Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
October 7, 2005

Robotic Research Goes Off-Road

Nicole Hemsoth

For decades, scientists and engineers have worked on how to overcome the challenges of applying robotic technology to replace or supplement human intervention in a variety of environments.

Now, a $2 million prize competition sponsored by the U.S. Department of Defense to build a fully autonomous vehicle that can drive on- and off-road through the desert without human drivers could impact a host of civilian applications for this kind of robotic technology, including the potential of enabling blind people to “drive” cars. However, the primary goal of this DoD-sponsored field test is to accelerate research and development in robotic ground vehicles that will ultimately help save soldiers' lives on the battlefield.

Currently, U.S. troops in Afghanistan use robots to explore enemy caves, and in Iraq robots help to detect roadside explosive devices and to inspect weapons caches. In the future, the U.S. military is looking to ground-based robotic vehicles to play a broader role in combat situations in order to keep soldiers out of harm's way.

Toward this end, SGI visualization and storage technology is helping a Carnegie Mellon University-led team to prepare for a 175-mile race this fall sponsored by the Defense Advanced Research Projects Agency (DARPA) that is designed to demonstrate the efficacy of autonomous ground vehicles to navigate treacherous desert roads and trails – without human drivers – and in the least amount of time.

The 2005 DARPA Grand Challenge will be held on Oct. 8, 2005, in the Mojave Desert and will include both natural and man-made obstacles throughout the racecourse. The team that develops an autonomous ground vehicle that finishes the designated route in the shortest amount of time within the 10-hour time limit will receive a $2 million prize.

Because the exact route will not be revealed to competing teams until two hours before the race begins, the ability to quickly, effectively and accurately plan the best route will be a strategic advantage for achieving success on race day. No remote control is allowed in the DARPA Grand Challenge race, so vehicles must be able to sense and drive autonomously by computing where and how to drive, relying on sensors and other advanced technologies to navigate the challenging desert terrain.

A four-processor Silicon Graphics Prism power configuration and a two-processor Silicon Graphics Prism deskside configuration, each with two graphics pipes, will provide Carnegie Mellon University's Red Team with the computing muscle and visualization capabilities to enable route planning for its two robotic vehicles competing in this year's DARPA Grand Challenge. The Red Team uses a combination of archived terrain data and preplanned route locations referenced and tracked via the Global Positioning System, as well as on-board sensors to help decide the best path and how to overcome obstacles along the way.

A 10 TB SGI InfiniteStorage TP9500 storage array from SGI and Engenio will store satellite imagery data of the route. The Red Team will preload millions of points of terrain data on the Nevada desert race site for pretesting the vehicle, and for the actual race the team will rely on the industry-leading speed of the TP9500 to readily process the satellite data for mapping to the final race route.

“SGI's contribution to the Red Team with its powerful Prism and InfiniteStorage systems has huge advantages for the planning and simulation that ultimately determines the intended route and speed of our unmanned vehicles at race time,” said Carnegie Mellon University's William “Red” Whittaker, captain of the Red Team. “With ready access to the satellite data and SGI's advanced visualization features and Intel Itanium 2 processors, Silicon Graphics Prism is truly a high- performance system that blazes through our biggest and most demanding sets of terrain data, providing our team with a highly accurate map of the route and what we believe will be the winning edge for our vehicles to run a successful race.”

According to Paul Temple, SGI Federal's senior business development manager, “We are excited to contribute to an innovative new class of robotic vehicles through Prism's ability to rapidly digest and visualize multi-terabyte terrain data sets stored by the TP9500 array, giving the Red Team a powerful tool and groundbreaking insights for planning its race routes. The Red Team's productivity has been transformed by Prism's lightning fast ability to load data sets at multi-gigabyte-per-second rates that will no doubt serve the team well in the 'golden hour' before the DARPA Grand Challenge when time is of the essence.”

The 40-member Red Team, which includes sponsors such as AM General, Boeing, Caterpillar, Intel, SGI and SAIC, will compete in this year's Grand Challenge with two entries based on AM General's Humvee vehicle design. The Red Team's “Sandstorm” vehicle will return to defend its 2004 distance and speed record, and a new H1 design dubbed “H1ghlander” will also compete.

In June, DARPA announced that 40 teams, including the Red Team, were selected to advance to the semifinals of the 2005 DARPA Grand Challenge autonomous ground vehicle competition. DARPA selected the semifinalists from a field of 118 entrants (all vehicles are developed without government funding). The Red Team's H1ghlander and Sandstorm vehicles both made the cut and have been invited to participate in a week of qualifications in late September and early October at the California Speedway in Fontana that will narrow the field to 20 competitors for the winner-take-all race on Oct. 8.

If the 2004 Grand Challenge is any indication, the Red Team is seen as the frontrunner for this year's competition. Last year's race, in March, consisted of a field test of autonomous ground vehicles that ran from Barstow, Calif. to Primm, Nev. and offered a $1 million prize. Although in last year's Grand Challenge race no vehicles were able to complete the difficult desert route, the Red Team's Sandstorm vehicle traveled 7.4 miles before getting stuck; no other entry got as far.