How self-driving cars will cut accidents 90 percent (Q&A)

For Road Trip 2015, CNET talks with the University of Michigan's Peter Sweatman about the rapid merging of computers and cars, and the fake city in Ann Arbor where it's being put to the test.

Stephen Shankland principal writer
Stephen Shankland has been a reporter at CNET since 1998 and writes about processors, digital photography, AI, quantum computing, computer science, materials science, supercomputers, drones, browsers, 3D printing, USB, and new computing technology in general. He has a soft spot in his heart for standards groups and I/O interfaces. His first big scoop was about radioactive cat poop.
Expertise processors, semiconductors, web browsers, quantum computing, supercomputers, AI, 3D printing, drones, computer science, physics, programming, materials science, USB, UWB, Android, digital photography, science Credentials
  • I've been covering the technology industry for 24 years and was a science writer for five years before that. I've got deep expertise in microprocessors, digital photography, computer hardware and software, internet standards, web technology, and other dee
Stephen Shankland
10 min read
Peter Sweatman, director of the University of Michigan's Transportation Research Institute
Peter Sweatman, director of the University of Michigan's Transportation Research Institute. Stephen Shankland/CNET

ANN ARBOR, Michigan -- Peter Sweatman isn't in charge of the computing revolution that's sweeping the auto industry, but he's at the center of it.

As director of the University of Michigan's Transportation Research Institute (UMTRI) in Ann Arbor, he's 30 miles west of Detroit and in the heart of car country. Ford, General Motors, Fiat's Chrysler and countless suppliers are all nearby, and they're getting even closer: a new project at the University of Michigan called Mcity will help carmakers develop the automated navigation systems of their self-driving vehicles.

Mcity is a 32-acre microcosm of motoring complete with faded stop signs, roundabouts, lousy weather and out-of-date traffic signals. There, automakers and others can test not just self-driving cars but also radio communications that link cars to each other and to road infrastructure.

By concentrating the work of many companies and academics, Mcity is an important facility for an auto industry embracing the powerful forces of the computing industry. It embodies not just Detroit's cooperation with Silicon Valley but also its competition with the tech hub for engineering talent and customer enthusiasm.

The Mcity effort unites the academic world of research -- such as UMTRI's study about how much self-driving cars could cut car ownership -- with real-world tests like the vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) study that's been under way in Ann Arbor since 2012. Backers include not just auto industry giants but also insurance company State Farm and high-tech companies like wireless chipmaker Qualcomm and network operator Verizon.

Sweatman, an Australian with a specialty in trucking safety who's worked in the public and private sector, has led UMTRI since 2004. He discussed the Mcity project with CNET News' Stephen Shankland. The following is an edited transcript of their discussion.

Q: Let's take it from the top. What is Mcity, and what is it going to accomplish for the industry?
Sweatman: Mcity is an environment for advancing connected and automated technologies. They're transformational and the public wants them, so we're going to have to move quickly. We created a Mobility Transformation Center (MTC) to concentrate on connected and automated technology, all its ramifications, and how we can move it faster into massive deployment. A key element of that is to have a safe off-roadway facility for testing automated vehicles. So we've created Mcity. It's a fake downtown -- a physical simulation of a dense, complex urban environment. It has many real situations packed into a 32-acre area -- city blocks, suburban streets, suburban arterials, rural roads, freeways and ramps, roundabouts, traffic circles and complex skewed intersections.

Click here for more Road Trip 2015 stories.

It's an interesting combination of thinking about human factors -- how the human interacts with vehicles and the infrastructure -- and about robotics -- how the machine behaves. We bring experts who understand these different worlds together at one test facility. The idea is to be able to reproduce the most challenging situations that vehicles find themselves in cities. We can keep repeating the same scenario so we can move forward quickly.

How many miles of roads are there?
Sweatman: We've got 4.2 miles.

What can you do at a place like this that General Motors or Toyota couldn't do on their own test tracks?
Sweatman: For one thing, we have different varieties of real traffic signals. There aren't too many test facilities that provide that. We have a range of lighting conditions. We've got building facades we can move around. We've got mechanized pedestrians and will have mechanized cyclists. We've got a range of roadway surfacing -- concrete, asphalt, simulated brick. It's not just a track, it's an environment. All that detail is important, because the sensors in an automatic vehicle are trying to recognize the situation.

Self-driving cars are one big piece of this. Another is vehicle-to-vehicle [V2V] communications and vehicle-to-infrastructure [V2I] communications. How will that be tested here?
Sweatman: We started with connected vehicles with dedicated short-range communication -- DSRC -- nearly three years ago. We've still got a large number of vehicles in Ann Arbor operating in that mode. We decided to bring together that DSRC connected technology with various levels of automation based on sensors in the vehicle. There's a huge power in doing that. You can think of DSRC as the ultimate sensor that you're adding to the vehicle: not only does the vehicle have machine vision, the other vehicles are talking to your vehicle and giving it info that it otherwise wouldn't have. We see that convergence of connectivity and automation as critical. That's one of the first things we'll be investigating at Mcity -- to see what additional benefit we get with combining DSRC with automation.

There's a plan to expand that V2V and V2I along Interstate 96 and 696 [in the Detroit area]. How many vehicles will be equipped, and how much infrastructure will there be?
Sweatman: As we move from Mcity to Ann Arbor and to southeast Michigan, we get more vehicles and get to a point where we can see crashes being avoided. The Michigan Department of Transportation [MDOT] announced a smart corridor on I-96 and I-696 last year. That's already being deployed to provide the wireless DSRC communication.

We need to get to a point where drivers know and appreciate every day that their vehicle is communicating with intersections, ramps and so on, and that the driver is benefiting every day. Safety is important, but safety incidents only happen infrequently. It's easy for someone to forget why they have that system in their vehicle.

It's important to have smart corridors, we're going broader across southeast Michigan with the Michigan Department of Transportation. We're going to put 500 radios in that infrastructure within southeastern Michigan. Then, with our partners which happen to have large company fleets located throughout that area, we're going to capitalize on that opportunity to deploy 20,000 or more equipped vehicles across southeastern Michigan. That reaches the scale of a real deployment, not just a model deployment.

Walk me through how it works. If you're driving a car equipped with this radio communication ability, it's talking to other cars, to the highway, to onramps and offramps, traffic lights -- what actually happens?
Sweatman: You're getting warnings and information presented to you as the driver. One we found to be very popular in Ann Arbor is the electronic brake light. Imagine a car two or three vehicles in front of you in the traffic stream is braking suddenly. You can't see their brake lights or the vehicle, but the signal coming from that vehicle is picked up by your vehicle, so you're able to preempt that sudden deceleration.

A look at Mcity, a test site for self-driving cars (pictures)

See all photos

In some cases what's valued by the user is the knowledge that other vehicles know they're there. This applies to motorcyclists -- they're very happy to know the car or large freight truck that normally wouldn't see them knows they're there.

It's not trying to get your private information. It doesn't even know who you are, but it knows you're there, and you know other players around you. The computer in your car is constantly sifting through information looking for something unusual or a sudden change that would indicate someone's going to crash into you.

Do you need protected spectrum for V2V and V2I? There's this 5.9GHz radio-frequency band carved off for this in 1999.
Sweatman: We love the idea of protected spectrum, but we also understand that spectrum is very valuable. If it's going to be proved possible to have unrestricted and uninterfered-with safety messages still being sent while other uses are taking place, then we're willing to consider that. It would require a fair bit of testing. We'd like to test it in our facilities here.

Are you concerned about the Federal Communications Commission opening that to discussion -- that maybe these airwaves should be publicly usable for any wireless activity?
Sweatman: I wouldn't say we're concerned. We're very aware of it and involved in the process. We've got great faith in the FCC that they're doing an in-depth investigation. At some stage it's going to be important to know exactly what technology is available to reliably share the spectrum, so some movie being downloaded can be interrupted so your brake-light message from three vehicles ahead of you can get through. We need to make sure that happens every time in a reliable way. And we're going to need to know it'll work with a large number of vehicles in a small area. Those things need to be tested not just in a lab, but in a real environment.

A lot is happening in Silicon Valley with self-driving car technology. Google gets a lot of attention, Apple perhaps is noodling around with it and automakers have opened up labs there. Is there some concern that the intellectual center of this work is on the West Coast and not here at the center of the US car business?
Sweatman: Michigan is the global center of the automotive industry. The future of mobility -- particularly as it's influenced by connected and automated technologies -- is playing out here. The facilities and capabilities we're providing are helping the industry in this part of the world move forward quickly. A little bit of healthy competition is a good thing.

We recognize it's not just about automakers. It's a new ecosystem of companies who aren't used to working together. We've got the automakers and tier-one suppliers, we've got the traffic signal and traffic sensing community, we have insurance, we have telecommunications, we have big data, we have IT. We pull them all together here in Michigan. We deploy these technologies in a big enough representation that we can learn fast, then we can improve it and do an even bigger deployment. We call that rapid learning cycles. That's something the automotive industry has honed over many decades.

Vintage illustration of a futuristic three-wheeled self-driving "dream car," 1961.
Vintage illustration of a futuristic three-wheeled self-driving "dream car," 1961. GraphicaArtis/Getty Images

What about testing in the real world? That seems to be Google's approach -- you build a car and see what it encounters, like a woman in an electric wheelchair chasing a duck in the street.
Sweatman: That kind of thing Google is doing is still testing. What we're trying to do with Mcity is to be able to produce unusual and challenging situations but then make them replicable so we can do them many, many times. To get just one episode of a particular situation isn't that helpful in quickly developing your algorithms and system. You want to say, "Let's re-run that again and change something," and get it right.

We want to get these vehicles on the roadway, operating with real users, as fast as we can. Google isn't doing that with real users, they're doing it with employees. What we're doing here, and what we're good at in Michigan, is doing these bigger deployments where we recruit volunteers. Then we learn much faster what people really want. What do people love about this?

It's going to be a revolution in safety. That's sufficient to get anybody to get this done as soon as possible.

It seems to me the way to build acceptance in the marketplace for self-driving cars is showing that although they might not necessarily be 100 percent safe, they are safer than human drivers. How well does that sales pitch hold up when the first kid playing street hockey is killed by a self-driving car?
Sweatman: When we get to high density of automation, serious crashes will reduce by an order of magnitude -- a factor of 10. There will still be crashes. We're going to have a mix of automation and conventionally driven vehicles, and we don't know how other drivers view other vehicles. Will they know that other vehicle is an automated vehicle and behave differently?

Humans cheat. They're pushing the envelope all the time, whether speeding or going through traffic signals. The machine does what it's programmed to do. It doesn't cheat.

There's no doubt the automated future, when we get to it, will be much much safer. Ninety percent of crashes are caused by human error. That's not going to happen with a machine, but there will be situations where the machine is confounded or something happens where there are crashes, but they will be far fewer. As soon as we see significant deployment, we're going to see crash rates going down. The people themselves are going to know it's safer.

What about sci-fi ideas like platooning [with multiple vehicles linking electronically into energy-efficient groups] or green waves where cities coordinate traffic into smooth traveling?
Sweatman: I'd call those applications, like you get on your computer or your phone. There are going to be endless applications. Platooning freight trucks on long journeys is one.

For years we've been trained by the auto industry that it's great to have a car because you get freedom, this autonomy. How comfortable will people be yielding that to a computer? People feel more worried about accidents when they're a passenger who's not in control, and they feel more secure when they are the driver who's in control. With self-driving cars, it's not the freedom of the open road on Route 66 anymore -- it's more like public transit.
Sweatman: I don't think people are going to be having this internal debate whether they should relinquish control. Americans more than anybody else around the world will leap at the chance to spend their time better. In the early stages, people who are nervous about it can resume control. That'll happen less and less frequently. The draw of using their time better will take over and people won't be so concerned.

There's a debate in the industry about whether self-driving cars are a complete break, with totally different technology, or whether there's a gradual shift to self-driving cars -- a continuum of steadily more sophisticated driver-assistance technology like cruise control on steroids.
Sweatman: I think it is a very steep continuum. It has to be. The reason the slope will be steep is the demand will be strong.