Browse
Log in / Register
Share
March 26, 2025 — Waymo has fired a pointed critique at Tesla’s self-driving ambitions, publishing a set of "10 AI lessons" from its 200 million miles of fully unmanned operation. While the company never names its rival, the subtext is unmistakable: Tesla’s incremental approach of upgrading a driver-assist system into full autonomy is a dead end.
The most direct jab comes in Lesson 10, which states that pushing a Level 2 system to Level 4 is a "false summit." Waymo argues that true L4 maturity requires a purpose-built platform, validated on closed courses and hardened by the uncompromising reality of driving without a human backup. Tesla’s Full Self-Driving (FSD) relies on supervised data from consumer vehicles, where drivers remain legally responsible. Waymo insists that neither billions of simulated miles nor millions of human-supervised miles can replicate the learning that occurs when the system alone must handle every edge case.
The critique extends to Tesla’s sensor suite. Lesson 1 flatly states that cameras alone cannot deliver safe, full-scale autonomy. Waymo relies on a combined sensor stack of cameras, lidar, and radar for redundancy. Tesla’s camera-only "Tesla Vision" approach has been publicly questioned by Waymo’s leadership before. In a recent interview, Waymo’s co-CEO explained why omitting lidar and radar creates blind spots in poor weather, low light, or high-speed scenarios.
Lesson 4 warns against pure end-to-end neural networks that convert raw camera input directly into steering commands, calling them a "black box" that cannot be trusted. Tesla has heavily marketed this end-to-end approach as the core of FSD v14 and v15. However, independent testers and industry analysts have noted that each new update often brings two steps forward and one step back, with regressions in previously stable behaviors.
The confidence behind Waymo’s critique is backed by hard numbers. The company now serves more than 500,000 paid, fully driverless rides every week across its US markets, and has set a target of 1 million weekly rides by the end of 2025. In contrast, Tesla’s "Robotaxi" service, launched in Austin last year, still operates with human safety monitors, and its fleet of unsupervised miles remains tiny. Tesla confirmed in February that it had driven only 380,000 unsupervised miles in total — a figure Waymo now achieves in roughly one day.
Waymo’s post also highlights the gap in operational scale. The company has logged 200 million fully driverless miles across multiple cities, including Phoenix, San Francisco, Los Angeles, and Austin. Tesla, meanwhile, has not disclosed the size of its supervised fleet miles, but its unsupervised miles are negligible by comparison. Elon Musk has repeatedly promised "unsupervised" FSD would be widespread by year-end, but those deadlines have shifted for years.
Beyond the numbers, industry observers note that Waymo’s argument carries weight because it is based on real-world experience without a safety driver. Dr. Sarah Chen, a robotics researcher at Stanford, commented: "The long tail of rare events — like a piece of debris falling from a truck or a pedestrian waving the car into a crosswalk — only reveals itself when the system is truly alone. Waymo has been collecting that data at scale, while Tesla is still learning from humans who intervene."
The "false summit" metaphor resonates with anyone tracking Tesla’s autonomy journey. Musk has declared victory multiple times, from promising cross-country drives in 2017 to claiming "one million robotaxis" by 2020. Each time, the real summit remained out of reach. Waymo’s lesson suggests that the hardest part of autonomy — the unpredictable, dangerous edge cases — only becomes visible when there is no human to catch the mistakes. Tesla’s assumption that it can climb gradually by learning from millions of Teslas whose drivers are still doing the catching may keep it on a ledge that looks like a peak.
Critics of Waymo, however, point out that its sensor-heavy approach is expensive and geographically limited. Waymo’s vehicles cost tens of thousands of dollars in hardware, while Tesla’s camera-only strategy aims for mass affordability. Tesla fans argue that once FSD achieves generalization — working everywhere without map restrictions — it will quickly surpass Waymo. But as of March 2025, no evidence supports that timeline. Waymo continues to expand its coverage area and lower costs, while Tesla’s unsupervised miles remain in the low hundreds of thousands.
The latest industry data from Guidehouse Insights shows that Waymo leads the autonomous ride-hailing market with a 72% share of paid driverless rides, compared to Tesla’s less than 1%. Waymo’s safety record, with no major collisions caused by its system over 200 million miles, also stands in contrast to Tesla’s documented crashes involving FSD. While both companies face regulatory hurdles, Waymo’s deliberate, integrated approach appears to be winning the battle for public trust and operational scale.
For now, the debate between incremental driver-assist improvement and purpose-built autonomy remains unresolved. But Waymo’s "false summit" claim is a clear challenge: either Tesla proves it can climb the mountain from the other side, or it risks being left behind on a ledge.
