Chuck Thorpe runs the Navlab group in the Robotics Institute of Carnegie Mellon
University. His interests are in computer vision, planning, and architectures
for outdoor robot vehicles. Since 1984, the Navlab group has built a series of
10 robot cars, HMMWVs, (aka "Hummers") minivans, and full-sized passenger
buses. The research is funded by the Defense Advanced Research Project Agency
(DARPA) for building off-road scout vehicles, and by the U.S. Department of
Transportation for traffic safety and automated highways.
NOVA: Why do we need automated cars and highways?
Thorpe: We need cars that are much more intelligent than the cars that we have
today. This is not because we want to take people out of the cars, but because
we want to help people get to where they need to get to much more safely and
much more efficiently than we can today. People have done a good job driving.
But there's so much traffic congestion, and there are so many accidents still
left that the only way to get better surface transportation is to have
automated cars, automated trucks, and automated buses with all of the sensors
that can help people do the driving.
This computer simulation illustrates one way an Automated Highway could be designed.
NOVA: Can you tell us how you got involved with highway safety?
Thorpe: My interest in highway safety really goes back to when I was a little
kid in 1963. My dad is an emergency room physician. In 1963 he declared a woman
dead who was thrown out of her car in an accident and run over by her own car.
My dad went out that night and bought seatbelts and put them in our car. See,
in 1963, that's how you got better safety. You put seatbelts in cars, you built
Interstate highways, and so forth. Now our cars have seatbelts and airbags and
anti-lock brakes. The cars have gotten a lot safer. We have the Interstate
highway system, which is now officially complete. So we've reduced a lot of
those causes of the accidents. We're left with the 90 percent of the accidents
that are caused at least in part by people. We haven't figured out a way to
make the people smarter. That's why this gives us a chance to make the people
smarter. We have the sensors around the car. We have some of the processing and
some of the intelligence. We can help people drive, we can automate part of the
driving systems. We can have smarter cars and smarter roads.
NOVA: Can you tell us more about the extent to which human error is involved in
Thorpe: Ninety percent of all of today's accidents are caused at least in part
by human error. About 70 percent of those are caused predominantly by human
error. The other 20 percent have some kind of component of human error that
helped cause the accident. So anything that we can do to give people eyes in
the back of their head by having other sensors around—to give them a better
feel for the road, give them better warning systems—any of that will help
reduce fatalities. Anything that we can do then to further automate the system
makes it even safer and makes it even more convenient.
This line of computer-driven cars traveling at 65 miles per hour is a powerful demonstration of how close we may be to automated driving.
NOVA: Why are you doing this?
Thorpe: This is hard work, there's no doubt about it. We've had people out here
all night, every night for the last two months developing this technology
getting ready for the demo. It's also great fun. The work is fun, because it's
a great kick to go out there and ride around on these automated vehicles and
see how well it works. But even more importantly, it is for mortal stakes. We
still have 40,000 fatalities per year in the United States alone, about the
same number in Europe. Anything we can do to reduce that number of accidents is
a big win.
NOVA: Is there any way to quantify how many accidents could be avoided as a
result of the technology?
Thorpe: Trying to figure out how accidents would be reduced requires a lot of
very careful work. For instance, part of the project I'm working on involves
single vehicles that run off the road. So we've gone through the accident
statistics trying to figure out why people drive off the road. Then we've gone
through the technology to figure out how long a warning we can provide them.
Now what we're starting to do is to go through the road database and find out
how wide the shoulders are, to find out if we could provide a two-second
warning—or on what proportion of roads will that actually help save people
from having an accident. When we're done doing all of those statistical
analyses, we'll have some good statistics on how many accidents we can
NOVA: Is this purely an American project, or is it worldwide?
Thorpe: This project has worldwide significance. The problems that we're
dealing with here with accidents and fatalities are really too big to get
nationalistic about. We cooperate quite well with other people. For many years
I have gone to the technical conferences and shared very freely with the
Europeans, the Japanese, other Pacific Rim people. It's an international
community. We're working on the same problems. We have the same motivation. I
hope that we can cooperate and solve the problems together.
It may take many highly visible demonstrations like this to convince the public that the automated car is a safe and practical idea.
NOVA: What are some of the most difficult engineering problems you have
overcome or have yet to overcome?
Thorpe: The most difficult research problem in this whole project has been
perception. How do you see? How do you give the computer the chance to see
what's going on around you? Some things, like radars, are pretty good for
finding big metal objects, but you want not just to know there are big metal
objects, you also want to look at pedestrians and at traffic zones and at
people on motorcycles and lots of things around you. So we've put a lot of
effort into computer vision. We've put a lot of effort into designing new laser
scanners and other sensors like that. That's one set of issues. There's a whole
second set of issues involving things like defensive driving. So we've not been
able to take it out here and run on the roads, but we have running in
simulation back in the labs all of the rules of defensive driving that you were
taught when you learned how to drive. Don't drive next to somebody if you can
drive behind them or in front of them, so you have a safe place to move to.
Don't drive too close to the person in front of you. Be polite to the other
drivers around you. Some of that we've been able to demonstrate today because
we have the sensors, and we can tell what some of the other drivers are doing.
Some of that we're still working on in the lab.
NOVA: Anyone who's used computers knows they crash inevitably. What kind of
chaos would that create if the system breaks down?
Thorpe: The system is designed to have lots of levels of redundancy in it.
People are used to riding on things like elevators that are computer
controlled, and elevators are reliable. A hundred years ago people weren't sure
that they trusted elevators until Mr. Otis had himself hauled up 45 feet in the
air and had the cable chopped and showed that his safety brake held. We have to
do the same kind of things as we get ready to introduce this. We have to show
that there are multiple levels of safety so that if any one thing fails there's
enough redundancy left in the system to keep it running safely to get you
safely off the road or turn control back to the driver.
NOVA: A lot of people really like to drive, they really like the control of
driving. Why would people want to give up that control?
Thorpe: For people who like to drive, we're not going to force them to give up
control. They can still drive. I get two reactions from people like that. One
is, "I don't want to have any of this automated stuff. But I sure wish that guy
who was tailgating me did." And certainly some of the other people in the
traffic flow will have the automated systems and that will make their vehicles
much more polite and much safer and so forth. The second reaction that I get
is, "I really like to drive but every once in a while I don't notice something.
It sure would be nice to have these sensors just to alert me. I can still drive
but I can use these sensors to make me a safer driver."
Thorpe hopes new automated features would be considered similar to today's cruise control.
NOVA: What do you think people's fears are going to be, and will you assuage
Thorpe: People are going to be excited about this once they get a chance to
come out and actually experience it. Part of our strategy in the way that we
built our scenario is to treat it very much like cruise control. So you can
turn it on and off like cruise control, you can go faster and slower. Then
there's a couple of other buttons that do things like close up the gap to the
car in front of you or back away or make you change lanes automatically. So by
giving a very familiar environment like cruise control, by introducing these
functions one at a time, people will get used to it, they'll be comfortable,
and they'll be looking forward to the next feature as it shows up.
NOVA: Do you think this is going to give people more free time, more time with
their families, more time to do the things they enjoy doing while they have to
commute or drive?
Thorpe: I would certainly love to have an automated vehicle so that when I'm
doing my daily commute I can be talking on the phone or working on my laptop
computer, so when I get home I'm done with all that and I have time to spend
with my family.
NOVA: What happens if you get a flat tire or suddenly the driver becomes ill,
has a heart attack, something like that? How do you troubleshoot those
Thorpe: There are a lot of contingencies that you need to worry about when
you're building a system like this. If you have a flat tire or a blowout—first of all that's going to be much less likely because we'll have the
computer sensing the tire pressure all the time. If that does happen some of
the research that's gone on here shows how you can control a vehicle well
enough to bring it over to the side of the road and stop even if it has a flat
tire. If the driver becomes sick, or when it comes to the off ramp at the end
of the automated travel, and asks the driver if the driver is ready to take
over, if the driver doesn't take over and do a competent job, the system will
automatically pull the vehicle over to a safe spot, bring it to a halt, and
send out a distress signal.
An on-board computer checks into the Automated Highway System.
NOVA: How will people get on and off the highway?
Thorpe: How you enter and exit the automated highway depends on which system we
end up building. Certainly for some of the congested downtown areas it makes
sense to have a dedicated lane that only has computer-controlled vehicles on
it. There may be a check-in station similar to the gated on-ramps that exist in
congested areas, or to a toll booth where the roadway checks the vehicle to
make sure that all of the automated systems are functioning and then lets it go
onto the road with all the other automated vehicles. For some of the rural
applications like driving across the country we can't have a dedicated lane
because there are only two lanes. So if you took away one of the lanes and made
it just for the automated vehicles, that would only leave one lane for the
manual passengers. For there we'll treat the system just like cruise control.
You'll drive the vehicle on the road and when you want to turn on the automated
system you'll push the automated button. The system will do a self-test on
board its own system and will see that the radar is working, see that the
controllers are working and so forth before it takes control.
NOVA: When do you think this will actually occur? When will people actually be
driving on automated highways?
Thorpe: Parts of the system are going to be ready very soon. Some of the
warning systems on the bus and on the truck are commercially available now. You
can go out today and buy a radar that warns you if you're getting too close to
somebody in front of you. Some of the warning systems for running off the road
will happen in the near term. The ultimate, automated highway, completely hands
off, feet off—it's hard to predict when that's going to happen. There are
all of the technical issues, but there are also all of the social issues and
the liability issues. Those are some of the questions that we in the consortium
are going to be dealing with over the next four years. If you look at the time
scale to build a new road, it takes about 20 years to plan a road, go through
the environmental clearance, get the funding, and actually go out and build it.
So certainly within the time frame of building brand new roadways we'll have
the vehicles ready to run on them.
The highway of the future? Only time will tell.
NOVA: What is the next step for this program?
Thorpe: After the demonstration is over we have a chance to sit down and
analyze what happened. We'll look at the other research that is going on in
parallel with the demonstration. We'll start to decide which of the
technologies and concepts are most promising. Eventually we'll select a single
prototype, build it, and spend several months in really rigorous testing.
NOVA: What advice would you give young people who are interested in getting
involved in this field?
Thorpe: The most important way to prepare for this kind of work is to cultivate
your curiousity. Building robot cars is both lots of work and lots of fun. It
takes mechanical engineering, electronics, computer science, psychology, and
civil engineering. It also takes imagination and just plain common sense.
Certainly it's important to study science and math. But it's also important to
try building things, to get a sense for what might work and what probably
won't. I spent a lot of time as a kid with building blocks, erector sets, model
airplanes, and eventually tinkering with mini-bikes and motorcycles. I probably
broke more things than I fixed. But I had fun, I figured out how things worked,
and I developed a sense of what it would be fun to build for a career.
Photos: (1) NOVA/WGBH Educational Foundation; (2-7) National Automated Highway System Consortium.