A race to the technological frontier
DARPA’s annual robotic vehicle competition has reached the edge of the technologically possible
- By Josh Rogin
- Feb 12, 2007
DARPA's Grand Challenge Web site
In 2001, Congress gave the Defense Department a seemingly impossible order: Make one-third of DOD’s military ground vehicles autonomous by 2015. The Pentagon handed that mission to the Defense Advanced Research Projects Agency, and that agency responded in 2004 with a simple idea: It would stage a Grand Challenge for robotic vehicles — a competitive race across a desert.
Since then, DARPA’s program has become something of a phenomenon. As teams prepare for the program’s third iteration, the 2007 Urban Challenge, DARPA’s franchise is a high-profile event for the robotics community, the premier training ground for a new generation of software developers and a self-contained cottage industry.
Its potential benefits are huge. In Iraq and Afghanistan, most U.S. deaths occur during convoy operations, often because of improvised explosive devices. The military has deployed thousands of small, unmanned robots to do the dull, dirty and dangerous work of detecting hidden IEDs, but autonomous vehicles would be the technological next step.
“The mission is to save lives on the battlefield. That’s what drives us,” said Norm Whitaker, DARPA’s Urban Challenge program manager. The technology could also prevent some of the 43,000 annual deaths caused by vehicle accidents here at home, he said. People in vehicles are vulnerable, whether in Iraq or on U.S. highways.
No team crossed the finish line in the first Grand Challenge in 2004. But 18 months later, DARPA proved the feasibility of its concept with Grand Challenge II when five cars completed a 130-mile course through a desert with no human intervention.
But in both Grand Challenge races, each vehicle contended with only the course and Mother Nature. The Urban Challenge brings the search for autonomous driving a step closer to reality, Whitaker said. DARPA is building a model of a fictional downtown area at an undisclosed location.
The mock city will include paved roads, stop signs, traffic circles and sharp corners. Rather than racing in turns, all vehicles will be on the course at the same time. Each vehicle must contend with the other vehicles, with DARPA-provided traffic and with other unknown moving obstacles.
Each vehicle must perform a series of military supply missions, guided only by its sensors and software. The fastest three teams to finish the 60 miles of assignments will win $2 million, $1 million and $500,000, respectively. Eighty-nine teams are vying for the chance to compete in November.
DARPA’s $2 million Urban Challenge prize is a hefty carrot. But the reasons that more than 1,000 people are flocking to enter the Urban Challenge are as eclectic as the people themselves. And many of those who don’t win the prize money still cash in.
Some teams have their roots in the most prestigious robotics programs in academia. Those institutional programs use the Grand Challenge as a rallying point and often cross-pollinate other research applications with what they learn from the DARPA competition.
For example, perennial Grand Challenge contender Carnegie Mellon University has built a robotic city on campus for several autonomous vehicle projects. Carnegie Mellon’s Red team entry Sandstorm finished second and its Highlander finished third in the 2005 Grand Challenge.
Other teams in the competition have little or no connection to university robotics programs. Velodyne is a Silicon Valley company run by two brothers who sell high-end home audio equipment. In the first Grand Challenge, the brothers developed a rotating light detection and ranging system, which they now sell to other contestants.
Major defense contractors, some of which build unmanned vehicles, have flocked to the Urban Challenge. For example, iRobot, which has sold hundreds of Packbot robots that detect improvised explosive devices in Iraq, teamed with Raytheon to create Team Scorpion.
General Dynamics, Lockheed Martin and Boeing all have teams. Science Applications International Corp. is working with the Georgia Institute of Technology. Northrop Grumman is part of a team that the California Institute of Technology leads.
This year’s event has attracted an influx of auto industry companies. Every major car manufacturer is sponsoring at least one team. Carnegie Mellon recently announced a collaboration with General Motors. Ford and Honeywell have a well-established partnership. Volkswagen is providing significant support for the Stanford Racing Team.
DARPA said it welcomes teams with nonmilitary objectives to enter the race. Safety systems in automobiles are a stop on the road to autonomous vehicles, Whitaker said. “When the driver falls asleep or isn’t paying attention, you are riding in an autonomous vehicle.” A winning strategy
In its 2005 winning effort, the Stanford Racing Team focused almost entirely on software development. Its Global Positioning System hardware, laser-range scanners and sensors were all commercial equipment, said David Stavens, one of Stanford’s lead developers. The Stanford team won, he said, because it used a rigorous testing regimen that effectively debugged the car, named Stanley.
The 2007 Urban Challenge will be exponentially more difficult, Stavens said. In a city environment, GPS technology is less reliable. Compensating for the loss of GPS and correctly identifying moving obstacles will be the most challenging aspects of the event, he said. Stanford plans to offset its dependence on GPS by using additional inertial measurement units (IMU), wheel encoders and integrated sensors. They are alternative methods for determining position and orientation.
Stanford sees itself as a pioneer in the effort to make streets safer with the use of driverless cars. “It’s really relevant for improving the quality of life in this country,” Stavens said.
Carnegie Mellon’s Tartan Racing Team plans to meet the challenge by relying less on lasers and more on radar and vision systems, said Chris Urmson, director of technology for Tartan Racing. The team will upgrade its robotic vehicle by adding support for advanced reasoning and multiple behaviors. “It’s a much more complicated set of behaviors that we have to implement,” Urmson said.
Teams with a military focus willingly sacrifice their chances of winning. Oshkosh Truck provides thousands of medium and large trucks to the Army and Marine Corps. Its monster entry, the TerraMax, is built for reliability, not speed.
“Our focus at this point is on developing the technology for supporting the soldiers and the Marines,” said Chris Yakes, director of Oshkosh Truck’s Advanced Products Group.
Oshkosh is developing a ruggedized field kit for upgrading military vehicles by adding autonomous functions. Military convoys could use the kits, allowing drivers to perform other tasks. Refuse collection trucks and airport runway snow plows could also use the technology, company officials said.
Other teams have new product development as their primary objective for entering the Urban Challenge. Torc Technologies is comprised solely of Grand Challenge alumni from Virginia Tech. Torc and Virginia Tech have united to build a vehicle for the Urban Challenge.
Torc has experience converting Grand Challenge technologies into profit for the company. It now sells compact GPS and IMU systems for locating missing vehicles.
Small businesses enter the Grand Challenge to get exposure and market their products to the robotics community. Cybernet Systems, a Michigan company with 50 employees, wants to polish its image as a robotics leader by equipping an Urban Challenge team. The small company also provides consulting services to large competitors such as Ford/Honeywell.
“If we can do this, it enhances our reputation,” said Cybernet Systems President Charles Jacobus. The Grand Challenge often yields unanticipated benefits. In the early 1990s, Cybernet researched how to create a virtual sense of touch — known as force feedback — for NASA. It then patented and licensed the technology, which is now a feature in nearly all video gaming systems.
Several component companies market extensively to the Grand Challenge contestants, offering components such as Crossbow inertial navigation systems, SICK laser radars and NovAtel GPS devices. Roboteq offers a discount and engineering help to any team that uses its motor controllers. A class system emerging
DARPA is subsidizing Stanford, Carnegie Mellon and nine other teams with $1 million grants as part of its new Track A Grand Challenge designation. Based on proposal submissions, the government gives certain teams funds in exchange for usage rights to the technologies for which taxpayers paid.
“This money has really given us an opportunity to compete on a much more level playing field with some of the more well-funded teams,” said Al Wicks, associate professor of mechanical engineering at Virginia Tech.
But other teams left to fend for themselves say the DARPA funds give an advantage to those teams who need it least. Professor Ernest Hall leads the Cincinnati Bearcats team and is director of the Center for Robotics Research at the University of Cincinnati. Lacking a DARPA grant, Hall has been forced to seek patchwork funding from individuals and local companies.
“We’re trying to do with $10,000 what other teams are doing with $1 million,” Hall said.
DARPA supports the Track A teams with funding because they are safe and well-established, Hall said. But in science, breakthroughs often come from nontraditional sources, he added.
It makes sense for DARPA to invest in those teams that have the most potential, said Tony Mordica, co-leader of the team Trobo, a self-funded team from Missouri with two full-time members. “If they gave $1 million to every team, there would be 1 million teams,” he said. But, he added, even a small amount of funding for teams such as his would go a long way.
DARPA’s prize money and grant awards should not take away from other valuable programs, Urmson said. “It’s important to balance these Grand Challenges with funding for fundamental research,” he said. Carnegie Mellon’s Urban Challenge vehicle is based on years of work funded through traditional methods, he added.
The greatest value of DARPA’s Grand Challenge program is that it gives exposure to the latest advances in robotics technology, Urmson said. Projects that would otherwise go unnoticed shine in the competition.
Come race day, he said, “it’s put up or shut up.”