesla by name. It didn't need to.

On Wednesday, the company published a blog post laying out 10 takeaways from 200 million miles of autonomous operation. Written by Srikanth Thirumalai, Waymo's vice president of onboard software, the post reads like a point-by-point rebuttal of just about every pillar of Tesla's self-driving philosophy — the camera-only sensor suite, the end-to-end neural network, the belief that no HD maps are needed, and the idea that a supervised driver-assist system can gradually mature into full autonomy.

The Sensor Argument

The headline jab is about hardware. Tesla runs vision-only, and Elon Musk has famously called lidar a "fools errand." Waymo's latest system, in its Ojai van, takes the opposite view with 13 cameras, 4 lidar units, and 6 radars, plus microphones.

"Cameras are incredible, but they aren't enough. By combining inputs from cameras, lidar, and radar, the Waymo Driver creates a rich, redundant world view that no single sensor can replicate." — Srikanth Thirumalai, Waymo

Thirumalai breaks down the division of labor: lidar "capture[s] 3D geometry with millimeter precision," radar tracks velocity and picks up objects in conditions cameras struggle with — heavy fog, for example — and cameras excel at reading street signs and traffic lights. He also defends HD maps, which Tesla says it doesn't need, calling them "incredibly helpful in poor visibility and complex thoroughfares."

The Black Box Problem

Waymo's second big swing is at end-to-end AI, the approach where raw sensor data goes in one side of a neural network and steering commands come out the other. It's central to how Tesla builds FSD, and other players have adopted it too.

"Pure end-to-end (E2E) neural architectures, where a model takes in raw pixels and directly outputs steering commands, run the risk of black box failures. That means, it's hard to understand how decision making happens in full E2E systems."

"A False Summit"

The sharpest line is saved for Tesla's core promise: that Full Self-Driving (Supervised) will keep improving — fed by an enormous fleet of customer-driven miles — until the human can finally step out of the loop.

"Simply improving a driver-assist system (L2) for full autonomy is a false summit. You can run billions of miles in simulation or with human supervision, but an AV system only truly matures when it is solely responsible for the driving task."

Thirumalai's point is that full autonomy "exposes the system to the true gravity of its decisions and reveals novel situations that humans or simulations might unconsciously smooth over." In other words, a supervised system never quite finds its own limits because a person keeps rescuing it before it does.

Why Now?

Of course Waymo thinks Waymo is right. But the timing is worth noting. The company is making this argument more explicitly than it has before, right as Tesla is finally landing the autonomy milestones it has hyped for years. Tesla appears to have removed safety operators from its Robotaxi service in several markets, and it has a launch event scheduled for the purpose-built Cybercab, though exactly what that event will cover isn't clear.

Tesla's counterargument is that a simpler, cheaper, camera-only stack is far easier to scale. Waymo's counter-counterargument is scale it already has: over 500,000 driverless trips per week. The next few years will settle it.

What This Means If You're Still the Driver

Here's the practical translation for the rest of us. Whichever philosophy wins, the car in your driveway today is a Level 2 system, and both companies agree on that much — Waymo explicitly, Tesla by way of the word "Supervised" in FSD's name. Your hands and eyes are still part of the safety stack.

A few things worth keeping in mind:

• Keep your cameras clean. If your car sees the world through cameras alone, a film of road salt, bug splatter, or a fogged-up pillar cam directly degrades what the system can do. A quick wipe of the B-pillar and rear cameras before a long drive costs nothing.
• Bad weather is where vision-only struggles most. Waymo's whole radar argument is about fog, spray, and low visibility. If FSD or Autopilot feels less confident in heavy rain, that's not your imagination — plan to take over sooner in those conditions.
• Supervision means supervision. Nag-defeat gadgets that fake steering-wheel torque undercut the one thing that makes an L2 system safe. They're a bad idea, and we don't sell them.
• Carry roadside gear regardless. No sensor suite prevents a nail in a sidewall or a dead 12V. A tire repair kit, jump pack, and warning triangle are the least glamorous and most useful things in your frunk.

The lidar-versus-cameras fight is fascinating to watch, but it's being settled in fleets, not in personal cars. Until it is, the smartest accessory for any driver-assist system remains an attentive driver — plus a clean set of lenses and a well-stocked emergency kit.