Autonomous SUV to compete in DARPA Urban Challenge

The Georgia Tech Sting Racing Team left for the Defense Advanced Research Project Agency's (DARPA) Urban Challenge in California this past Wednesday. The team, which is composed of faculty and students from the College of Computing and the College of Engineering, has been working on a driverless Porsche Cayenne for the past year to prepare for the 60-mile urban course competitions.

The semifinal national qualifying event will be held Oct. 26 at the former George Air Force Base in Victorville, Calif. The top 20 teams will then compete in the final event Nov. 3 for a possible grand prize of $1 million.

"Everyone who works on it also would really like to be known as someone who worked on a successful project, so [the publicity] could be a great thing for all of us involved," said Matt Powers, research scientist for Interactive Computing.

The Porsche itself has been augmented with nine laser sensors, eight dual Pentium processors and five radar sensors. Many of the sensory and software systems were provided by Science Applications International Corporation (SAIC), who was partnered with Tech in research and development on the car.

"I would say 80% of the teams have corporate partnerships. I know Stanford is partnered with Volkswagen, Carnegie Mellon is partnered with Toyota and there are some corporations competing on their own," Powers said.

The Porsche Cayenne was chosen based on its strong air conditioning system and the integrated computer systems already present in the car. The Porsche's systems already included electronic feedback on the speedometer, engine RPMs and turn signals.

"We looked at using the Toyota Prius, because it promoted environmental values, but at the end of the day it just didn't have the size and power that we needed. We were able to get the Porsche at a large discount, and its computer system saved us hours of work," said Henrik Christensen, KUKA Chair of Robotics in the College of Computing and director of the Center for Robotics and Intelligent Machines. The funding for the car and the additional systems used in the project came from the Office of the Provost, GTRI and the Colleges of Engineering and Computing.

The automated car drives using a combined sensor system and standard GPS map. The multiple sensors placed on the car track the double yellow lines and white line of the road, eliminating the two to three meter inaccuracies within the GPS maps. The team has been working on research and development for the car for over a year, putting in six days of work a week and 10-hour days for the past few months.

"I really hope practice goes well every time we do it, since when we practice parking we park the automated Porsche next to my Porsche," Christensen said.

In order to qualify for the semi-finals, Sting Racing had to pass an initial site test by DARPA. The test in June included a technical report detailing both the design of the car and the algorithms involved in the software, as well as safety and performance tests. The car had to negotiate an intersection with other cars, complete a lap, make a U-turn and pass stationary cars on the course.

"We didn't have any trouble until we had to pass a stationary car parked just past a tight curve. Our car's camera radius didn't extend out far enough to pick it up, so it accelerated through the turn, and then noticed the car in its lane and slammed on its break within maybe a foot of the parked car. We have worked on that since and improved, but I was in the chase car behind the Porsche that day sweating bullets, hoping that we wouldn't hit the stationary car," Powers said.

The team will receive a standard GPS map of the area for the semi-finals two days before the competition, and a USB stick with their specific mission five minutes before the start of the competition. Each car will have a series of locations within the area that it has to reach in a specified order. After five minutes of programming in destinations and routes, the team will sit in tents for up to 10 hours while DARPA officials follow the automated Porsche through the 60 miles of course in a chase car. The chase car has the ability to pause or disable the vehicle for safety reasons or for necessary stops. Should the vehicle have to be disabled, the team will be disqualified.

Cars are not allowed to exceed 30 mph during the course, and the chase cars may pause the automated contestants for breaks, extending the completion time beyond the maximum of six hours per vehicle.

"30 mph really under-utilizes the Porsche's abilities, but we will be able to get to 30 mph faster than any other contestants. Maybe that will be part of our advantage," Christensen said.