Work around the clock. Spend tens of thousands of dollars. (At least.) Do the impossible. And get well rewarded by Uncle Sam. Dozens of small groups are driving to do just that by March 13, 2004. The impetus is the Grand Challenge, a race sponsored by the U.S. Defense Advanced Research Projects Agency. Show up with a vehicle that can direct itself from Los Angeles to Las Vegas in ten hours-sorry, no human help allowed-and you could win serious bragging rights. Oh, and a million dollars cash.
DARPA has long funneled money for esoteric research projects to large defense contractors and universities and has launched some amazing technical jumps including the Internet, which was originally called ARPANET. Now the agency has initiated an unusual effort to develop self-guiding military vehicles by pitting teams against each other in a race for new technology. The hope is ultimately to develop autonomous technology that will replace people with machines in many dangerous situations-and save money at the same time.
DARPA first became interested in autonomous devices in the late 1960s. In 1987, one agency-funded project created a vehicle that could travel 4.5 kilometers on varying pavement types and road widths at an average speed of 14.5 kilometers per hour. But none of the usual research suspects have come close to the performance that DARPA really wants: high speeds sustained for hundreds of kilometers over any type of terrain. So it has decided on something completely different from its usual approach of sending large checks to large organizations. Instead, the agency has invited any and all U.S.-based groups to build autonomous vehicles and enter a race from California to Nevada. Officials hope the competition will tempt not only academic and corporate researchers, but also some unknown genius. “This is the first time in about 40 years that DARPA [has opened] its doors for new technology [in which the project] was not run through either a campus or industry,” says Jaime Teuffer, owner of 5BEST, a San Diego consulting firm competing in the challenge.
The deal is good for DARPA: a million dollars is chicken feed in defense industry contracts, and the government pays only for demonstrated results. Contestants, on the other hand, have their work cut out for them. “We haven’t been given near the amount of time” that such a project would ordinarily entail, says Paul F. Grayson, chief engineer of American Industrial Magic of Traverse City, MI, which is fielding Team AV. Although Grayson started his effort after advance word of the contest leaked in the summer of 2002, most teams did not begin work until the official announcement in February.
The competition isn’t called the Grand Challenge for nothing. The exact route will remain unknown to the teams until two hours before the starting gun on March 13. It can run as long as 480 kilometers-both on roads and off-and must be completed in ten hours for an average speed of 48 kilometers per hour. But obstacles could disrupt the driving, and no human intervention will be allowed. “There will be spots that we will stop-either there’s an obstacle in the way or there will be a problem in the algorithms,” says David Anton van Gogh, leader of Team Caltech and a university staff member in engineering and applied science. “To make up the time we’ll have to be going 60 [miles per hour] at points.” (That translates to over 96 kilometers per hour.)
High speed with no one at the steering wheel sounds like a recipe for flaming mounds of wreckage. The teams are employing varying strategies to solve the attendant problems. Caltech and Carnegie Mellon have an obvious wealth of resources, especially in the software that will be critical to controlling the vehicle. Grayson’s group, on the other hand, has split its design into tiny bits and is farming them out around the world to qualified individuals who have a little time and want to be involved. “You can impose on some of the world’s experts when you are dealing with short times,” he says. For a brief period, Grayson says, he had “the number eight guy in neural networks and the number three guy in computer game coding working on his team.”
Differing tactics are also abundant in solving the problem of autonomous steering. Instead of paying tens of thousands of dollars for an inertial guidance system, for example, Grayson plans to buy a surplus model for scrap metal prices. Both the CyberRider team, based in Irvine, CA, and Team 1010Delta, run by robotics consulting firm VIDEOptions of Lake Oswego, OR, will use video processing to detect obstacles. Virtually all the teams, however, are using gasoline-based vehicles to obtain the speed, power, and endurance that the winner will need.
With cars, computers, guidance electronics, and software to procure, the expenditures are serious, and paying for the projects is a fundamental problem. Grayson and American Industrial Magic plan to sell advertising space on their vehicle. Other teams are self-funding their projects. “We have four business owners on our team that can contribute hard cash when needed,” says Ivar Schoenmeyr, leader of the CyberRider team, which is based in Irvine, CA, but has members at MIT and also in Stuttgart, Germany. Schoenmeyr thinks that his team’s cost could run $100,000. Caltech has budgeted $400,000 for its effort-half earmarked for salaries and the rest for hardware, including a $14,000 1996 Chevy Tahoe.
Making money go a long way is tough. “I think we paid too much [for the car], because when I was writing the check, the salespeople were a bit too happy,” Caltech’s van Gogh remembers. All the teams are depending on sweat equity: Grayson’s typical day runs from 5:30 in the morning until 11:30 at night, while Caltech will have used roughly 13,000 person-hours by race day in March. “A lot of the kids sacrificed some of their school work during the spring term,” van Gogh says. “They could have gotten A’s easily, but they worked on this too long.”
What are the chances that anyone might win? It depends on whom you ask. Van Gogh thinks that there is a 90 percent chance that his team’s final vehicle could run the course if there were no time limit. The real challenge, he says, is the ten-hour countdown. “If we take high risks, I think the chances are over 50 percent,” he says. “If we don’t take any risks, then they’re definitely under 50 percent-probably 20 percent.” And yet VIDEOptions’ Team 1010Delta reported a successful practice run in late June and now is confident that it can complete the task.
Still, the outcome is far from certain and the costs are serious. So why are so many groups taking a stab at the Grand Challenge? Some want to publicize their companies. But all are driven by the chance to say that they beat the best technical talent in the world. “This is an opportunity to, if you were an athlete, enter the Olympics,” Schoenmeyr says. “The only prerequisites are if you think you’re smart enough to pull it off.” And if no one wins, there’s always next year to show who’s really in the driver’s seat.
How AI is reinventing what computers are
Three key ways artificial intelligence is changing what it means to compute.
These weird virtual creatures evolve their bodies to solve problems
They show how intelligence and body plans are closely linked—and could unlock AI for robots.
We reviewed three at-home covid tests. The results were mixed.
Over-the-counter coronavirus tests are finally available in the US. Some are more accurate and easier to use than others.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.