Tesla’s Autopilot is supposed to deliver full self-driving, so why does it feel stuck in the past?

Photo by Amelia Holowaty Krales / The Verge
A year ago, Tesla CEO Elon Musk stunned the automotive world by announcing that henceforth, all of his company’s vehicles would be shipped with the hardware necessary for “full self-driving.” By 2019, Tesla drivers would be able to sleep in their cars, he suggested. Musk one-upped that a few months later, vowing to demonstrate a completely autonomous, cross-country trip in a Tesla by the end of 2017.

But since then, the company has fallen behind in updating its flagship semi-autonomous driving system, Autopilot, irking some among its legions of obsessed fans, and raising questions about Tesla’s ability to deliver on the promise of a fully self-driving car. In an August earnings call, Musk admitted the cross-country trip may be delayed. “It is certainly possible that I will have egg on my face on that front,” he said.

IT’S BEEN NEARLY FIVE MONTHS SINCE THE LAST AUTOPILOT UPDATE
Since Musk’s announcement, Autopilot 2.0’s over-the-air software updates have been infrequent and scattered, Tesla owners say. For the first half of 2017, there were updates generally every three weeks, such as the ability to use Autosteer at speeds up to 90 mph on the highway and 35 mph on local roads. It’s been nearly five months since the last significant Autopilot update, and many of the features from the previous version of the advanced driver assist system (ADAS) are still missing.

Perks such as rain-sensing wipers have yet to be added, but more operational tasks, such as the ability to read speed limit signs and differentiate between vehicle types on the road, are also absent. Other features, like automated lane changes on the highway, better steering on tight roads, and auto-summon, were promised by the end of 2016, but have yet to be rolled out. Tesla has since removed the date from its website.

In his announcement a year ago, Musk showed video of a Tesla leaving a garage, driving across town, and finding its own parking spot — all autonomously. While there was someone behind the wheel during the demonstration, the implication was clear: this was a car that will drive itself, and sooner than you think possible.

tesla autopilot cluster
But “Enhanced Autopilot,” as it was called, wasn’t complete. As a result, Tesla vehicles built since October 2016 have many fewer safety and convenience features enabled than in older models. Musk acknowledged as much during his announcement, noting that “it will probably sometime next year” when Tesla cars with the newer hardware suite caught up to the previous versions in functionality. In early 2017, Musk tweeted that most of those features would be rolling out soon, but months later, many are still missing.

Given the lag in Autopilot updates, some Tesla owners now question whether the company can deliver on what’s ultimately been promised: full self-driving. “It just leaves you with no faith that they are anywhere on that path,” said Ian Jordan, an electrical engineer from Seattle who owns a Model S. “Today they can’t reliably detect a speed limit sign, so it just seems like an enormous gap.”

“IT JUST LEAVES YOU WITH NO FAITH THAT THEY ARE ANYWHERE ON THAT PATH.”
Jordan, who is also an investor in Tesla, said his Model S would routinely “freak out and dive for oncoming traffic” when Autopilot becomes confused by certain intersections in his neighborhood. “Don’t know what it is,” he said. “Every time.” Unlike most other major driver assistance systems, Tesla owners can decide to turn on Autopilot regardless of where they are driving; other ADAS, like Cadillac’s Super Cruise, restrict use to divided highways. In order to gather the maximum amount of data, it would behoove Tesla to encourage customers to use Autopilot as much as possible in safe conditions.

That’s probably because the company’s ambitions to offer full self-driving depends on its drivers using Autopilot all the time. Before Teslas can start driving autonomously, the company needs to collect a lot of data to prove to customers (and regulators) that the technology is safe and reliable. So, its cars run Autopilot in “shadow mode” in order for Tesla to gather statistical data to show false positives and false negatives of the software. In shadow mode, the car isn’t taking any action, but it registers when it would have taken action.

It also casts a spotlight on the major difference between Tesla and practically every other automaker and tech company pursuing autonomous driving: Tesla is breaking with its competitors in the self-driving space by declining to include LIDAR sensors under the assumption the equipment will be too expensive for personal ownership. Musk has said previously that LIDAR sensors “don’t make sense in a car context.” Instead, the cars rely on eight cameras for 360-degree vision, as well as 12 ultrasonic sensors and forward-facing radar.

Photo: Tesla
Most companies assume self-driving cars will be shared and used in restricted, geofenced areas, like downtowns or college campuses. Tesla is banking on autonomous cars being cheap enough (and reliable enough) for people to own. But to get there, it needs to convince people that Autopilot is a better, safer way to drive. As a highly advanced version of adaptive cruise control, it has a lot going for it. But there are internal and external forces that suggest things aren’t going as planned.

AS A HIGHLY ADVANCED VERSION OF ADAPTIVE CRUISE CONTROL, IT HAS A LOT GOING FOR IT.
Internally, Tesla has experienced a lot of upheaval as it relates to Autopilot. The company’s team of engineers working on Autopilot reacted with dismay to Musk’s announcement last year, believing Autopilot lacked the capability to deliver full self-driving, according to the Wall Street Journal. In addition, CNN reported that Musk reportedly brushed aside certain concerns as negligible compared to Autopilot's overall lifesaving potential, but employees who worked on Autopilot struggled to make the same leap.

A major cause of this conflict has apparently been the way Musk chose to market Autopilot. The decision to refer to Autopilot as a “full self-driving” solution — language that makes multiple appearances on the company’s website, especially during the process of ordering a car — was the spark for multiple departures, including Sterling Anderson, who was in charge of the Autopilot team during last year’s announcement, the Journal reported. In June, Chris Lattner left his post running software for Tesla’s Autopilot division after just six months.

Externally, federal regulators were still sifting through the wreckage (both literal and digital) from the 2016 crash that killed a man driving a Tesla Model S while using Autopilot. The National Traffic Safety Board recently concluded that Autopilot contributed to the crash and recommended that all automakers, Tesla included, consider more serious limitations to where its semi-autonomous systems can be used and how they market such products.

There are other factors likely contributing to Tesla’s infrequent Autopilot updates. The company is in the midst of what Musk has called “production hell” with the Model 3. Customers and investors are watching intently to see if Tesla can deliver on its promise of a mass-market, battery-electric car. And the effects of Tesla’s decision to part ways with the supplier Mobileye in 2016 are still being processed. With the introduction of the second-generation Autopilot in October 2016, Tesla eliminated Mobileye’s computer vision technology powered by its EyeQ3 chip. The company replaced that with its own computer vision system called “Tesla Vision” powered by Nvidia’s Drive PX2 onboard computer.

Autopilot 2.0 is “still missing a lot of the functionality of the original version that relied more heavily on Mobileye from their vision system,” said Sam Abuelsamid, a senior research analyst at Navigant. “And from everything that we’ve seen, it does not sound like they have caught up to where Mobileye was a year and a half ago.”

“IT DOES NOT SOUND LIKE THEY HAVE CAUGHT UP TO WHERE MOBILEYE WAS A YEAR AND A HALF AGO.”
Tesla creates the impression that these are changes that can be fixed with over-the-air software updates, an area where the company has proven to be more nimble and willing to take risks than legacy OEMs — and a feature that customers rate highly. But autonomous driving, especially Level 4 and 5 that Musk has promised, requires a very specific hardware suite in order to be safe and reliable. Recently, it was reported that Tesla quietly developed a new hardware suite with more computing power to help achieve its goal of full automation, which would seem to contradict Musk’s October 2016 vow that all vehicles would be shipped with “full self-driving capabilities.”

Tesla argues that customer safety undergirds all its decisions, and that it prioritizes features that would have the most immediate impact on that. Nonetheless, a spokesperson acknowledged that its Autopilot updates have been few and far between, and that the company was working to correct that.

"For our customers who opted to purchase our Enhanced Autopilot and Full Self-Driving packages, we’re working super hard on new features and rapid progress is definitely being made,” a spokesperson said in a statement to The Verge. “As is the case with any new technology, validation of first-of-its-kind technology is subject to a high number of variables which can impact product readiness timelines significantly. While it's taken longer than we originally expected to roll out all Enhanced Autopilot features, the product already provides significant assistance to drivers, and this year our platform was re-engineered to facilitate improved data collection, mapping, and fleet learning – important foundational changes which we expect to enable more dramatic improvements for our customers over time. With safety at the core of everything we do and every decision we make, we will be rolling out advanced new features as soon as they’re validated, and we’re grateful for everyone’s patience.”

“WE’RE WORKING SUPER HARD ON NEW FEATURES.”
But as Tesla expands to the mass market, that attitude may not be enough. “The customers they’ve had up till this point are clearly within the early adopter group,” Abuelsamid said. “They are willing to cut Tesla much more slack than I think a typical customer would be, in terms of both functionality and quality... Tesla really needs to get all that sorted out if they’re going to become a mainstream brand.”

Any blowback from customers and critics so far has been mild. The company continues to grow sales of its Model S and Model X amid a soft luxury car market, and Tesla’s stock continues to climb. And despite serious questions about the company’s ability to deliver the promised number of Model 3s, Tesla’s brand remains relatively untarnished. Fans of the taboo-shattering company seem to have internalized the fact that Musk’s mouth can sometimes outpace his company’s ability to deliver. If anything, it makes them love him more — even if it makes him seem like a buffoonish character from HBO’s Silicon Valley.

“I want to believe he's Henry Ford,” one commenter on Reddit wrote in a recent post about the lack of Autopilot updates, “but it's really hard to shake the feeling he's Gavin Belson.”