Mar 8 2005
Students from Duke University's Pratt School of Engineering are partnering with Carnegie Mellon University's "Red Team" in an effort to win a $2 million prize from the Defense Advanced Research Projects Agency (DARPA). All they have to do is complete the toughest ground course ever devised for a self-guided robotic vehicle.
The contest, called the DARPA Grand Challenge http://www.darpa.gov/grandchallenge/, is a race between fully self-guided ground vehicles to be conducted in the Southwestern United States on Oct. 8, 2005. DARPA said it hopes the contest will stimulate significant improvements in technology that may be adapted for future military use.
The race route could be as long as 175 miles over desert terrain featuring natural and man-made obstacles. The exact route will not be revealed until two hours before the event begins, and the course must be completed within 10 hours. Winner takes all -- there is no second place.
The Red Team is developing two Hummer vehicles for the competition -- Sandstorm and H1ghlander. Duke's contribution will be a "Duke Radar Object Identification Device" (DROID) unit for each vehicle. The DROID uses radar to identify objects on the ground that must be avoided to prevent damage to the vehicle. DROIDs will provide the locations and classifications of obstacles to the Red Team vehicles' navigation systems.
Radar bounces radio waves off a target to get a reflection. That reflection is received by the radar's antenna and is then analyzed by a computer. Radar systems are essentially range devices that measure distances to each radar reflection.
Radar is routinely used to track aircraft and ships, but those applications are fairly simple because the sky and ocean are essentially featureless. Identifying and distinguishing among objects on cluttered landscapes is far more challenging, and radar has never been successfully adapted for off-road ground vehicle use.
The Red Team investigated radar last year but ultimately dropped it from the project because of the immense challenges, particularly ground clutter issues, involved in adapting the technology for ground vehicle use, said Josh Johnston, lead student engineer for the Duke team. Johnston, from Boise, Idaho, is a senior double majoring in electrical and mechanical engineering, and president of Duke's Robotics Club.
Duke is taking a novel approach in its quest to make radar a contributing part of the Red Team's vehicles, he said. Instead of directing radar pulses from the top of the vehicle down, the team is positioning the radar for a horizontal field of vision similar to what a human driver would see. DROID will be placed on the front of the car.
"Other onboard systems can figure out topography, so we're focusing on the location of above-ground objects," Johnston said. "Our job is to identify obstacles that could damage the vehicle." But he notes that's not so easy.
William "Red" Whittaker, robotics professor at Carnegie Mellon University and leader of the Red Team project, explained that "radar can sense fast and far through desert dust, but cannot yet distinguish fine detail, or reliably distinguish between content and clutter. Radar interpretation is a grand challenge in its own right, and the software challenge is profound.
"Bright ideas and computer simulations are not enough to win a race, so Duke must push radar interpretation beyond the laboratory, onto robots, and out to the world."
The Duke team is analyzing different obstacles -- fence posts, barbed wire barricades and natural hazards such as rocks, ravines and trees -- and identifying distinguishing features to create "obstacle profiles." Those profiles become part of a catalog that the computer program consults when trying to identify and classify an object detected with radar.
One man-made obstacle called a tank trap still stumps the team, however, said Robert Kielb, one of two faculty advisers for the Duke students and a senior research scientist in the mechanical engineering department. (Adjunct electrical and computer engineering professor Jason Janet is the other adviser.) A tank trap looks like a man-sized "jack" from a child's jacks-and-ball pickup game. Tank traps can slow down, block or impede the movement of armored fighting vehicles.
"Tank traps are difficult for radar systems to locate because they are unintentionally stealthy designs, because they don't provide strong reflections to the radar system," Kielb said.
Kielb and Johnston plan to create a full-size tank trap on campus so they can hammer away at the reflection problem.
It's been slow going, but the team is making progress, Johnston said. "We're essentially trying to program a computer to tell the difference between a tumbleweed and a tangle of barbed wire -- it's incredibly difficult," he said. Johnston reported Duke's accomplishments to date to the entire Red Team in mid-February in Pittsburgh.
Aaron Mosher, a Boeing Co. engineer who is working with the Red Team and with the Duke University team on the DROID project, said the team was developing advanced filtering and reasoning logic for the radar system.
"A person can look at the radar output and manually pick out important features while ignoring most of the clutter," Mosher said. "But we have to build all of that intelligence into a computer system that will perform that operation automatically. The clutter that comes from the ground makes it hard to separate out legitimate targets. Our DROID system's logic must walk a fine line between sending out too many false alarms versus missing important obstacles.
"So far, we've had some promising results," Mosher said. "This radar system's intelligence has grown significantly over what we used in last year's race."
Duke students have been working on the DROID project since September. Faculty in Duke's Pratt School of Engineering support the effort by providing expertise, laboratory space and by incorporating aspects of the project into senior design classes for electrical engineering and mechanical engineering majors. Robotics Club members have been working on the project constantly.
Five Duke students went to the Nevada Automotive Test Center in Carson City over the holiday break in December to work with Red Team members and get experience in the field.
This is the second attempt at meeting the Grand Challenge for the Red Team, an alliance of individuals, non-profits and for-profits. On March 13, 2004, Red Team was one of 15 robotic vehicles that attempted to navigate a challenging 142-mile route through the Southwest desert. None was able to complete the mission. Red Team beat all competitors, but managed to travel less than 10 miles.