As Tesla gears up for the official launch of its steering wheel and pedal-less Cybercabs, Waymo is issuing a stark warning about Elon Musk’s approach autonomous driving.
Srikanth Thirumalai, Waymo’s VP of Onboard Software, doesn’t specifically call out Musk or Tesla in a blog post, published Wednesday, entitled “10 AI Lessons from Driving 200+ Million Fully Autonomous Miles.” But his intention is clear: Tesla’s system is insufficient to the task of fully autonomous driving.
“Cameras are incredible, but they aren’t enough,” Thirumalai writes. “For years, there’s been a debate over whether cameras alone could solve full autonomy. Now, after more than 200 million real-world miles, the data is clear: safe, fully autonomous operations at scale require more.” Waymo uses cameras, in addition to lidar and radar, to create “a rich, redundant world view that no single sensor can replicate,” he adds.
Musk, of course, would beg to differ. The occasional trillionaire has over the years made his distaste for lidar widely known, calling the laser-based sensor “a crutch” and predicting that any company relying on the powerful but expensive tech is “doomed.” The crux of his argument is that humans make driving decisions primarily through the use of their eyes, so why not autonomous vehicles?
In recent months, Tesla has doubled down on this position. In a recent earnings call, Tesla’s head of AI Ashok Elluswamy ripped into other companies who have adopted a multi-sensor approach. “Historically, the so-called experts have always claimed that you need lidars, radars, HD maps, and the entire kitchen sink to drive safely,” he said. “Here we show that such is not true. You can have safe, comfortable, and affordable autonomy with just cameras. This record should be a huge validation of Tesla’s entire AI approach.”
Of course, Tesla is mostly alone in this position. The vast majority of AV operators, from Waymo to Zoox to Motional, use lidar, in addition to radar and camera, as part of their on-board suite of sensors. They believe that autonomous vehicles need to be better than humans with regard to their perception systems, which requires a redundant multi-sensor approach. If one sensor fails, the vehicle wouldn’t be fully blinded as other sensors can help guide it to safety.
Tesla’s camera-only approach could shut it out of certain markets, depending on how certain policy debates turn out. New Jersey’s state legislature is currently considering a bill that would legalize robotaxis, but only if they include multiple sensors. Tesla would be banned under such a law.
Thirumalai goes on to outline other differences between Waymo and Tesla. While Musk thinks the use of high-detail maps to help chart an autonomous vehicle’s course is “a really bad idea,” Waymo believes that HD maps are a powerful “prior.” Waymo uses HD maps to “jump-start our validation process, so we can provide a fully autonomous service to riders from our first trip,” Thirumalai says. “As we drive, we treat our maps as another input — like our sensors, but acting as a mental memory.”
Musk believes autonomous vehicles need to be able to drive without a map; after all, roads change, lanes get repainted, construction zones appear, as do potholes. Musk believes that you need an instant analysis system that’s so good it gains little from maps — and that HD maps in particular can slow down the development of that necessary system.
But Waymo isn’t resting on its mapping laurels. The robotaxi company uses an AI-driven mapping system to ensure its maps are “continuously updated,” Thirumalai writes, “providing the vehicle with a reliable, high-fidelity reference to lean on during complex maneuvers.” In other words, mapping is useful, but it’s not an end-all-to-be-all.
Thirumalai goes on to describe some of Waymo’s other principles — closed-loop simulation can help identify more edge cases, Vision-Language Models are great reasoning tools — before concluding with his most direct swipe at Tesla yet: you can’t build a fully autonomous system on a Level 2 driver assist one.
The final lesson is that there is no substitute for actual fully autonomous miles. 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.
This, of course, gets at the heart of the Tesla proposal: that with enough data and enough miles and enough training, eventually its advanced driver assist systems, Autopilot and Full Self-Driving (Supervised), will evolve into Level 4 capable, fully autonomous platforms.
Today, Tesla’s FSD system is still classified as a Level 2 system, meaning drivers need to monitor the driving at all times and be ready to take control at a moment’s notice. If a crash occurs while FSD is being used, Tesla will always blame the driver, claiming that the company’s terms of service require constant driver monitoring. But that’s not how Musk and other executives talk about FSD. In fact, Tesla was forced to add “Supervised” to its marketing materials after regulators complained that the company was misleading customers.
So who’s right? Waymo or Tesla? Waymo operates its fully autonomous vehicles in 11 cities, conducting 500,000 paid trips a week. Tesla has its own robotaxis, some of which are supervised by human safety drivers, but an increasing number of which are unsupervised. And the company is on the cusp of launching its purpose-built Cybercab – though its unclear how widespread it will be at launch.
But after boasting about his company’s autonomous capabilities for years, Musk is still very clearly trailing behind Waymo. It’s unlikely he’s going to suddenly shift his tactics to embrace lidar, but he is undoubtedly aware that he is lagging behind Alphabet’s robotaxis. What he plans on doing about that will become increasingly clear in the months ahead.
