|Photos by Ryan Donahue|
Bob is having a bad day. Rain has left a slick of water across the asphalt, and reflections bouncing off the surface are confusing his sensors. Directed to avoid a simple barrier, he loops past it, then proceeds to drive in circles like a puppy chasing its tail. A collective guffaw goes up from his handlers, a group of Caltech students who have taken on the mind-numbingly difficult task of trying to build an autonomous robotic vehicle. Bob is a car — a 1996 Chevy Tahoe to be precise — that can drive by itself. Or at least that’s the idea. Soon the reflections are affecting its ability to steer, and Bob is veering badly to the left, having previously demonstrated perfect aim. Dave van Gogh, the team’s manager, shakes his head stoically. Bob, it seems, is discombobulated.
In 10 days’ time, and assuming it passes the qualifying trials, Bob will enter the first great race of the 21st century, the DARPA Grand Challenge, which calls for a vehicle to drive itself from Barstow to Vegas (some 200 miles) in under 10 hours. Team Caltech is one of 25 groups that have signed on for this tournament, which will pit robotics researchers against industrial engineers, high school students, hot rodders, battle-bot builders, and a guy in Berkeley who is making the task even more insane by halving the number of wheels — his vehicle is a motorbike.
The DARPA Grand Challenge, or DGC as it’s known, was conceived by the Pentagon’s Defense Advanced Research Projects Agency (DARPA) as a response to a congressional mandate that by 2015 one-third of ground-combat vehicles should operate autonomously. Over the past decade, DARPA has handed out millions of dollars to military contractors and university departments to research unmanned vehicles, yet the agency has gotten precious little return on this investment. Surely the great American entrepreneurial spirit could be brought to bear on the problem. As DARPA project manager Colonel Jose Negron told The New York Times last October, the feeling was that “There are solutions out there in the community and nation that people weren’t offering, because they don’t deal with the military complex. So we are inviting little mom-and-pop folks out there to help spur advancement and take us where we need to be.” The incentive: a million dollars, plus the bragging rights — which for some of the competitors is worth even more.
Next week at the California Speedway in Fontana, all DGC challengers will be put through their paces in the grueling Qualifying Inspection and Demonstration round, in which each vehicle will have to show that it can autonomously navigate a one-and-a-half-mile course littered with obstacles. Only those that make it through will be allowed forward to the actual race, which starts at the Slash X Café near Barstow on March 13. It so happens that halfway through the qualifying week, term finals begin at Caltech, which sums up the problem van Gogh is facing. “It’s pretty hard to get students to do anything when they’ve got assignments and exams,” he says. Then again, “The great thing about Caltech students is they aren’t afraid to try anything.” And so it is that on a rainy Sunday afternoon in the car park of the Santa Anita raceway, engineering professor Richard Murray and seven students are sitting in a tent hunched over their laptops, trying to decipher what’s going on in the “brain” of a highly computerized but decidedly feral SUV.
CALTECH hardly counts as a mom-and-pop shop; the team has more than $400,000 in sponsorship, much of it from Northrop Grumman. It’s also getting help from scientists at the Jet Propulsion Lab, the folks who built the Mars rovers. Bob is actually using some of the same algorithms the rovers use, so he shares, as it were, a genetic heritage with the intrepid bots now boring away at rocks on the red planet. But where the rovers only have to move 100 feet a day, to complete the Grand Challenge Bob will have to average 20 miles an hour — more than 9 meters a second! All that while avoiding boulders, trees, embankments and abandoned cars.
The race is being held across the desert, the bulk of it on dirt backroads. Participants will be told the final route only two hours before the race, at which time they’ll be given a CD with the GPS coordinates defining the actual course. Teams will have to input those data and make any final adjustments before handing their vehicle over to DARPA personnel. During the race there can be no contact with the vehicle, and in some sections vehicles will be required to remain within a corridor as narrow as 10 feet — any straying will result in immediate disqualification.
Of the 25 teams formally accepted into the DARPA race, 14 are from California and nine from the L.A. area. Employees of Rockwell Scientific in Thousand Oaks are fielding two separate vehicles, both based on a dune-buggy concept. Also from L.A. is the competition’s only kiddie team, a group of students from Palos Verdes High, Palos Verdes Peninsula High and Beverly Hills High, spearheaded by 16-year-old boy genius Joe Bebel and assisted by their moms and dads in high-tech industries. Few people think the kids have a chance of winning, but the fact they even qualified is cool enough.
Contributing the indispensable gonzo factor without which no auto event would truly rate are the A.I. Motorvators, who describe their approach as “somewhere between a neural net and a brick on the accelerator.” The Motorvators have been motivated by West L.A.–based hot rodder Chris Pedersen, a mechanical-design consultant whose leisure activities involve building and racing “gravity-powered race vehicles, flying wings and NASCAR stock cars.”
Above all these valiant but, by their own admission, essentially amateur entries stands the Goliath known as the Red Team. Based out of Carnegie Mellon and headed by robotics legend William “Red” Whittaker, the Red Team has more experience, more expertise and more money than probably the rest of the field combined. A long Tom Wolfe–ish piece in the current issue of Scientific American details their expenditures, including $725,000 on the vehicle itself, a converted military Humvee named “Sandstorm” packed to the gills with technology. The Red Team’s Web site lists several dozen sponsors headed by Boeing and Caterpillar, and a number of companies have donated major pieces of hardware including a $60,000 position-tracking system from Applanix and a $47,000 top-of-the-line Navtech radar that can see through dust. On top of that, Space Imaging, a satellite mapping company, has given the Reds a vast cache of high-
resolution data worth almost $200,000.
For the Red Team, the prize money is of little consequence; it’s their reputation that’s at stake. Whittaker and his colleagues at Carnegie Mellon’s world-renowned Robotics Institute have designed autonomous vehicles that map underground mines, crawl up volcanoes and search for meteorites in Antarctica. It seems to have been taken for granted that if anyone is going to win the Challenge Race this year, it will be them. (If no one finishes in the requisite time, DARPA will hold the race again in 2005 and again in 2006.) But against all the hoopla surrounding this juggernaut stands at least one team that has been virtually ignored, and might just give the Reds a run for their money.
Axion Racing, based out of Westlake Village, sounds like a most unpromising candidate for a cutting-edge robotics challenge, but according to the team’s representative, its vehicle has already driven every possible variation of the DARPA course. Heading the Axion team is Bill Kehaly, the affable North American operations manager for Tropical Waters, a bottled-water company based in the tiny Micronesian state of Kosrae. Kehaly came to the DARPA Challenge when he was trying to think of a way to get Tropical Waters into the consciousness of the U.S. consumer. Suddenly he had a vision: “Wouldn’t it be great if you had this giant water bottle driving across the desert?” To that end, Kehaly assembled a team of managers, accountants, vehicle experts and computer specialists who have automated a Jeep Cherokee painted like a water bottle.
From the moment he decided to enter the race, Kehaly knew what his personnel policy must be: “I remember saying that if you’re going to do this, you can’t do it with people who were first in their class. You’d have to get people who were second or third, people who are young and hungry and can listen.” The Red Team, he says, is full of people who are “the best of the best,” and that, according to him, is their problem. On the Axion team, nobody ever assumed they could ace this task, yet they apparently have. For the past month, says Kehaly, the “Spirit of Kosrae” has been test-driving through DARPA’s course area clocking up hundreds of unmanned miles. There are always things that can go wrong on race day, like a nail puncturing a tire, but unless Kehaly is exaggerating wildly, Axion has already achieved the goal. All on a budget of $200,000.
Like all of the teams, Axion is anchored by its ability to integrate large amounts of sensor and mapping data. In this case, the task is being overseen by Melanie Dumas from the Artificial Intelligence Laboratory at UC San Diego. Most of Kehaly’s technical team are based in San Diego, and it also includes Greg Jones, a mechanical engineer with 30 years’ experience designing specialty vehicles, including driving systems for high-level quadriplegics. DARPA’s main interest in autonomous vehicles is battlefield use, but Kehaly sees a huge potential business here in the mining industry and in vehicles for the disabled.
No one I spoke to for this piece mentioned what must surely be a major application — the car as designated driver.
“It would be great if some small team beat everyone,” Caltech’s van Gogh told me the first time we met. And indeed, in many ways it will be a letdown if the Red Team wins, though it’s hard to believe it won’t. Many people seem to be hoping that the Carnegie Mellon colossus will be beaten by Caltech, yet in the parking lot at Santa Anita two weeks before the trials, it’s clear that Bob is going to need a few more driving lessons. A month earlier I asked Lars Cremean, one of the team’s three group leaders, how far he expected they would go in the race. At the time, he said he’d be happy if they made “double digits.” On that rainy Santa Anita Sunday, the feeling seems to be that everyone will be ecstatic if they make it through the qualifying round.
Building a vehicle that can drive itself 200 miles cross-country is a monumentally difficult undertaking, and any team that even gets to the starting line will have done something incredible. But it takes more than brilliant minds and fast chips to make a smart car. One thing that becomes increasingly clear is the importance of good old project management. Bill Kehaly may not be a tech guru, but he appears to be a fantastic manager — and he’s recruited an experienced team of accountants, lawyers and other professionals. Unlike the Red Team, which has been run like a military boot camp (Whittaker is a former Marine), the Axion team have been having fun. “No one is being paid here,” Kehaly says; they are doing it because they are enjoying it. From the start, Kehaly realized, if he wanted to get the best from his technical crew he would have to make sure they were properly supported. “I told them, you make the magic happen, and I’ll take care of the mundane stuff.”
It is on this mundane front that the Caltech team have faced their most serious problems. Elliot Andrews, a Caltech administrator and former Harley-Davidson dealer who has also run race-car teams, is advising the students on mechanical matters. At Santa Anita, he tells me, “The goal at Caltech is not so much to win the DARPA Challenge as to provide the students with a good learning experience.” If you really wanted to win, he says, you’d have to run the team much more tightly.
Which raises the issue of where is the right place to do these kinds of projects? Universities are critical crucibles for new ideas and advanced concepts, but are they the best environments for the development of practical applications? As Kehaly sees it, such tasks don’t belong in academe and should be farmed out to small businesses. He admits this would require a serious shift in direction for DARPA, and he personally doubts the agency is ready for such change. “I’m the clone of what DARPA say they want. The trouble is, how much of me could they really stand?” On March 13, when the robots roll, it won’t be just the cars that are racing against one another — we’ll also be watching a competition between different cultures.
The DARPA Grand Challenge qualifying trials will be held from Monday, March 8, through Thursday, March 11, at the California Speedway in Fontana. The race itself starts at Slash X Café near Barstow early in the morning of Saturday, March 13. Spectator seating is limited. For more information on the trials and the race, see www.darpa.mil/grandchallenge/.