Amazon's self-driving unit just hit a major setback
Zoox's custom robotaxi lacks mandatory features like brake pedals and mirrors.
The Ride AI conference is coming up on April 2—barely two weeks away. If you’re involved in the autonomous vehicle industry—as an executive, engineer, investor, or otherwise—I’d love to see you in Los Angeles. Click here to get more details and purchase a ticket.
In the closing weeks of the Biden Administration, federal regulators wrote a report that was bad news for Zoox, Amazon’s self-driving venture. The Washington Post obtained a copy of the report last week.
Zoox uses a custom-designed vehicle that lacks standard driving controls. It has no steering wheel, pedals, or side mirrors. And that could be a problem because federal safety regulations effectively require road-legal vehicles to have many of these features.
Yet a custom-designed vehicle has always been a core part of Zoox’s pitch to investors (it was acquired by Amazon in 2020). So in recent years, the company has barreled ahead with its toaster-like custom robotaxi despite the legal challenges.
In 2022, Zoox asserted that its vehicles were compliant with federal regulations. The next year, the National Highway Traffic Safety Administration opened an investigation into Zoox’s compliance with these rules. And thanks to the Washington Post, we now know that in December 2024, NHTSA concluded that Zoox’s vehicles were not compliant.
Zoox continues to insist that its vehicles are legal under federal regulations.
“Our recent discussions with NHTSA are about mirrors, windshield wipers, a defroster, and a foot-activated brake pedal — equipment that makes sense for vehicles with human drivers, but not for the Zoox purpose-built robotaxi,” Zoox told the Washington Post. “Our purpose-built design means that the robotaxi can never be operated by a human driver, and our AI driver doesn’t rely on this equipment to view the world.”
But Zoox is now in an awkward position. The company was on the verge of launching commercial robotaxi services in the Las Vegas and San Francisco metro areas. Now it may not be able to do so legally.
NHTSA hasn’t formally recalled Zoox’s vehicles, but that could happen in the coming months. And the Washington Post says that officials in California are unlikely to allow local use of a vehicle that’s not compliant with federal regulations.
NHTSA’s statement to the Washington Post was non-commital. “We will continue to support transportation technology innovation while maintaining the safety of America’s roads,” the agency said.
In theory, Zoox could toss out their custom vehicle and instead add self-driving hardware to a conventional vehicle. This is the approach Waymo has taken, and it has worked well for the Alphabet subsidiary. The problem is that this would likely require years of expensive engineering work, and it’s not clear Zoox can afford to wait that long.
And this means that Zoox is highly dependent on the goodwill of the Trump administration. There are a few ways Trump could help the Amazon subsidiary:
NHTSA could re-write the rules that require cars to have brake pedals, side-view mirrors, and other driver-related safety features. Last November, Bloomberg reported that Trump’s team was getting ready to overhaul federal regulations to streamline the deployment of self-driving cars, so this process may already be under way. But even with a friendly administration, this process could take several years. NHTSA would need to draft new rules, get public input, revise the rules, publish them, and then fight off the lawsuits that would likely be filed by skeptics of self-driving technology.
Zoox could ask the Trump administration for a Zoox-specific exemption from the rules. Federal law allows NHTSA to grant such exemptions, but they are limited to 2,500 vehicles per year. That would be enough for large-scale testing (Waymo still has fewer than 1,000 vehicles active in its commercial fleet), but it could become constraining once Zoox wants to scale up its service nationwide.
A third option would be to ask the Trump administration to simply declare Zoox’s vehicles to be in compliance with existing rules. It wouldn’t be the first legally dubious waiver to come from this administration.
As I wrote last November, Tesla is in a similar bind. Tesla is hoping to begin manufacturing the Cybercab, which has no controls for human drivers, as early as next year. Even with pressure from Elon Musk, it will be difficult for NHTSA to finish a re-write—and resolve any resulting lawsuits—in less than two years. And Musk presumably wants to manufacture a lot more than 2,500 vehicles, so a narrow exemption wouldn’t be very helpful.
But if NHTSA blesses Zoox’s declaration that its vehicles already comply with federal regulations, that could create a precedent for Tesla to follow when it releases the first Cybercabs a couple of years from now.
Waymo has taken a more pragmatic strategy: it buys conventional vehicles and adds sensors to them. The strategy forces Waymo to ban passengers from the driver’s seat, reducing each vehicle’s capacity. But starting with compliant vehicles allows Waymo to entirely sidestep federal safety regulations.
I also wonder if it’s premature to remove steering wheels and pedals from self-driving vehicles. There were a number of incidents in 2023 when Waymo or Cruise vehicles got confused at the scenes of car crashes, fires, or other emergency situations. In an emergency, Waymo allows a police officer or firefighter to take manual control and manually drive a Waymo vehicle where it needs to go.
I expect this to become less necessary over time. But it doesn’t seem like such a bad idea to keep steering wheels in self-driving cars for at least the first few years.
I am also skeptical of the strategy to go straight to a purpose built AV for the same reason Tim mentions. It’s a difficult engineering problem to reduce vehicle retrieval events down to zero and I would imagine it’s expensive to rescue a stalled out Zoox.
Another strategy they could pursue would be to convert their test fleet of Toyota Highlanders into commercial vehicles and perhaps manufacture more. Probably isn’t great unit economics but it could maybe get them on the road with paying customers faster
The same applies to Tesla Cybercab, it does not have any the conventional gear.
I estimate it will be approve by 2H 2025.