Three United States senators on Monday proposed legislation that would require all new cars in the United States to have driver-monitoring systems within six years. Two of the legislation’s sponsors—Ed Markey (D-MA) and Richard Blumenthal (D-CT)—recently sent a letter to federal regulators expressing concern about last week’s fatal Tesla crash in Texas.
It’s not clear how a 2019 Tesla Model S wound up crashing into a tree at high speed in a residential neighborhood outside Houston. Police reported that neither of the vehicle’s two passengers was in the driver’s seat: one was in the front passenger seat, while the other was sitting in a rear seat.
The crash has drawn more attention to the long-running debate over adding driver-monitoring technology to cars. A few carmakers have already adopted robust driver-monitoring technology. Cadillac’s Super Cruise driver-assistance technology, for example, uses a driver-facing camera to verify that the driver’s eyes are focused on the road. Drivers can take their hands off the wheel while Super Cruise is active. But if they stop looking at the road ahead, Super Cruise will warn them and eventually disengage.
Tesla vehicles use a more rudimentary technique to measure driver engagement: a torque sensor on the steering wheel. Unfortunately, hands on the wheel are no guarantee that a driver is paying attention. Also, torque sensors are easily defeated by hanging a weight from the steering wheel.
Legislation would mandate driver monitoring by 2027
Safety advocates have argued that any car with an advanced driver-assistance system (ADAS) should have a driver-monitoring system. That’s to combat the risk that drivers will start to trust their ADAS so much that they stop paying attention to the road. And many argue that a Tesla-style torque sensor isn’t good enough. Drivers’ failure to focus on the road likely contributed to three Autopilot deaths over the last five years, as well as a death caused by a prototype Uber self-driving vehicle in 2018.
And of course, distracted driving is a problem that extends well beyond Tesla or any manufacturer’s ADAS. The National Highway Traffic Safety Administration estimates that distracted driving is responsible for more than 3,000 deaths every year.
The Markey/Blumenthal bill, also co-sponsored by Sen. Amy Klobuchar (D-MN), would require the Secretary of Transportation to draft rules requiring all vehicles to have driver-monitoring technology installed. The final version of the rule would have to be ready within four years, and it would give automakers another two years to comply. If the legislation passed this year, every new car would need to adopt some version of the technology by 2027.
All three senators are members of the House Commerce, Science and Transportation Committee, which is holding a hearing on Tuesday about the future of automotive safety.
Requiring DMS for all vehicles, not just ADAS ones, might seem like a dramatic step. However, ADAS will only become more popular and more capable over the next six years. So many, perhaps even most, vehicles are likely to have some kind of ADAS by 2027. Indeed, cars in 2027 will likely handle even more of the driving task than today’s best systems. That will make it even more challenging for drivers to pay attention while their car does most of the driving.
European officials have been more aggressive than their American counterparts about requiring driver monitoring. Colin Barnden, an auto industry expert based in the United Kingdom, told Ars that recently enacted European rules require driver-monitoring systems in all new car models introduced from mid-2024 onwards. European carmakers get a two-year grace period to continue selling car models approved before the deadline, but every new car sold would need a DMS starting in mid-2026.
Separately, Barnden said that Euro NCAP, a government-backed group that rates cars for safety, would require cars to have a driver-monitoring system in order to earn a five-star safety rating starting in 2023 or 2024.
The self-driving technology industry is in a strange state right now. A number of companies have been pouring millions of dollars into self-driving technology for years, and many of them have prototype self-driving vehicles that seem to work.
Yet I know of only one company—Waymo—that has launched a fully driverless commercial taxi service. And I only know of one company—Nuro—that’s running a driverless commercial delivery service on public roads. You’d expect these companies to be capitalizing on their early leads by expanding rapidly, but neither seems to be doing that.
Meanwhile, several other players, including Cruise and Mobileye, say they’re planning to launch large-scale commercial services by 2023. But plenty of self-driving companies have blown past self-imposed launch deadlines in the past, so it’s not clear if that will actually happen.
In short, predicting what the next couple of years will bring is a challenge. So rather than offering a single prediction, here are eight: I’ve broken down the future into eight possible scenarios, each with a rough probability. Hopefully, breaking things down this way offers a good overview of the many different strategies being pursued by self-driving companies today. A decade from now, we’ll be able to look back and say which companies or approaches were on the right track. For now, we can only guess.
1. Waymo wins (20 percent)
Waymo has been viewed as the self-driving industry’s technology leader ever since it started as the Google self-driving car program more than a decade ago.
In the optimistic scenario, Waymo will maintain and expand its current lead. It will grow its current taxi service from one corner of the Phoenix metro area to all of Greater Phoenix, then steadily expand to other metro areas. Running the largest driverless taxi service could give Waymo access to more real-world driving data and operational experience than any other company has, which could allow it to further improve its software and maintain its lead.
So why do I only give Waymo a 20 percent chance? While Waymo still seems to be the technology leader, it hasn’t capitalized on its lead as well as many people—apparently including Waymo’s own leadership—expected a few years ago.
In 2018, Waymo announced deals to purchase “up to” 82,000 vehicles for use in its taxi fleet, suggesting the company thought it was on the verge of large-scale commercial launch. Yet today its fleet still numbers in the hundreds of vehicles.
I don’t know why Waymo is moving slowly. Maybe its software has become excessively optimized for suburban Phoenix. Maybe its hardware or back-end support costs are too high to operate profitably. Maybe there are lingering safety or reliability concerns that Waymo wants to squash before expanding in a big way.
But whatever the issue, it may not go away any time soon. Which could leave an opening for other companies.
2. Another robotaxi company wins (25 percent)
Plenty of other companies are pursuing the same basic strategy as Waymo—building and operating a robotaxi fleet. These include:
Motional (owned by Hyundai and auto-parts supplier Aptiv).
Zoox (a startup that was recently acquired by Amazon)
Aurora (a startup that recently acquired Uber’s self-driving program)
If Waymo falters, I think it’s most likely to be on business execution: Waymo continues to have industry-leading technology but fails to expand rapidly enough to take full advantage of it. Running a taxi service with a few hundred vehicles in one metro area (as Waymo is doing now) is a very different proposition from running a taxi service with hundreds of thousands of vehicles in dozens of cities.
Automaker-backed companies like Cruise, Argo, and Motional might have a greater ability to rapidly scale up production of self-driving vehicles. Amazon obviously has a lot of experience with large-scale logistical problems. And Aurora has a close relationship with Uber, which might provide Aurora with preferential access to its ride-hailing network.
3. Tesla (and Comma.ai) wins (5 percent)
This will make Tesla fans mad, but I think it’s true: Tesla is a long shot.
The bullish case for Tesla is that it has access to a vast trove of real-world driving data harvested from customers’ vehicles. If you think limited training data is a major bottleneck for improving self-driving algorithms, then this might be a significant advantage. Tesla CEO Elon Musk also has a bigger appetite for risk than most of the other companies working on self-driving technology. Musk’s willingness to put unproven technology on public roads may accelerate Tesla’s progress even as it creates a greater risk of fatal accidents.
On the other hand, Tesla has significant disadvantages. The company’s business model—selling cars to end users—puts lidar sensors and high-density maps financially out of reach. Elon Musk has tried to spin this as a positive, calling lidar a “crutch.” But the fact remains that almost every other company is using lidar and HD maps because it believes they are helpful.
More fundamentally, it’s hard to watch videos of Tesla’s software in action and conclude that Tesla is in a leading position—or even that it is catching up to the leaders. Tesla’s unfortunately named “full self-driving beta” software routinely flubs scenarios that Waymo’s cars have been able to handle for years.
If I’m wrong and Tesla’s strategy does succeed, that would be very good news for Comma.ai, a self-driving startup founded by legendary hacker George Hotz. Comma is building an open source self-driving system designed to run on a smartphone. Comma’s strategy is to enable early adopters to modify their own cars to take steering inputs from Comma’s smartphone-based software—and then use the data harvested from those early customers to further improve the software in much the same way as Tesla. Like Tesla, Comma has eschewed lidar, arguing that it can achieve adequate performance with smartphone-grade cameras.
Hotz’s ultimate goal is for Comma to be the Android to Tesla’s Apple. That is, if Tesla emerges as a clear leader in self-driving technology, other automakers will need to license their own self-driving technology to compete with Tesla. Hotz hopes that Comma’s software will become an industry standard among automakers, much as Android is an industry standard for smartphones not made by Apple.
4. Mobileye (and its partners) wins (10 percent)
Mobileye, an Israeli company acquired by Intel in 2017, has some big advantages. As the leading provider of advanced driver-assistance systems (ADAS), Mobileye has relationships with a bunch of automakers. The company has leveraged those relationships to get data from numerous customer cars, giving Mobileye access to a king’s ransom in real-world road data. Mobileye’s dominant position in the ADAS market also gives it a deep bench of engineering talent.
Like Tesla (and unlike Waymo and some other companies in the space), Mobileye believes that full autonomy can be achieved through incremental improvements to ADAS. Mobileye is hoping to gradually deliver better and better ADAS until, at some point, customers can take their hands off the wheel and take a nap in the driver’s seat.
At the same time, Mobileye isn’t as hemmed in by financial constraints as Tesla. With Intel as its parent company, Mobileye has the resources to explore multiple technological approaches simultaneously. Like Waymo, Mobileye is experimenting with lidar and testing fully self-driving technology on public streets with safety drivers behind the wheel.
Mobileye’s plan is to license its chips, sensors, and software to a range of customers. So if Mobileye’s technology works, it could allow smaller companies to ride Mobileye’s coattails. The delivery startup Udelv, for example, recently announced plans to use Mobileye’s technology in at least 35,000 delivery robots between 2023 and 2028. Mobileye is hoping to have dozens of customers like Udelv in a few years.
An industry favorite?
A Mobileye victory would also be pretty good news for other automakers, because it would mean they’re unlikely to get disrupted by Waymo, Tesla, or other upstarts. Mobileye’s main business is licensing technology to automakers, so if Mobileye develops a leading self-driving stack, carmakers should be able to easily integrate it into their existing product lines.
Mobileye has released some impressive videos of its technology in action. However, it’s hard to judge self-driving technology based on a few videos. Lots of companies have released impressive videos, but most of them haven’t been confident enough in their technology to launch driverless commercial services.
There’s also a risk that Mobileye’s business model—licensing its technology to a wide variety of customers, from automakers to companies running delivery fleets—could hamper the company’s progress. Developing self-driving technology for one specific application is hard enough. Mobileye is hoping to develop a single technology stack that will perform well in a wide range of applications. That might not be realistic.
5. China wins (15 percent)
There’s a lot of international collaboration on self-driving projects, with overseas companies like Honda and Volkswagen investing in American self-driving projects and Intel buying the Israeli Mobileye. But China is mostly in a world of its own.
China has its own stable of self-driving companies that have largely developed independently from Western nations. Leading self-driving companies in China include AutoX, Pony.ai, WeRide, search giant Baidu, and ride-hailing firm Didi Chuxing.
Last December AutoX became the first company to test fully driverless vehicles on public roads in China—though this was merely a test of its technology, not the launch of a commercial driverless service.
I haven’t done very much reporting on these companies, so I can’t talk at length about their prospects. But what seems clear is that, if American self-driving companies face serious overseas competition, it’s most likely to come from China. (The most formidable company without American or Chinese ties is probably Russia’s Yandex.) China’s authoritarian but relatively competent government could theoretically find ways to speed the development of Chinese self-driving technology, whether that’s by building supportive infrastructure or shielding companies from liability concerns.
But so far, Chinese companies don’t seem to be dramatically ahead of their American counterparts, so I don’t rate this an extremely likely possibility.
6. Self-driving trucks win (5 percent)
Self-driving cars get the most attention, but a substantial minority of self-driving companies believe that trucks, not cars, will be the first application for self-driving technology.
Companies like Kodiak, Embark, and TuSimple focus on automating long-haul trucking routes. A couple of other companies—Waymo and Aurora—are hedging their bets, simultaneously working on trucking and taxi projects. Nuro, a delivery startup I’ll discuss more below, is also working on long-haul trucking via its <a href=”https://techcrunch.com/2020/12/23/nuro-acquires-autonomous-trucking-startup-ike/”>acquisition</a> of Ike last year.
Some of these companies are designing their systems to work autonomously only on freeways. Under this model, a human driver will drive the truck to a transfer point near the freeway. The trailer will then be transferred to an autonomous truck that will haul it on the freeway to another city. Then the trailer will be switched to another human-driven truck to navigate the last few miles of tricky urban driving to get the truck to its final destination.
This model has a number of advantages. Freeways are controlled environments that are generally free of pedestrians, cyclists, and other obstacles. They are generally well-marked and lack complex intersections. These factors could make the task of automating freeway driving relatively simple.
And in theory, a company could parlay the data and experience garnered from early success with self-driving trucks to other markets, growing to eventually be a major player in a range of self-driving applications.
However, I remain a skeptic of self-driving trucks for a simple reason: while freeways are easier to navigate in some ways, the potential costs of failure are extremely high. If a fully loaded semi truck loses control at 70 miles per hour, it could cause real damage. And there’s no obvious way to test this technology incrementally. At some point, each of these companies is going to have to push a button and send a truck driving on its own at freeway speeds. I’m skeptical anyone is going to feel confident enough to take that risk that any time soon—or that regulators in many jurisdictions will feel comfortable enough to allow it.
7. Delivery robots win (10 percent)
Nuro is another company that’s planning to haul stuff instead of people. But instead of automating large, high-speed trucks on the highway, Nuro is building small, low-speed delivery robots optimized for streets in residential neighborhoods (Nuro’s vehicles are bigger and faster than sidewalk robots). This is an attractive initial application for self-driving vehicles because a smaller robot traveling at a top speed of 25 miles per hour (40 km/h) is less likely to kill someone.
Two years ago, I made the argument that startups like Nuro that focused on low-speed applications could disrupt Waymo. The other startup I pointed to in that piece was Voyage, which was building a low-speed taxi service at a retirement community in Florida. I argued that focusing on low-speed operations in a controlled environment would greatly simplify the technology challenge, allowing these companies to enter the market much sooner than companies trying to build full-speed taxi services serving entire metro areas.
I then argued that once these companies had mastered these simpler applications of self-driving technology, they might be able to use the data and operational experience they glean to move upmarket and eventually beat companies (like Waymo) focused on more challenging driving tasks.
This prediction hasn’t held up well. It’s true that Waymo has struggled to bring its own technology to market at scale. But it’s not clear that startups like Nuro and Voyage are doing better. Voyage was acquired by Cruise earlier this year.
Nuro is still an independent company and appears to be going strong. However, Nuro seems to be in a similar holding pattern as Waymo. Earlier this week, for example, Nuro announced a pilot project delivering pizza for Domino’s—a partnership Nuro first announced way back in 2019. Like Waymo, Nuro has seemed to be on the verge of large-scale commercial operations for two or three years.
This doesn’t necessarily mean Nuro is doing anything wrong—it might just inherently take years of practice to get an autonomous delivery vehicle running. But this pace does suggest that focusing on low-speed delivery robots isn’t much of a shortcut to commercialization.
Still, the fact remains that Nuro is one of the few companies actually running a commercial self-driving service. And it may be gaining experience that will allow it to expand rapidly in the next few years. If Nuro does succeed, that could be good news for Zoox. While Zoox is primarily focused on building a taxi service, it could easily pivot to building driverless delivery vehicles for its parent company, Amazon. Also, if Nuro succeeds we can expect a lot of other companies building self-driving stacks to court Udelv as a potential partner or acquisition target.
8. Somebody else wins (10 percent)
Technology markets have a way of surprising us. It’s possible that some startup I haven’t noticed yet is destined to embarrass the well-funded companies mentioned above.
Another possibility is an “everyone wins” scenario. Maybe self-driving isn’t one problem but a bunch of individual problems, each requiring its own carefully designed technology stack. Maybe driverless taxis, driverless vehicles, and driverless trucks are three distinctive markets that will each be dominated by its own group of companies. And maybe early versions of the technology will require so much human supervision that we’ll have a slow and gradual transition from human-driven vehicles to driverless ones. I don’t think this is a likely outcome, but it’s possible.
Another possibility is that nobody wins: maybe self-driving is an even harder problem than people appreciate, even after years of setbacks, and it will take decades, rather than years, to get working. In that case, we could have yet another “AI winter” where a lot of companies scale back their research in this area. Again, I don’t think this is likely, but it’s a possibility.
Few people have been working on self-driving cars longer than Chris Urmson. Urmson played a key role on Carnegie Mellon’s team in all three of DARPA’s famous Grand Challenges between 2004 and 2007. He then led Google’s self-driving project for several years. Urmson left Google after being passed over to become the CEO of the spin-off company that became Waymo.
“I’d been leading and building that team and, for all intents and purposes, general managing it for years,” Urmson told Bloomberg in a Thursday interview. “Of course I wanted to run the program.”
Bloomberg asked Urmson about Tesla’s Autopilot technology—and particularly Elon Musk’s claim that Tesla vehicles will soon be capable of operating as driverless taxis.
“It’s just not going to happen,” Urmson said. “It’s technically very impressive what they’ve done, but we were doing better in 2010.”
That’s a reference to Urmson’s time at Google. Google started recruiting DARPA Grand Challenge veterans around 2009. Within a couple of years, Google’s engineers had built a basic self-driving car that was capable of navigating a variety of roads around the San Francisco Bay Area.
A couple of years later, Google started letting employees use experimental self-driving cars for highway commutes—an application much like today’s Autopilot. Google considered licensing this technology to automakers for freeway driving. But the technology required active driver supervision. Urmson and other Google engineers decided there was too great a risk that drivers would become overly reliant on the technology and fail to monitor it adequately, leading to unnecessary deaths.
No time to waste
After leaving Google, Urmson co-founded the startup Aurora with two other prominent self-driving executives. Former Autopilot boss Sterling Anderson reportedly left Tesla in 2015 after clashing with Elon Musk over Musk’s aggressive timeline for developing fully self-driving technology. Drew Bagnell was a senior member of Uber’s self-driving project.
Late last year, Uber sold that project to Aurora, more than doubling Aurora’s headcount and cementing Aurora’s status as the largest remaining independent self-driving startup.
For the last couple of years, Aurora has focused on long-haul trucking as its first commercial product. Urmson predicted to Bloomberg that Aurora would be the first company to deploy self-driving technology for long-haul trucking routes at a “meaningful” commercial scale.
But the Uber deal could also make Aurora a contender in the self-driving taxi business. Not only has Aurora absorbed dozens of engineers with expertise in this area, but a close relationship with Uber will give Aurora an easy way to scale up once its technology is ready.
At the same time, Aurora’s swelling headcount of 1,600 souls puts Urmson under a lot of pressure. At this point, most of Aurora’s rivals are majority-owned by huge companies—either car companies like General Motors and Ford or car companies like Alphabet and Amazon. These companies can continue pouring money into self-driving technology for as long as it takes to get it working.
But Aurora doesn’t have a parent company with infinitely deep pockets. So if Aurora can’t bring a product to market soon, it’s going to need to raise additional money on top of the more than $1 billion it has already raised.
“Urmson doesn’t shy away from the possibility the company may need to raise more money,” Bloomberg reports. “And he’s confident it would be able to do so.”
Of course, that’s what any startup CEO is going to say. But the reality is that investors are fickle. If Aurora can’t demonstrate substantial progress toward a viable commercial product, it might not be able to raise another round of funding. That seems to have been the fate of Zoox, a promising startup that was forced to sell to Amazon at a fire-sale price last year.
John Krafcik, the former auto industry exec who took over Google’s self-driving car project in 2015, is stepping down as CEO of Waymo. Waymo, which spun off as a separate Alphabet subsidiary in 2016, accomplished a lot during Krafcik’s five-and-a-half-year tenure. Still, Krafcik failed to meet the lofty expectations he faced when he took the helm.
Until 2015, the Google self-driving car project was led by engineer Chris Urmson. At that point, Google CEO Larry Page believed the technology was nearly ready for commercialization, so he hired a car guy—Krafcik—to manage the practicalities of turning the technology into a shipping product.
Krafcik spent his first few years negotiating partnerships with automakers. Talks over a potential partnership with Ford fell apart in early 2016. Krafcik then inked a smaller deal with Fiat Chrysler to buy 100 hybrid Pacifica Minivans—a deal that was later expanded to 500 minivans.
In short, Waymo expected its self-driving taxi service to be a big business by around now.
Things haven’t gone according to plan
If that had happened, Krafcik would have been well-positioned to lead Waymo as it scaled up from a small pilot project in Arizona to a large business with tens of thousands of vehicles in dozens of cities. With a deep understanding of auto industry logistics and strong relationships within the auto industry, Krafcik could have ensured that the process of integrating Waymo’s technology into Jaguar and Chrysler vehicles, and then manufacturing a bunch of them, went smoothly.
But that didn’t happen because commercializing self-driving technology proved to be more difficult than Waymo’s leaders—and a lot of outside analysts, including me—expected in 2018. Waymo did launch a commercial service in December 2018, but it came with a big asterisk: at launch all vehicles still had a safety driver behind the wheel, all but ensuring the service would be unprofitable.
It would take almost two more years—until October 2020—for Waymo to stop using safety drivers for most commercial rides. There are now some signs that Waymo’s service is finally expanding beyond its initial market. In recent months, the company has stepped up testing in San Francisco, prompting speculation that the Bay Area could be Waymo’s second market after Phoenix.
But the pace of growth seems glacial compared to the expectations the company set a few years ago. A Waymo spokeswoman told Ars that the company’s fleet has “well over 600 vehicles across all of our locations.” Six hundred vehicles is fewer than 1 percent of the 82,000 vehicles Waymo ordered three years ago.
It’s not clear why. Perhaps Waymo is expanding gradually for safety reasons. Maybe the vehicles require so much human oversight on the back end that the service is unprofitable even without a safety driver. Maybe it will just take time for Waymo to build out the infrastructure necessary to support thousands of vehicles in a bunch of cities.
And to be fair, it’s not clear if any of this is Krafcik’s fault. It’s possible that self-driving is just an inherently difficult problem and Waymo would have struggled to bring its technology to market under any leader. It’s not like anyone else in the industry has leapfrogged Waymo.
But the slow pace of self-driving technology certainly makes Krafcik’s auto industry expertise less relevant. Whatever constraints there might be to Waymo’s growth, an inadequate supply of vehicles certainly isn’t among them.
Krafcik will be succeeded by a pair of longtime Waymo executives who will serve as co-CEOs. Dmitri Dolgov is an engineer who has been part of the Google self-driving car project since 2009 and was previously Waymo’s Chief Technology Officer. Tekedra Mawakana joined Waymo to lead its policy shop in 2017 and rose to be Chief Operating Officer in 2019. Dolgov will focus on improving Waymo’s technology while Mawakana will be responsible for business strategy.