The Pentagon’s Defense Advanced Research Projects Agency is famous for its pursuit of high-risk, high-reward technologies. When DARPA bets on technology, the wins or losses can be spectacular. Its latest big bet came in the form of the third edition of the DARPA Grand Challenge.
Known as the “Urban Challenge,” the contest took place on a sunny November weekend at DARPAtown, a specially setup racecourse on the grounds of the vacant George Air Force Base in Victorville, Calif.
There, some of the best and brightest makers in the country were lured into tackling a difficult problem that strongly appeals to their love of making things. Unlike DARPA’s typical top-secret projects, the Urban Challenge was designed to publicly showcase the talents of top-level makers. And the $2 million top prize doesn’t detract from the appeal either.
DARPA’s war planners want a way to keep soldiers safe during dangerous supply missions. Their goal is to develop a vehicle that can drive itself to a dangerous place and do what it needs to do — drop off gear, deliver supplies, or pick up soldiers — without risking the life of a human driver.
Toward this end, DARPA organized the Urban Challenge. The barriers to entering the contest are relatively low — any qualified team of engineers and mechanics can compete. But winning the challenge is a task of Herculean dimensions. To win, the team must build a car that can autonomously maneuver a 60-mile course in an urban environment, “executing simulated military supply missions while merging into moving traffic, navigating traffic circles, negotiating busy intersections, and avoiding obstacles.”
Several rounds of tough competition eliminated most of the contestants prior to the competition’s final day. Just 11 vehicles out of the 36 semifinalists were still in the hunt. Those remaining were given three missions to complete. Moments before race time, the teams were provided with details of their secret missions, the information provided as a computer file on a USB jump drive handed to each team’s leader. Each mission was different, requiring the vehicle to negotiate through the sometimes heavy DARPAtown traffic. To be eligible to win, a vehicle had to successfully complete all three missions in less than six hours.
At first look, designing a self-driving car may seem impossible. But really, “it’s just an extension of present technology,” says Michael Darms, an engineer with the Tartan Racing team. “Cruise control, which is a first step, has been around for decades.”
Darms lists numerous examples of existing cars that handle more and more of the tasks of driving. Many luxury cars have smart cruise control that automatically maintains a safe distance from the cars ahead. Further, some have anticipatory braking systems that use radar to anticipate a crash and pre-charge the brakes for a faster stop.
But the vehicles at Urban Challenge must go beyond simply removing the driver’s foot. They must remove the driver’s brain. Doing so requires a great deal of technology. In use here are radar, LIDAR (light detection and ranging), gyroscopes, and machine vision sensors, all of which paint an incredibly detailed digital image of the area surrounding the vehicle for use by the onboard computer. Control schemes include sophisticated “fly-by-wire” systems that turn the car’s wheels and apply its brakes. And the GPS systems used are incredibly accurate, telling the car where it’s located to within a few centimeters.
The teams at the Urban Challenge range from huge groups from corporations and universities with millions of dollars in funding behind them, to groups of five or six tinkerers who modified their personal cars. As one might expect, the greater the resources, the better the self-driving cars perform, for money does matter. But all of the robot cars here, even the low-budget ones, perform admirably. Here are some typical Urban Challenge entries, ranging from the simplest to the most elaborate.
While other entrants have paint jobs proclaiming Ford, Caterpillar, and Google as sponsors, Ody-Era’s decals include Papa’s Italian Bistro and Mac’s Fabrication Shop. Ody-Era, a 2008 Mercury Mariner from Kokomo, Ind., may not be the most sophisticated, but the fact is, its makers legitimately qualified to compete at DARPAtown against teams a thousand times larger and richer. They showed that a few makers working in a home garage can still attempt great things.
“We’ve spent less than $20,000 in total on our vehicle,” says Rick Bletsis, the driving force behind this self-driving car. Yet Ody-Era made it through several levels of competition in order to compete in the Urban Challenge at DARPAtown.
Unlike most other competitors, Ody-Era relies mostly on machine vision to guide itself through the course. “We use inexpensive, off-the-shelf digital cameras and a desktop computer to sense the environment. A local farmer lent us the John Deere GPS from his farm tractor,” says team member and software engineer Mahesh Chengalva.
With simplicity as its watchword, the car utilizes a program consisting of only 4,000 lines of Visual Basic code to drive itself. Contrast that with the hundreds of thousands of lines of code inside the control algorithms of many competitors.
Ody-Era passed several key hurdles but was stopped well before the finals, undone by a problem with the computer that controlled its drive-by-wire steering.
The team has few regrets. “We did our best, that’s all we can say,” says Bletsis. “Our technology is so novel, nobody else here has much like it.”
Simpler: Plan B
If the first idea didn’t work, then maybe it’s time to consider Plan B. Many would find it surprising that two brothers who run an insurance company in New Orleans would have the interest and the technical know-how to develop an autonomous vehicle able to compete with bigger, richer teams like Tartan Racing and Stanford University. But the Gray brothers are serious about robot drivers, and they’ve developed their own autonomous vehicle called Plan B.
Plan B is a Ford Escape Hybrid packed with sensors and Linux-based computer intelligence. If your map is good, they say, you merely need to tell the vehicle where to go and let the SUV find its own way.
For just $365,000, give or take a few thousand, Team Gray Racing will sell you a real, live autonomous vehicle, complete with an Oxford Technical Solutions GPS system accurate to 10 centimeters, a Velodyne 3D high-definition LIDAR sensor, a fly-by- wire steering and braking system, and Gray Racing’s own AVS-2 intelligent driving computer with obstacle detection and dynamic rerouting capabilities.
Team Gray Racing did very well in the 2005 DARPA Grand Challenge, being one of only five teams to finish. It would have been a miracle if little Team Gray had reached the finals in the 2007 Urban Challenge. But there was no miracle this time, for Plan B was axed in the final cut, the day before the final event.
Not Simple: Boss
Boss, an autonomous 2007 Chevy Tahoe, is like the New York Yankees: it’s a huge monetary investment whose sponsors expect a big payoff. The Tartan Racing team, an amalgamation of academics, engineers, and students from Carnegie Mellon University, General Motors, and Continental AG among others, built Boss, outfitting it with a complex array of radar, laser sensors, and drive-by-wire control systems.
Technology like this doesn’t come cheap, and a multimillion-dollar project like Boss requires the combined financial resources of several global conglomerates. Say what you will about design elegance and resourcefulness; in the end, money talks. That’s why the Yankees win so often.
First place and the $2 million grand prize went to Tartan Racing’s Boss, for completing all three missions 20 minutes ahead of the competition. Stanford University’s Junior came home with a check for $1 million, and Virginia Tech’s plucky Odin got half a million.
All of these teams were deep-pocketed and loaded with the best technical minds available anywhere. But big-thinking amateur makers shall not be dissuaded. As they say in New Orleans and Kokomo, wait until next time.