Elon Musk’s heterodox robotaxi philosophy gets put to the test

Believe it or not, the Tesla Cybercab is coming.

Nearly two years after Elon Musk first unveiled the gold-tinted, gull-wing door sporting, steering wheel-free two-seaters as the future of Tesla’s autonomy efforts, the company is finally putting them into operation as part of its robotaxi service in Austin, Texas. Public rides are here, and as Musk fans will assert, the era of Tesla’s robotaxi dominance will begin.

Or will it? Musk has bet his company — and his trillion-dollar pay package — on a stripped down, barebones autonomous driving system that eschews multiple sensors used by all of his rivals in favor of a camera-only approach. The Cybercab, as a purpose-built autonomous vehicle without traditional controls, embodies this bet. Musk believes he can forge his own path to robotaxi success — and the Cybercab is his gilded vehicle of choice. We’ll soon know if he’s in for a crash.

The Cybercab will be using a modified version of Tesla’s Full Self-Driving (FSD). For the company’s millions of vehicle owners, FSD remains a Level 2 driver assist feature that requires human supervision. In the Cybercab, it will be monitored by remote operators who theoretically can step in if there’s a problem.

FSD and Autopilot have been linked to dozens of fatalities and thousands of crashes. Using only cameras and an AI-powered neural net, the Cybercab will need to navigate a complex, unpredictable environment. Without conventional controls, there will be no human fallback in case the autonomous system fails. And after years of blaming crashes on their drivers, Tesla will now bear full legal responsibility if something goes awry.

For years, Musk has coasted on brash promises for an autonomous future. With tomorrow’s Cybercab launch, those promises are going to face harsh scrutiny that’s not easily overshadowed by rows of gleaming robot cars hitting the road.

For over seven years, Musk has thumbed his nose at the conventional wisdom around self-driving cars — specifically that more sensors equals more safety. At Tesla’s inaugural “Autonomy Day” event for investors in 2019, he laid out his vision for a sleek, simplified autonomous driving system that avoided the protruding sensors of the Waymo cars of the time. The top target for his ire: lidar.

The light beam sensor that practically everyone views as an essential ingredient for self-driving cars is “a fool’s errand,” Musk said at the event in 2019. And “anyone relying on lidar is doomed. Doomed. Expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices… you’ll see.” The crux of his argument is that humans use their own eyes to make driving decisions, so why not autonomous vehicles?

Tesla once relied on radar, in addition to cameras, as part of its perception system, only for Musk to do away with the sensor in 2022. The company’s own engineers tried to persuade Musk to keep radar. They worried that without radar, Tesla’s self-driving system would be susceptible to basic perception errors if cameras were obscured by rain or sunlight — which could lead to crashes. They were rebuffed.

The reason everyone else uses lidar underscores the limitations of Musk’s camera-only approach. Cameras only provide flat, two-dimensional images, requiring Tesla to use computer vision and neural networks to approximate depth and distance. A camera-only system leaves itself vulnerable to sunlight glare, while a multi-sensor system has additional data in case its perception is blinded. And cameras struggle in adverse weather, like snow or fog, while lidar and radar can often pierce through these conditions.

Few of Musk’s predictions have come true, including his condemnation of lidar. Waymo, which uses lidar in addition to cameras and radar, is operating a fleet of 4,000 fully driverless vehicles in up to 14 cities across the US. This is the largest, more fully realized deployment of self-driving cars in the US — and possibly the world. Lidar hasn’t slowed Waymo’s progress; in fact, its engineers strongly believe the sensor helps accelerate it.

In a recent talk at Y Combinator’s Startup School, Waymo co-CEO Dmitri Dolgov agreed that cameras have gotten good enough to closely approximate human driving. But if the goal is to exceed humans and build a system that truly excels at safety, then cameras alone aren’t good enough. If you’re targeting full autonomy and superhuman performance, Dolgov said, “you find that weak sensing just leads to a safety curve that flattens out way too early.”

But Musk claims that a camera-only system is more affordable to mass produce, and thus will enable Tesla to deploy Cybercabs more quickly than Waymo. There’s no arguing that the Cybercab is cheaper to produce than Waymo’s sensor-laden robotaxis, but the lack of lidar isn’t the only reason. Tesla’s also saving money by leaving out traditional controls like steering wheels, pedals, and mirrors. The company argues this helps create more passenger space, but the hope is that it will also make the vehicle more profitable to manufacture.

At Tesla’s “We, Robot” event in 2024, Musk promised that people could buy the Cybercab for a price “less than $30,000.” Someone could theoretically buy a small fleet of Cybercabs to operate as a bespoke robotaxi service, he said.

But to sell the Cybercab to regular people, Tesla will need an exemption from federal safety rules that require vehicles to have steering wheels and pedals. If it tries to self-certify, it may run into the same trouble that tripped up Zoox for several years. (The Amazon-owned company eventually got its exemption, with heavy conditions.) Tesla will also need a permit from the California DMV to operate fully driverless vehicles on public roads. And in order to do that, it will need to demonstrate that its vehicles can operate safely — which so far, the company has not done.

What happens when a Cybercab is disabled or breaks down? Other AV operators like Waymo have procedures in place for remote operators to try to move the vehicle out of the way. And if all else fails, they dispatch crews of technicians who come out and manually drive the vehicle away. For a vehicle without steering wheel or pedals, teleoperation becomes the primary failsafe. That explains why Tesla has taken to installing Starlink connections in all its Cybercab — a convenient, if not double-dealing, solution for Musk.

The Cybercab may find itself stuck in Austin while Tesla works out these technological and regulatory hurdles. The Texas capital is where Tesla’s robotaxi service features the most unsupervised vehicles. It’s where Tesla feels most confident in its driverless capabilities.

There is no federal policy dictating driverless car deployments in the US. Legislation that could open the gates to more Cybercab-type vehicles has been stalled in Congress for years, with no sign of budging. That leaves Tesla to fight for permission state by state, or sometimes city by city. That’s expensive, not to mention time consuming.

Tesla’s current Robotaxi business has been extraordinarily difficult to nail down. The fleet’s size and supervised-vs-unsupervised status has vacillated wildly for months. Sometimes there are dozens of robotaxis available, depending on the city, but more often less than a handful are actually doing paid passenger trips. In Tesla’s most mature market of Austin, where it has been running robotaxis since June 2025, the company has approximately 110 Model Y vehicles in operation, according to Robotaxi Tracker.

With the addition of the Cybercab, we should expect those numbers to grow dramatically. But as we’ve seen in the past, just because the vehicles are on the road doesn’t mean Tesla is necessarily using them for passenger trips.

If you ask Tesla, its safety record is unimpeachable. The company’s dashboard of 2 billion miles claims that Tesla vehicles that use the supervised version of Full Self-Driving are involved in 40 to 90 percent fewer crashes than manually driven vehicles.

But its actions also speak louder than words. Despite marketing materials that suggest the system is more autonomous than it actually is, FSD remains just a Level 2 feature that requires constant human driver supervision. The driver, not Tesla, is legally responsible when something goes wrong.

Meanwhile, fatal crashes involving FSD continue to rack up. Under a 2021 standing general order by the National Highway Traffic Safety Administration, Tesla is required to report crashes where its Level 2 driver assist system was engaged within 30 seconds of impact and someone was hurt, killed, or the vehicle was towed.

According to Electrek, Tesla accounts for roughly 85 percent of the entire industry’s reports — nearly 4,000 crashes — thanks to its status as the automaker with the most ADAS equipped vehicles. And independent trackers like Tesla Deaths record roughly 65 fatalities involving Tesla Autopilot or FSD systems between 2013 and 2025. (There could be more; Tesla redacts most of the details in its reports to NHTSA.)

Tesla likes to trumpet the good stats, while hiding the bad ones. Its vehicles enjoy some of the highest safety ratings in the country, but its autonomous driving statistics remain a black box. Unlike Waymo, Tesla hasn’t submitted its safety data to any independent group for validation. Its own AI trainers have told Reuters they don’t trust its safety stats. And the company is currently under investigation by NHTSA for how it reports these crashes.

Autonomous driving is all about trust, and Tesla currently doesn’t enjoy a lot of it. A recent survey found that most people think the company is misleading its customers by using terms like “Autopilot” and “Full Self-Driving.” They also don’t think the company is doing enough to ensure its drivers follow the rules — and a quick survey of social media backs up this belief. And they don’t want it to deploy fully driverless taxis.

The Cybercab is sure to scramble some of these perceptions. With its future at stake, Tesla is barrelling ahead with fully driverless vehicles. A well-managed launch will no doubt be enough key-jangling to impress investors. But the public may be harder to convince.