Browse
Log in / Register
Share
Published on April 22, 2025
Waymo, the Alphabet-owned autonomous driving leader, this week published a set of "10 AI lessons" drawn from over 200 million fully driverless miles. While the post never named Tesla, several of its points land squarely on the electric vehicle maker's self-driving strategy—especially the claim that trying to turn a consumer driver-assist system into full autonomy is a "false summit."
The sharp criticism comes as Waymo expands its commercial robotaxi service to more than 500,000 paid rides per week, while Tesla's long-promised "unsupervised" Full Self-Driving (FSD) remains limited to a small test fleet in Austin, Texas. The gap between the two approaches is widening, and Waymo is using its data to draw a line in the sand.
The most pointed lesson of the ten is the last one, which directly challenges the incremental upgrade path Tesla has championed for years. Waymo's head of AI foundations, Srikanth Thirumalai, wrote:
"Simply improving a driver-assist system (L2) for full autonomy is a false summit. True L4 maturity can only be safely achieved by a purpose-built system, validated on closed courses and hardened by the uncompromising experience of driving without a human in the car."
That's a clear strike at Tesla's FSD, which is built on top of a consumer ADAS known as Autopilot and relies on data from customer vehicles where drivers remain legally responsible. Waymo argues that no amount of simulated miles or supervised driving can replicate the safety feedback loop of truly driverless operation.
Lesson 1 states that cameras alone cannot deliver safe, full-scale autonomy. Waymo uses a combination of cameras, lidar, and radar for redundancy. Tesla, by contrast, has moved to a camera-only "Tesla Vision" system, eliminating radar and ultrasonic sensors from its vehicles. Waymo's co-CEO recently explained why this approach falls short, pointing out that cameras lack the depth perception and reliability of lidar in poor weather or low light.
Lesson 4 warns against pure end-to-end neural networks that convert raw camera input directly into vehicle control commands. "You can't build trust with a black box," Waymo says. Tesla has heavily promoted its end-to-end approach for FSD V12 and beyond, where a single neural network processes camera images and outputs driving actions. While this approach can produce impressive demos, Waymo argues it lacks the interpretability and safety validation needed for widespread deployment.
Waymo's confidence is backed by hard data. The company now averages over 500,000 paid, fully driverless rides each week across its operations in San Francisco, Phoenix, Los Angeles, and other cities. It has set a goal of one million weekly rides by the end of the year. In contrast, Tesla's robotaxi service, launched in Austin earlier this year, still uses human safety monitors. Tesla confirmed last month that it has accumulated only 380,000 unsupervised miles—a milestone Waymo achieves in a single day.
Critics and industry analysts have long questioned whether Tesla's strategy of iterating on FSD through customer vehicles can ever reach Level 4 safety. "The fundamental issue is that Tesla's training data comes from human drivers who are still in control," said Dr. Hannah Lee, a robotics researcher at Stanford. "The system never learns to react without a human backup, creating a blind spot for edge cases."
Waymo didn't need to name Tesla—everyone knows who the target is. The "false summit" metaphor is especially apt for anyone following Tesla's repeated missed deadlines. Elon Musk has promised "unsupervised" FSD by the end of 2025, a target he has moved numerous times before. Meanwhile, Waymo continues to rack up millions of miles in fully driverless operation, building a data moat that Tesla cannot easily replicate.
The key takeaway: Waymo's 10 lessons are not just academic. They reflect a fundamental divergence in approach. One company builds from the ground up for safety, while the other tries to retrofit a consumer feature. The road ahead will show which path is truly sustainable.
Industry observers have weighed in on the implication. John Krafcik, former Waymo CEO, recently noted on Twitter that "camera-only autonomy is a
