A year before the 2005 DARPA Grand Challenge, no one in the robotics world had
heard of the Gray Insurance Company. And no one from Gray Insurance, including
the company's lead software developer Paul Trepagnier, knew anything about
robotics. "We were nobodies," Trepagnier says, so it wasn't surprising that
NOVA's cameras weren't focused on them during the race. They are nobodies no
longer. Against all odds, and in the wake of two devastating hurricanes, the
Gray team finished close behind two robotics superpowers. Their "Kat-5" was one
of only five vehicles from the starting lineup of 23 to even finish the course.
As Trepagnier explains in the following interview, if it weren't for a "$2
million bug," they might have even won it all.
An impossible goal
NOVA: How did an insurance company from suburban New Orleans get
involved in the Grand Challenge?
Trepagnier: Well, Eric and Michael Gray, two brothers who are owners of
the company, are avid readers of Popular Science. There was an article
in Popular Science about the first Grand Challenge and how they were
having a second Grand Challenge. Eric read the article and left it on Michael's
desk with it circled saying, "Are we crazy enough to do this?" Then Michael
sent an e-mail out to a bunch of people asking, "Are we crazy enough to do
this?" And apparently enough people said, "Yes."
So they took the next step, which was to do a little research to figure out
what it would cost, the logistics behind it. They figured out, "Okay, this
seems doable from a monetary and logistics standpoint." Then they approached
me. My main job before the Grand Challenge was to develop software for our
e-business site, to let our clients see their insurance data online. The Grays
thought highly, I guess, of me, and they came to me to see if I could program
it. And I said—and these are the famous last words that I'll never, ever
be able to live down—I said, "No, it's impossible."
NOVA: That didn't stop them?
Trepagnier: No. They said, "Okay," and they kept moving along. So I saw
that they were serious. Then DARPA had a conference out in California where
they announced the 2005 Grand Challenge and the rules. I flew out there with
them. And throughout the whole thing I kept thinking, "You know, it is doable.
We can do this." So when we got back, I started working on it.
NOVA: Did the Gray brothers have any experience with robotics or
computer science?
Trepagnier: No.
NOVA: None at all?
Trepagnier: No. Before meeting up with [team members from] Tulane, I was
the most knowledgeable in computer science. I had a masters in computer
science.
NOVA: Had you ever worked in robotics?
Trepagnier: No. No robotics, no artificial intelligence, nothing like
that. I'm pretty much a business programmer. But I had the theoretical
background in computer science. So I'd touched on a lot of the things that the
Grand Challenge involves.
“Our main strategy was to keep it simple.”
NOVA: When you joined Gray Insurance five years ago, did you have
any idea that you would end up part of a team building a driverless
SUV?
Trepagnier: Absolutely not.
Playing catch-up
NOVA: The race was in early October 2005. When did your team really get
up and running?
Trepagnier: I would say mid-March. We had decided to use a Ford Escape
Hybrid, and I think we got number 30 off the assembly line. We sent it off to a
company called EMC, Electronic Mobility Controls, to get it set up with a
drive-by-wire system. [Such systems, which translate computer signals into vehicle operations, are
commercially available and widely used by physically disabled drivers.]
Then we got the vehicle back in mid-March.
NOVA: So you started late in the game. Did that put you at a
disadvantage?
Trepagnier: Oh, of course. From mid-March, we had just six weeks to get
ready for the site visit [when DARPA officials came to judge if the team was
worthy of competing]. We had six weeks to get the vehicle fully autonomous and
avoiding obstacles—an insane goal. Yet somehow—I mean, we didn't
have the best site visit, but we were good enough to make it.
NOVA: Were you working day and night?
Trepagnier: The last week and a half, it was not quite nonstop, but it
was close. In fact, we got our obstacle avoidance algorithms working at 3:00
a.m. the morning before the site visit.
NOVA: Is it true that you read up about video game programming to
prepare?
Trepagnier: That was the first thing I did. When I came back from the
California conference, I was still the only skilled programmer on the team. So
I said, "Where am I going to find all this AI robotics and stuff like that?" I
actually ended up going to Barnes & Noble and bought—I'm looking at
the titles right now—I bought Programming Game AI By Example,
Physics for Game Developers, AI Robotics. There's a couple more.
I bought game program books because they're a great source for artificial
intelligence algorithms, for pathfinding in particular.
NOVA: So these books had clues for how to program a robot to navigate
through tough terrain?
Trepagnier: Yeah. For our DARPA site visit, we actually used a couple of
algorithms that we got from the books. They got us past the site visit, at
which point we threw most of that code away and said, "Okay, now we've got four
months." Before, we had six weeks. "Let's step back and do this properly."
“We put a lot of blood, sweat, and tears into this car.
We didn’t want to stop
because of Katrina.”
NOVA: Most of your core development team was students from Tulane,
right?
Trepagnier: Yes. And when Gray joined up with Tulane, I decided to apply
there to get my Ph.D. in computer science. So it was myself, Jorge Nagel, who
is getting his masters in mechanical engineering, Powell Kinney, an
undergraduate in biomedical engineering, and Matt Dooner, a computer science
undergraduate.
NOVA: It's a pretty young team.
Trepagnier: Yes.
NOVA: What was your strategy for designing the robot?
Trepagnier: It was very, very simple. Our main strategy was to keep it
simple.
NOVA: You wrote in your application to DARPA, in your technical paper,
that you planned to be "integrators rather than inventors." What did you mean?
Trepagnier: It meant that, whenever possible, if there was a solution
already available in the commercial or academic sector, just grab that, write
the code to glue it together with the rest of our components, and move on.
NOVA: What did you originally name the vehicle?
Trepagnier: Originally we called it GrayBot because we didn't really
spend any time figuring it out.
NOVA: But the name evolved.
Trepagnier: Yes, well, Katrina hit. And that's when we named her Kat-5.
Surviving the storm
NOVA: In the days leading up to Katrina hitting New Orleans, did you
think it was going to impact your plans for the race?
Trepagnier: We actually didn't think it was going to hit us. Here's the
timeline for us: On Friday night, we were getting ready for a little mini race
against ULL's [University of Louisiana, Lafayette] CajunBot. It was Michael's
philosophy that the more competition you can have beforehand the better. So we
were going to have a mini-competition on that Saturday in a gravel pit in
Hammond, which is up north across the lake.
So Friday night we got up to Hammond after work at, like, 6:00, 7:00. True to
form, we stayed up all night doing an all-night stress test, because we had
never run the vehicle for 10 hours straight. We started a program to run the
car in the middle of a field on an endless loop. And we just watched it run and
run and run, and we had a good time and just watched it make circles.
Then, about 4:00 in the morning, we had the radio on. And all of a sudden we
hear on the radio that Katrina now has jogged to the left, and it's coming
straight for New Orleans, and they're likely going to call for an evacuation
either Saturday or Sunday.
So that kind of changed things. I left and got back home around 6:00 that
Saturday morning. Walked in the door and told my wife Jennifer that we were
evacuating in four hours and promptly went to sleep. Woke up three hours later,
packed everything in the car, and left. And we drove to Lake Charles, Louisiana
with our son Taylor, who was about a year and a half old then, to ride out the
storm. My wife was able to get a hotel room there.
But the technical paper for DARPA was due that Monday. So I'm sitting in a
hotel in Lake Charles the morning Katrina hits trying to get the data off the
server in New Orleans so that I can e-mail it to DARPA. I know that we're going
to lose power in our main building any minute, so I'm dialed into my computer
at work trying to get all the pictures and graphs and all the text that I've
written before we lose power. I put it all together in my hotel room and
e-mailed it from there.
NOVA: Where was the vehicle?
Trepagnier: The vehicle was in Hammond. We actually had a very strong
structure, a Quonset hut up there. They parked the vehicle in the hut.
NOVA: Was Hammond hit?
Trepagnier: Hammond's north of the city. It got the winds and the rain,
but not the flooding.
NOVA: So the vehicle was fine.
Trepagnier: Oh, the vehicle was fine. Everything was fine, at least in
Hammond. They got power back, I think, about a week and a half, two weeks after
the storm.
NOVA: In those horrible few days when the city was flooding, what were
you and the other team members thinking about in terms of the race?
Trepagnier: We really weren't in communication. I was in Lake Charles
until Tuesday. Our reservations were up on Tuesday because we thought it was
just an evacuation. We'd evacuate for two days, drive back home. We've done it
before. We did it during [hurricane] Ivan last year. We did it earlier this year—no
big deal.
After two days, though, we didn't have a hotel room, so we drove to Shreveport
to stay with one of my wife's friends. That's when I finally got in touch with
Michael Gray, who was in Tennessee. He said The Gray Insurance Company was
moving offices to Baton Rouge temporarily. So I had to go to Baton Rouge to
help move servers to get the company back up and running.
“We lost about three to four weeks of time just because we were all scattered
to the four winds.”
NOVA: Did it ever seem like you just couldn't move forward with the
DARPA Challenge?
Trepagnier: Yes, it did. But when I got back to Baton Rouge and talked
with Michael, he had talked with Matt and Jorge and Powell. Matt and Jorge were
in Memphis. Powell was in Slidell. It looked like Tulane was going to be out,
so they didn't have anywhere else to be. And they didn't want to stop working
on it.
I felt the same way. I mean, we put a lot of blood, sweat, and tears into this
car. We didn't want to stop because of Katrina. And then Michael said that the
three Tulane students could stay up in Hammond. He'd pay for all their food and
lodging. And when I was finished getting the company servers and stuff up in
Baton Rouge, then I could get back on it as well. He committed to spending what
it took to make sure the project went forward.
At that point, though, it became really just the four of us working on it.
Before then, I'd say over 50 percent of the team was the IT department at Gray.
But they had to get the business functioning in Baton Rouge.
NOVA: How many people on your team lost their homes?
Trepagnier: About 75 percent of us either had our homes destroyed or had
enough damage to keep us from living in our homes, which was my case.
NOVA: Where did you live?
Trepagnier: We ended up at four different places during the evacuation.
We'd go stay somewhere three, four nights, and then it was like, "Okay, we're
kind of in your hair." It was an awkward situation. Some of our friends who
kept us were just amazing, though. I'll be indebted to them forever.
Eventually, I sent my wife and son to Austin to be with my sister and my
parents, who had evacuated there. I moved to Hammond and just stayed there,
worked 24/7 trying to get things done.
Katrina's impact
NOVA: How did the storm change your plans for the vehicle and the race?
Trepagnier: Well, we lost about three to four weeks of time just because
we were all scattered to the four winds. We lost half our team, who had to do
stuff for the company. We were working off generators. And for a while, we
couldn't even go to stores to buy stuff because they were all closed. A lot of
our lofty plans for how we were going to be the best just didn't make it.
NOVA: When did you rename the vehicle?
Trepagnier: I'm not sure. I just remember that DARPA sent us an e-mail
asking if we were okay. And we said, "Yes, we're fine, but we're renaming our
vehicle Kat-5."
NOVA: I doubt it was in honor of the storm.
Trepagnier: No. It was more an act of defiance: "You can mess
with us, but you can't put us down. We're going to get back up again." It's
also a pun, because while it's obviously after Katrina, which was a category 5,
the whole car is actually wired with cat-5 cabling.
NOVA: Given that at the first DARPA Grand Challenge not a single vehicle
made it more than seven or eight miles—
Trepagnier: 7.3.
NOVA: (laugh) 7.3 miles. How far did you think—
Trepagnier: We said 7.4. When we first got involved in this, we said 7.4
miles would make us all happy.
NOVA: You didn't think you would even make it to the finish line?
Trepagnier: Remember I said it was impossible at the beginning? No one
will let me live that down, especially not Eric and Michael Gray. But honestly,
especially after the NQE [the National Qualification Event, where the
competitors for the final race were chosen], we were feeling, I don't want to
say confident, but we were feeling like we had a chance. And then our chase
vehicle driver, who worked for DARPA, came up to us the morning of the race and
said that of all the teams, he felt that we had a very, very good chance of
finishing. And we were, like, "Wow. He's been following all these cars during
NQE and he thinks that we've got a good chance?" And then after 10, 20, 30
miles, we started going, "Hey, we're doing well!"
The $2 million bug
NOVA: The winner this year, the Stanford team, took nearly seven hours
to finish the course. Kat-5 finished just 37 minutes behind that.
Trepagnier: Yes. Have you heard about our $2 million bug?
NOVA: No. What's that?
Trepagnier: Well, the director of DARPA said later that if we hadn't had
a bug where we slowed down in the dry lakebeds, we would have either beaten
Stanford or been very, very close to Stanford's car. The bug meant we went from
30 miles an hour to two miles an hour on all the dry lakebeds. We'd never
tested in an area 100 feet wide like that. We call it the $2 million bug.
Needless to say, it's been fixed.
NOVA: Wow. Despite the bug, were you happy with how you
finished?
Trepagnier: Oh, of course. When our car crossed the finish line,
we just went insane! And then it got even better when our chase car driver said
that our elapsed time was going to be somewhere in the seven hour
range—which meant we were in the same league as Stanford and CMU
[Carnegie Mellon University]. For us, with no previous experience and only
about five and a half months of development, to be in the same league as CMU,
which is just a robotics powerhouse, and Stanford—they've got some of the
smartest people in Silicon Valley, and they've also got Sebastian Thrun, who
worked at CMU. We were just so ecstatic.
NOVA: What made your team successful?
Trepagnier: Well, there are a few things. One, while we didn't have the
biggest budget, we at least had a decent budget [roughly $650,000]. We had a
very small team of very competent people who trusted each other implicitly.
That was so critical. I knew that if I asked Jorge to do something and he said,
"Yes, I can take care of it," that it was going to get done to either my
expectations or better than my expectations. And we all felt that way. A small
team can do so much so fast. As soon as you start adding 10, 20 people, you've
got a bureaucracy. You've got people going in different paths, and it's just
not the same. Also, a lot of our skills and talents meshed really well with one
another.
And after Katrina, we were on our own. We were an isolated group, and all we
did, night and day, was work on the car. Most of the team slept in the same
building as the car. I don't think we could ever achieve that again.
NOVA: After the race, The Wall Street Journal ran a story under
the headline "Amateur Team Defeats the Experts." The story was pretty
complimentary, but how do you feel about being characterized as amateurs? Is
that fair?
Trepagnier: Pretty much. I mean, I now know a lot more about
robotics—I'm getting a doctorate in it. But at the time, we were
amateurs. We really had no experts, true experts on anything related to
autonomous vehicles. I don't think we're amateurs anymore. We had a really
impressive trial by fire that's made us—I wouldn't quite call us experts,
but, you know—
NOVA: Seasoned?
Trepagnier: Seasoned, yes, is a good word.
NOVA: What's next for the Gray team?
Trepagnier: It looks like there's going to be another Grand Challenge.
NOVA: I hadn't heard that.
Trepagnier: Well, they sent out questionnaires to all the teams that
competed in 2005 asking our opinions on proposed formats and things like that.
It's not definite, but all signs indicate there's going to be a Grand
Challenge. Apparently they're very interested in dealing with moving obstacles
in an urban environment.
We're looking forward to that, hoping that we get the chance to compete. We're
confident that given a year, year and a half of lead time with our core team
still intact, that we'll show the so-called experts that we're no longer
amateurs. But then again, we'll also have a target painted on our backs.
Everybody will be gunning for us because we won't be the underdogs anymore.
|