Is the race to autonomy another race to the bottom?

This competition won't be won by who gets their technology on the road first, it'll be won by who gets it cheap, first.

Tim Stevens Former editor at large for CNET Cars
Tim Stevens got his start writing professionally while still in school in the mid '90s, and since then has covered topics ranging from business process management to video game development to automotive technology.
Tim Stevens
9 min read

The drive to develop autonomous technology has often been called a race, and a race is exactly what it is. Virtually every major vehicle manufacturer, and plenty of independent suppliers, are scrambling as hard and as fast as they can to make cars that can drive themselves. And though the race is well and truly on, the exact location of the finish line hasn't been agreed upon. There's a perception that the first company to the make the technology work will reap all the rewards, that this company will be carried forward on the crest of an industry-shifting wave big enough to sweep the competition right off the field.

There is some truth within that over-simplification, because given how tumultuous every industry is these days even the tiniest of advantages have a way of becoming integral. However, while this is a race, the real finish-line isn't actually getting this technology to market. It's a little further down the road.

The real winner will be the one who makes autonomous technology affordable.

Jaguar's I-Pace gets the Waymo treatment

See all photos

An autonomy primer

Self-driving cars are such a seemingly straightforward yet curiously nebulous concept that I hope you'll forgive a little expectation-setting here. We are, right now, living and driving in a world with millions of cars that can, in some limited circumstances, drive themselves. Automatic emergency braking, where a car brings itself to a stop faster than a human can, is an increasingly common feature that's been proven to reduce accidents and, ultimately, save lives.

Advanced lane-keep assist systems, like AutoPilot or ProPilot Assist, are another glimpse at this autonomous future, a hint at what's to come. However, despite all the yahoos on YouTube filming themselves riding in their Teslas with their hands off the wheel (including chief yahoo Elon Musk during his CBS This Morning interview), drivers of these cars still need to watch the road and maintain control. That, then, is not full autonomy.

What I'm talking about today, in this article, is proper autonomy of the sort where you can get in the car, tell it where you want to go, and then take a nap, watch a movie or, heaven forbid, get some work done. In industry parlance this is often called Level 5, though Level 4 can offer exactly the same functionality, just limited to certain geographic areas or even weather conditions. 

That's proper "driverless" autonomy, and thanks to Waymo it's actually coming to public streets, and to public availability, later this year -- though perhaps not in the way you might have envisioned.

The benefits

Why do we care about autonomy? Why this rush to create technology that the vast majority of Americans are downright afraid of? The reasons are too many to readily iterate here, and indeed the industry as a whole is still getting its head around the secondary and tertiary benefits, but here are a few highlights:

  • Safety: Without a doubt, this is the number one reason for cars that drive themselves. People may be afraid of giving up control, but with an estimated 40,000 driving-related deaths in the US last year and with human error the primary factor in 90 percent of all accidents, there's clearly room for improvement.
  • Senior mobility: Many of us would like to think we'll be driving ourselves well into our golden years, but the truth is that many senior citizens reach a point where they must turn in their licenses. Self-driving cars will ensure independence for more people.
  • Family mobility: Hey parents, how nice would it be if you could have your car go pick your little ones up from soccer practice, instead of having to make that happen yourself? 'Nuff said.
  • Efficiency: Autonomous cars will be more efficient drivers for a variety of reasons, including smoother acceleration and braking, meaning less consumption and greater range. They'll also decrease traffic congestion, because they'll be unaffected by commute-plaguing issues like rubbernecking -- and because there'll be fewer accidents in the first place.
  • Stress: Stuck in traffic? That sucks. Stuck in traffic while binging on season four of Expanse? That's not be so bad.

The cost

To do a proper benefit analysis you of course need to look at expense, too. At present, its the sensors that really drive up the cost of self-driving cars. While there's debate about how many sensors are needed, and indeed what type, the general consensus is "the more the merrier." Here are the most common types:

  • Imaging: These are basically cameras, working in the visible spectrum to virtually watch the road. Many cars already have basic imaging sensors built into their windshields to provide lane-keep and adaptive cruise.
  • Radar: These emit active pulses of radiation and use the responses to identify objects in three-dimensions, and likewise are increasingly common on cars for adaptive cruise systems. Radar is largely unaffected by weather and these sensors are affordable, but provide a relatively low-resolution view into the world.
  • Lidar: These shoot a field of lasers and measure the responses to create a proper, three-dimensional map of the world all around the car. Lidar scanners provide the most detail, but have limited range and are often quite expensive.
  • Sonar: Yes, even sound sensors are used. In autonomous cars, they can check blind spots and help for self-parking.

In an ideal world, an autonomous car would have all of the above and many multiples of each type for redundancy's sake. However, there are two key reasons why you can't simply festoon a car with all the scanners and imagers in the world.

The first is pretty basic, and that's aesthetics. I personally have no qualms getting into an ugly bus or train or other form of public transit, and I had no problems hopping into Waymo's autonomous van that looked to all the world like it was wearing a dunce cap and having a bad reaction to a series of bee stings.


Would you buy a car that looks like this? The good news is, you may never have to.


That said, I might have second thoughts about paying money to own a car like that, and I'm guessing a lot of you would feel the same way. The good news is that sensor packages from companies like Velodyne continue to get smaller, and eventually they'll almost totally dissappear.

The other reason that companies must be more selective in sensor application is the more troubling one: Cost. Getting a vehicle manufacturer to go on the record about how much their sensor packages cost is like pulling teeth, but you can be sure that the combination of laser, radar, imaging and sonar sensors, plus enhanced GPS receivers and a beefy onboard supercomputer to run all the numbers, typically costs some multiple of the sticker price of the car they're attached to. Prototype autonomous cars are often carrying in excess of $100,000 in additional hardware.

When Waymo CEO John Krafcik stepped on the stage at the North American International Auto Show in 2016, he wanted to show off the company's then-new Pacifica mini van. However, the big news was the promise to get Waymo's sensor package to a price below $7,500. That would perhaps make it the cheapest autonomous sensor package out there, but there's also the cost of the aforementioned supercomputer to make sense of all that data, which in Waymo's case will be sourced by Intel . That'll likely add at least a few grand more. 

After I was lucky enough to take a ride in Mobilieye's prototype autonomous car, the company's CEO Amnon Shashua promised to do even better: Between $8,000 and $12,000 for all the sensors and processing power to make them work. That car, in its current form, relies inexpensive cameras, costing $10 to $20 each. However, Mobileye's final car will have Lidar and radar sensors, too.

And that brings us to Tesla, the only company actually charging consumers for supposed self-driving functionality. Buyers of the Model 3 can opt for the $5,000 Enhanced Autopilot package and then pay another $3,000 for a promise of so-called "Full Self-Driving Capability." However, there's no guarantee for when that will be available, and many industry experts are extremely skeptical that Telsa's car, which lacks Lidar scanning, will ever offer the kind of redundancy required to make an autonomous car truly driverless.

Even if Tesla pulls that off, many of the consumers most eagerly looking forward to autonomy would have a hard time stomaching even that $8,000 cost. The good news is, many won't have to.

Intel Mobileye's autonomous cars in Jerusalem

See all photos

Business models

There are a lot of reasons why a company like Waymo doesn't want to build cars -- just ask Elon Musk about how he's sleeping these days. But even beyond the notion of avoiding "production hell," the sheer cost of these systems means that even if Waymo were building cars, few would be able to afford them.

And that's at least partly why Waymo is instead launching a service in Arizona this year, not a car. While the details are still slim, Waymo's service will provide personal, point-to-point transportation ala Uber or Lyft. When you figure Uber completed more than four billion rides in 2017 alone, amortizing the massive up-front cost of that sensor package doesn't look so bad.

Uber and Lyft will be desperate to follow as quickly as possible, because while adding new, human drivers allows those startups to expand quickly into new markets, driverless cars will solve a lot of their ongoing problems, things like background checks, employment rights and, that most pressing problem of all, profitability.

Even if these services spent $20,000 per-car to enable driverless operation, that would theoretically allow a single car to run 24 hours a day, quickly offsetting the salaries of multiple drivers of the meatbag variety.

Intel Mobileye Autonomous Ford Fusion

Mobileye's present autonomous car is among the most subtle looking visually, thanks to its reliance on imaging sensors.


Racing to the bottom of your driveway

So if you happen to live in the Phoenix area, you'll have access to autonomous cars very, very soon. But what about the rest of us in the rest of the world? And what about the kind of car we'd actually think about buying or leasing... or whatever it is we want to call services like Care by Volvo

For now, we're stuck waiting to see which of the world's automotive manufacturers -- nearly all of whom are trying to out-invest, out-research and out-hire their way to autonomy -- will get these sensors to market first. However, given Mobileye's solutions sit at the cheaper end of the spectrum, and given that company already has partnerships with most major car companies globally, don't be surprised if the true winner isn't actually an automotive manufacturer at all, but rather a supplier. 

And so, picking up on Mobileye's estimates, let's say that the first proper, autonomous cars come with an extra $8,000 to $12,000 option package. Many buyers wouldn't even bat an eyelash at ticking that box when ordering something like an Audi A8Mercedes-Benz S-Class or BMW 7 Series, and it will be interesting to watch the adoption of this technology among a new generation of post-millennial wealthy who want the luxury of being driven without the pains of dealing with a human driver. 

The really interesting part, however, will be what happens next. Adaptive cruise, when it first hit the S-Class 20 years ago, was a $2,800 option. That's about $4,200 in today's money for a feature that comes standard on the cheapest Toyota Corolla. That same, plummeting price trajectory will happen far more quickly with autonomy thanks to much higher demand. 

Even relatively expensive Lidar sensors will become inexpensive once they're churned out en masse, while the cost of advanced processing units from Intel, NVIDIA, Qualcomm and others will quickly follow suit. And then, after the massive, up-front research and development costs have finally been paid off, and some generous insurance discounts factored in, these systems will be very affordable.

What happens next? Will the automotive industry crumble after people realize that buying a car that spends 90 percent of its time sitting idle is a bad investment? Or, will owning a vehicle suddenly become an income opportunity as people send their cars out join a fleet during the 9-to-5? Most importantly: what are we going to do with all the free time we get back when we don't have to deal with traffic? For answers to these questions, we'll just have to wait and see. 

Update, 10:55 a.m. Eastern: Updated to clarify the distinction between Level 4 and Level 5 autonomy.