«  View All Posts

Demystifying Self-Driving Cars

From our leaders | April 19, 2019 | 5 min. read

By the Aurora Team

Aurora CEO Chris Urmson participated in a debate to build support — with a surprising result

Public education is an important responsibility if you’re a self-driving car company. Today, if you live in places like the San Francisco Bay Area, Pittsburgh or Phoenix, Austin or Miami — you may have seen a self-driving car on the road, and simply by virtue of that firsthand experience, the technology somehow becomes more normal. Less unknown. But most people across the country haven’t yet had that chance — and consequently, a lot of people have a lot of misconceptions about the technology, and how it will play out

We had the opportunity to directly confront some of those misconceptions recently when we participated in a debate about the self-driving car in New York City. The debate was held by the organization Intelligence Squared in association with the Adam Smith Society. Organizers asked the audience of about 200 whether they believed self-driving cars would, on balance, be good for society.

Before the debate started, 67% of the audience supported the self-driving car. The audience voted again about the issue after the debate was over. Read through a few of the key issues brought up during the debate — and then, at the end, discover how we fared.

Argument Against #1: The “Moose Problem”

Will self-driving cars simply not be able to recognize all the unusual things the world presents to us… like a moose standing in the middle of a winding Vermont road?

This is an issue that has confronted the industry for decades. At the DARPA Urban Challenge back in 2007, for example, one of the vehicles ran straight into a railroad-crossing arm — essentially because the engineers had never thought to design the software to deal with obstacles that don’t touch the ground.

While we don’t today generally think of railroad-crossing arms as a difficult problem, this general type of scenario is something we’ve never stopped considering. We call this the “long-tail” problem, and 12 years after that vehicle hit the railroad crossing, Aurora has come up with a much better solution. To summarize, basically the problem is, how do you ensure your vehicle will be able to handle the near-infinite variety of objects out there in the world, even if they are only encountered rarely?

Let’s say there is a kind of animal that none of our cars have ever seen before. Maybe a kangaroo. Maybe an extra-terrestrial standing along the gravel shoulder of a rural road. Regardless, the problem boils down to, not hitting the thing.

The beginning of the solution is to design the system so that sensors like lidar, radar, and cameras can accurately perceive matter all around the vehicle. That’s relatively straightforward. And then, regardless of whether the system understands what that particular thing is — keep track of it. We call this strategy, no measurement left behind.

Once our system sees the sensor data and understands that something occupies that particular location, the system tracks that matter, even if it doesn’t understand specifically what type of object it is. So if an alien suddenly decided to dart out into the road, our vehicle would stop. In fact, by enabling our system to classify an obstacle as a non-specific object, the approach we use at Aurora turns out to be broadly similar to the method a human driver might employ when confronted with a similar situation.

Argument Against #2: Won’t Self-Driving Cars Be Expensive?

Lidar, radar, cameras, computers — self-driving cars require a lot of equipment, most of it costly, one of my debate opponents observed. For example, isn’t the price of lidar up around $80,000 per sensor? So how will people be able to afford the technology? If driverless tech has the ability to save people’s lives, could it possibly save only the lives of the wealthy?

No. In fact, self-driving cars will help decrease the wealth-related disparities tied to personal mobility. At least at first, few people will actually outright purchase self-driving vehicles. Rather, we expect most will enjoy the benefits through ride-hailing services, such as Uber or Lyft, which already have boosted the mobility of low-income groups. For example, taxis are not easily available in some low-income Los Angeles zip codes. But a 2018 UCLA study showed that “ride hailing extends reliable car access to travelers and neighborhoods previously marginalized by the taxi industry.” In fact, the study showed that transportation network companies “served neighborhoods home to 99.8 percent of the Los Angeles County population.” The study goes on to say that “ride hailing provides auto-mobility in neighborhoods where many lack reliable access to cars.” That tells us that the fleet model provides alternatives to new populations and income groups who may have experienced compromised mobility in the past.

As for the expense of the sensors — sure, some lidar sensors may cost $80,000, but in general the price has fallen dramatically in the last few years, to just a few thousand dollars. And it will continue to fall. Today, the self-driving car industry demands just a few thousand lidar sensors a year. In a few years, when we need hundreds of thousands, or millions, of lidar sensors, the economies of scale will drive down costs. This is something that happens with technology. The first commercially-available VCR cost $50,000; you can get a DVD player for $26.99 today. According to one of my debate opponents, the first cellphones cost $9,000 in today’s dollars. Current smartphones have many orders of magnitude more features, memory and processing power for a price point that’s at least an order of magnitude smaller. There’s no real magic to this, just good, hard engineering and development work. It’s why we don’t worry about the component cost of the system, in the long term.

Argument Against #3: Won’t self-driving cars be bad for the environment?

Researchers at UC Davis hypothesized that overall vehicle miles driven may increase as riders move away from public transportation like the bus or a train to the privacy and productivity of self-driving cars.

That UC Davis study was fascinating. Essentially, to model how self-driving cars would affect personal mobility, they gave their test subjects access to a chauffeur and a vehicle, and then watched what happened. The single largest effect was for older people — they used their new form of mobility to go out at night, to socialize with friends. These were previously housebound seniors who were reluctant to drive in the dark, or were no longer able to drive. What that study showed was that some increased travel will come from people whose mobility is otherwise compromised. Not just older people but also people with various types of disabilities. Overall, this seems like a really good thing.

In terms of the environmental impact, Securing America’s Future Energy tallied up the various companies testing self-driving cars in California, and they discovered that more than 60% were using electric vehicles as their platform, and another 20% were using hybrids. In contrast, the US carpark today, is made up of only 1% electric vehicles. Which says to me that self-driving cars of the future will go a long way toward improving environmental outcomes. Furthermore, one of the largest influences on vehicle fuel efficiency is the driver; thus we expect that self-driven vehicles should be able to consistently drive in a more fuel-efficient manner, something that will benefit both the owner of the vehicles and the environment.

The Results

After the debate, the audience voted a second time on whether they supported self-driving cars. Recall that the first time, 67% voted to support, while 18% went against and 15% were undecided. After, just 2% were undecided, 10% were against and a whopping 88% of people were for self-driving cars. The support for a future with self-driving cars had gone up by 21%.

To me, the result speaks to the importance of education. At this debate, just 90 minutes of sharing information about the technology and its impacts created an enormous change in attitude. If people are alarmed or afraid of self-driving vehicles, I think much of that emotional reaction comes from a lack of understanding. Informing and engaging with the public is the duty of our industry. Not only because it makes good business sense, but because we all have a shared interest in making transportation safer, more efficient and more accessible.

Listen to All Hail The Driverless Car debate here

the Aurora Team

Delivering the benefits of self-driving technology safely, quickly, and broadly.