|
|
UCLA Computer Science Students Part of DARPA Grand Challenge
Independent Team Places Fourth
By Marlys Amundson
Last March, just outside of Barstow, California, a series of driverless vehicles set out for Las Vegas. One after another, though, they came to a halt, and the $1 million prize for the Grand Challenge went unclaimed.
Sponsored by the Defense Advanced Research Projects Agency, the Grand Challenge was a race among autonomous robotic ground vehicles along a preset on- and off-road course. The competition drew attention to technological issues related to autonomous vehicles, and spurred innovative new approaches.
The Golem Group, whose car traveled 5.2 miles, included two UCLA computer science graduate students who designed and assembled the vehicle’s vision system. Jason Meltzer was with the team from its inception and Eagle Jones joined the group a few months prior to the race.
“We were very pleased with our race results,” noted Meltzer. “Especially when you consider how much less we spent than the other teams. Our total budget was under $35,000 and one team spent more than $3 million.”
The DARPA Grand Challenge tested the ingenuity of groups from universities and corporations, as well as independent teams who conceived and built autonomous ground vehicles designed to complete the rugged 150-mile course.
Grand Challenge rules required all vehicles to complete the preset route without human control of any kind. The vehicles were required to have the ability to perceive terrain features, ground conditions, obstacles and other Challenge vehicles, and be able to intelligently control their speed and direction so as to avoid or accommodate them.
Of the 104 teams that applied, only 15 vehicles qualified for the race. Of those, two withdrew prior to the beginning of the race and four others were unable to make it out of the starting area. Only four teams, including the Golem Group, were able to complete more than five miles of the course, and none of them reached the eight-mile mark.
Richard Mason, the Golem Group’s leader, assembled a group of Caltech alumni and Aerospace Corporation employees to design and build their vehicle. He asked Meltzer, who is part of UCLA’s Vision Lab, to head up the car’s vision system. Jones, another member of the Vision Lab, assisted with the associated sensors and coding.
The Golem Group’s vision system consisted of several cameras - both video and infrared - mounted on the vehicle. By gathering data on objects around the car and calculating the distance as measured by each camera and the radar system, the car was able to determine the location and size of fixed and moving objects around it. This information was relayed to the autonomous driving system, which decided the vehicle’s course based on the input.
“Because we were working on the project on a volunteer basis, there wasn’t time for us to do everything that we would have liked,” said Meltzer. “This year we’ll have time for additional research and will be able to generate a new approach that can be applied to the vehicle’s vision system.”
In June, DARPA announced plans for a second race in October 2005, doubling the prize to $2 million. Meltzer and the Golem Group hope to capitalize on their positive results in the first race and secure additional funding from sponsors for the new race.
More faculty and students from UCLA will be actively involved in designing and building the vision system for the Golem Group’s new vehicle. Graduate students in computer science professor Stefano Soatto’s Vision Lab will design and equip the vision systems for at least one, and possibly two vehicles, working on them simultaneously.
“Last year I was their vision expert although I was only a second-year graduate student,” explained Meltzer. “Next year we’ll have more people involved, allowing us to try a more cutting-edge approach. We’re also interested in providing opportunities for undergraduates in computer science and other fields to be involved in the project.”
Meltzer’s role will be very similar to his work in the first race; he will develop algorithms for the vision system and test and implement them. Jones, too, will develop vision algorithms for the car and help select and test cameras to equip the vehicle.
While the Vision Lab will focus on the vehicle’s vision and artificial intelligence-related issues, the other Golem Group team members will be working on the mechanical and control aspects – determining how the vehicle moves and assembling the physical parts of the vehicle.
As they did in the first race, the team plans to retrofit a commercially available vehicle with actuators and the necessary electronics to enable it to run autonomously. They hope to use more advanced artificial intelligence, though, especially in the vision system.
“The Golem Team gave me an opportunity to apply what I’ve learned and to be involved in a potentially important competition with interesting and intelligent people,” said Meltzer. “This year we’ll have time to conduct more research, and create new knowledge that can be applied to the vehicle.”
For more information on the Grand Challenge, please visit http://www.darpa.mil/grandchallenge/.
For more information on the UCLA Vision Lab, please visit http://www.vision.cs.ucla.edu/.
More information on the Golem Group can be found at http://www.golemgroup.com. |
|
