DARPA sees inspiration as trophy of robot race

As director of the U.S. government's research and development arm, Tony Tether hopes today's races will attract tomorrow's engineers and scientists.

Stefanie Olsen
Stefanie Olsen Staff writer, CNET News
Stefanie Olsen covers technology and science.
10 min read
For Tony Tether, an upcoming race of robot cars isn't just about saving lives in the military. It's also designed to inspire a generation of technologists.

As director of the Defense Advanced Research Projects Agency (DARPA), the U.S. government's military research and development arm, Tether pioneered a series of driverless challenges that have wowed the public and four-star generals alike. His agency was also responsible for funding early development of the Internet and wireless infrastructure around the country.

Tether graduated from Stanford in 1969 with a master's degree in science and a doctorate in electrical engineering. After his college years, he started several companies, including Systems Controls, a technology supplier to the military, and the Sequoia Group consultancy. He was appointed director of DARPA in 2001.

CNET News.com talked to Tether ahead of the Urban Challenge, the third in DARPA's series of robot races, which will award $2 million to the winner. The finals will take place November 3 in Victorville, Calif.

Q: We're getting close to the Urban Challenge, and you've witnessed all of the others. So how do you suspect this one will vary from the others?
Tether: Well, it's really a much harder Challenge. In the previous Challenges, we proved that you could make vehicles that could travel long distances...with no drivers, (doing) hairpin turns and going through various obstacles. All the vehicles basically had the same experience. I mean they went on the same track; the same obstacles were there; the obstacles didn't move.

This one will be different. This one they also have a time--they have to get in under six hours in order to qualify for the prize. But the event itself is more than just speed because these vehicles now have to actually act like they're trying to pass the California drivers test.

By the way, we all have copies of the 2007 drivers manual from California. It's a pretty thick manual. So this time we have over a hundred people (out on the course) positioned at corners and so forth, actually gathering data on these vehicles. So, it is very possible that somebody could get through the course in less than six hours and flunk their driving test.

So if all the cars failed their driving test, you obviously wouldn't name a winner?
Tether: Right. They have to obey the rules and show that these vehicles can respond to the environment around them. What makes it different than the other Grand Challenges is that the environment itself is going to be random because each of the robots will have to contend with each other out in the course. The course itself is benign.

It's an urban area that is used by our services in training in urban warfare, but it's sort of a residential area. There are houses, street corners...traffic circles. If an unmanned convoy came to a city and had to go through it, these are the kinds of things you would have to contend with and so it has to do it safely. Safety is really more of the criteria here than speed. (But) speed is still the necessary condition.

I think the thing that's really been accomplished is that these vehicles have learned to recognize not only fixed obstacles, but obstacles that are moving.

Under what condition might the race run over into Sunday?
Tether: The weather. It should be good, but occasionally they do get these major windstorms and get winds that (gusted) up to 60, 70 miles an hour and the fact that we have people out on the course--behind K barriers. I mean the 'bots won't care about the weather, but having that many people out there, I have to be concerned about safety.

What will be the hardest thing about the course, without giving anything away?
Tether: Here's how it's going to happen: we're going to be sending them out on what we call a mission. And the missions are something like this: they will leave the grand stand area and they will be told to go to like five points on the course in (the) order that these points are laid out--go to this place and go this place. Basically, they have to calculate how they're going to get there with GPS technology that calculates a route for you.

(But) there will be a few surprises. Most of them will probably pick the shortest distance, to make (a) way through all these various stops. And we could close a route of the shortest distance and they might find that the road is closed, which means the robot has to recalculate a new path and make a three-way turn and come back out the way it came. So it's going to be fairly dynamic, but fair.

I've heard from people that DARPA made the first Challenge really difficult and nobody could finish it. Then the second year the race was manageable and then, everyone could do it, apart from a few mechanical failures...
Tether: I don't really think so. If you went and looked at the track the first year and compared it to the track on the second year, you would find that the second year was extraordinarily much more difficult. These things had to do hairpin turns, 180-degree turns, tons of right turns--these are the things that hurt the first year.

I think what happened between the first year and the second year is that...at least in the world of the media, it didn't look like much was accomplished in that they only went eight miles. But remember, this was the first time this ever was done, where they went eight miles at speeds of 20, 30 miles an hour.

I think they didn't quite realize how accurate they really had to be. They really had to have external sensors that were watching things like the edges of the road and where there might be obstacles. I think in the first year they relied a little bit too much on a brute force, an almost GPS approach. In the second year, they knew that that couldn't be done so they became much smarter. Plus they already knew that from the people doing this there was, "holy cow," we went eight miles.

In fact, I'll tell you a story. Gen. Kevin Byrnes--who at that time was the four-star general, the U.S. Army TRADOC, which is the training and doctrine command--he showed up unexpectedly one morning on the first one. I was sitting up in the stands with him, when the vehicles started off and Carnegie Mellon went out and another car went out. It was chilly, I mean it was spooky because they went down the road, they made a turn. And he turned to me and he said, 'Now look, there's nobody inside there right?' I said, 'No, no, there's nobody inside there.' He said, 'Now, and there's nobody controlling them remotely right' because it looked like they were being driven by somebody. Now these were the two vehicles that got the furthest, by the way.

So what do you think has been accomplished between the second and now?
Tether: I think the thing that's really been accomplished is that these vehicles have learned to recognize not only fixed obstacles, but obstacles that are moving.

I went to a couple of the site visits and the first thing (one of) the vehicles did for me was a three-way turn. Now, imagine you're watching this vehicle all by itself do a three-way turn and then come to an intersection, and there was a car there already and when it pulled up, another car pulled up after it. It knew enough to wait for the first car to go because by the rules, it knew that car had precedent. But it also knew that it had precedence over the other car that showed up after. It was stunning.

It was absolutely beyond my expectation to the point where I think we're going to have several vehicles finish the track.

Do you expect to name a winner this year?
Tether: Yes, I do--assuming that they don't flunk their driving test while they're doing it. I don't expect vehicles to have mechanical failures, I really don't. Like I said, the roads are smooth. They're just typical urban roads. There are no real tricks. It's just as if you were driving around in one of the suburbs of the Bay Area.

Can you tell us how this challenge came about?
The autonomous vehicle really came about for two reasons. One was that it's a serious mission for the military and that if we can reduce the number of people who are driving convoys in a place like Iraq or Afghanistan, we would definitely reduce the infrastructure to take care of those people. The second reason is that we are worried here at DARPA about the food stock: that the kids today in the United States don't seem to be going into engineering and science like they used to. Now, a lot of people say this is because there's not enough money and I don't believe that's the case. I believe the reason is that we haven't given them exciting things to do.

Everybody owns a car in the United States; you can buy these computers commercially; the sensors are even available commercially; and even the actuators to make the car do what you want are somewhat available commercially because of the handicapped market. The only thing missing now is somebody's imagination to create that secret sauce. In DARPA's arrogant way, we were trying to re-create the man to the moon mission, which created an enormous amount of excitement in this country. We couldn't give a contract to generate that excitement. And (excitement), I believe, is what was needed to have people be interested in science and engineering. I'll tell you, that really has paid off, I mean we have changed the lives of tens of thousands of kids.

We have shown that it can be done, and when somebody shows you that something can be done, you get brave about going and trying something on your own.

Are we close to reaching the military's goal of replacing one-third of ground vehicles with unmanned cars by 2015? And on the flip side, are you excited about seeing this technology emerge in the consumer car market, like many of the teams are working toward?
Tether: There's no question that it has a consumer value. And in fact, if you really think about it, what it will do in the consumer market is the same thing that we're looking at doing in the military market and that is to save lives.

So do you think we're on track to meet that 2015 goal?
Tether: I do. I believe that the military right now is using autonomous vehicles more than they have, but not necessarily direct descendants of any of ours. We have shown that it can be done. And when somebody shows you that something can be done, you get brave about going and trying something on your own. General Dynamics, for example, has a semi-autonomous vehicle that they're selling to the military--semi-autonomous in that if it gets into trouble, it sort of calls home and a person helps it out. The Israelis are using semi-autonomous vehicles for patrolling the Gaza strip.

A lot of professors say that the DARPA of the '60s and '70s doesn't exist anymore because funds have been radically cut. Do you agree with that assessment?
Tether: No. In the '60s, DARPA's budget was $200 million or $300 million, and most of that money was tied up in things like Ballistic Missile Defense. DARPA's budget today is $3 billion. Now, even with all your inflation figures, that just doesn't inflate the $3 billion. We have a lot of money and we spend it again on areas where we believe (it) will make the greatest impact.

We look for ideas that will make an enormous impact, and then we fund them not to bring them all the way, but to prove the technical feasibility. So anytime somebody gives you that (assessment of DARPA), you really need to ask them the following question: how many ideas have they submitted to DARPA that has not been accepted?

So what R&D concepts from DARPA are helping to protect troops now serving in Iraq?
We're doing things that range all the way from prosthetics that (an amputee can control) by his brain to developing techniques (that) will allow people to survive a blood loss of over 60 percent.

What are the top three advances to come out of DARPA in the last five years would you say?
Tether: Let's see, we've revolutionized the whole computer science industry by moving into cognitive processing, that is, computers that learn you as opposed to you having to learn them. Stanford Research, by the way, in Menlo Park is a major contractor in that area. We've also done a lot in biology, again for finding ways for people out in the battlefield to be able to survive their environment.

Then wireless, I guess. If you take your cell phone, you might think that you're wireless and you are. But there's a big infrastructure called towers that really make it work. And what we proved and have developed is the ability to have no infrastructure and still have total cellular wireless type of communication. That's important from a military viewpoint because when we go into an area, we don't have time to build the towers. Now that's also going to be a big commercial thing because if somebody doesn't have to build the infrastructure to have a wireless network, that means that the cost for it is much less than somebody who does, (and it) gives them a great price advantage. Those are three, but I'm not supposed to have favorites.

Will there be more Grand Challenges?
Tether: We don't know. We'll evaluate that after this one and see where we go from there.