Last week Waymo stated that fully autonomous vehicles cannot be safely operated using a single sensor type or solely "pure end-to-end" AI systems. The company did not name a rival directly, but the comments were understood as targeting Tesla.
The assertions appeared in a blog post and in an interview with Axios one week before Tesla is expected to formally unveil its two-seater Cybercab on September 3. The technical claims ignited a social media debate that lasted through the weekend.
What Waymo says
Srikanth Thirumalai, Waymo’s vice president overseeing driving software, wrote in the company blog that cameras are excellent but insufficient on their own. Citing more than 200 million real-world miles of data, Waymo says the evidence is clear: safe, fully autonomous operations at scale require more. By combining inputs from cameras, lidar, and radar, the Waymo Driver produces a rich, redundant world view that no single sensor can replicate.
Thirumalai told Axios that a pure end-to-end neural architecture that "takes in raw pixels and directly outputs steering commands" — a description aligned with Tesla’s autonomy approach — carries the risk of black-box failures. He noted that even the best AI models with trillions of parameters can still hallucinate, and in physical systems there is no simple software "click reboot" to mitigate consequences.
Waymo’s scale and operational experience
Waymo has long integrated a mix of sensors into vehicles built by other manufacturers. That conservative technological approach has allowed the company to scale to roughly 4,000 robotaxis across 14 U.S. cities and to provide about 500,000 paid trips per week. Operating at that scale, Waymo continues to uncover operational challenges such as handling severe weather, avoiding emergencies, and operating safely around school zones.
Tesla’s camera-first, end-to-end strategy
Elon Musk has long derided lidar as a "crutch," and Tesla has concentrated on building full autonomy using only cameras and AI. The company has spent years developing the Cybercab, a purpose-built two-seater gold sedan designed from the outset to be autonomous. The Cybercab lacks a steering wheel and pedals, uses a relatively small battery, and, according to a recent filing, Tesla aims to build more than 125,000 of them annually.
Tesla still must prove its self-driving software can achieve full autonomy. The company has fallen behind earlier timelines: Musk once promised there would be 1 million robotaxis on the road in 2020, a projection that did not materialize. Over the past year Tesla has piloted a small Robotaxi network in a handful of Texas and Florida cities using modified Model Y SUVs. Those trials have remained limited in scope, with Tesla saying it prioritized safety over scale; in recent weeks the company removed safety monitors from a majority of those cars.
Why the Cybercab reveal matters
The Cybercab’s September 3 unveiling is a potential turning point: if Tesla can show that a camera- and end-to-end AI–first approach scales safely, it would be a major validation for the company’s strategy and engineering teams. Such a result could give Tesla a competitive edge on costs because the company manufactures its own vehicles and plans to use far fewer sensors.
Cost and business competition
Beyond technical disagreements, cost is a central battleground. Waymo’s multi-sensor approach is inherently more expensive: it employs more sensors and retrofits vehicles produced by others, which involves purchasing the vehicles and sometimes paying import taxes before outfitting them with autonomy technology (as in the Zeekr-made Ojai example). Tesla builds its own cars and is betting that cameras alone will suffice for navigation. If Tesla succeeds, it could outcompete Waymo or other players such as Uber on price.
Conclusion
The dispute is both technical and commercial: the robotaxi market could be worth hundreds of billions of dollars. Waymo points to its hundreds of millions of real-world miles and to redundancy from multiple sensor types as key to large-scale safety; Tesla is betting on a lower-cost, software-first route. The Cybercab reveal on September 3 and the subsequent evidence of operational performance will be critical in showing which approach is likely to scale safely and economically.
