Making a Robotic Cross-Country Racer

Feb. 17, 2005
Designing a robot that can travel 140 miles across the desert is hardly child's play

Raoul Benoit
Contributing Editor

Team Ensco fielded David, which had some problems keeping to the flatter path. Ensco, Falls Church, Va., supplies engineering to the defense, security, transportation, environment, aerospace, and automation industries.

The proposed Path provided by Darpa challenged the robots intelligence and survivability.

Cliff, built by Virginia Tech students, got off to a good start in the preliminaries but had brake trouble that prevented it from finishing the race.

Ohio State University turned an Oshkosh MTVER/Model MK23 into an unmanned robot named Terra Max. To ensure it could make it over the path set up by Darpa, students had to remove the cab and shorten the exhaust stack.

At the start of last year's crosscountry Grand Challenge race sponsored by Darpa (Defense Advanced Research Projects Agency), 15 unmanned vehicles stood ready to make their way past 200 waypoints and the finish line. They were all designed to travel 140 miles across rough desert terrain while navigating through GPS checkpoints and avoiding a myriad of natural obstacles like lakes, cliffs, and boulders, as well as manmade structures like fences and drainage ditches. And they would do it without human intervention.

There was an array of competitors, everything from modified sport-utes, four-wheelers and six-wheelers, to a small motorcycle and 32,000-lb truck. They were built by teams from high schools, colleges, companies, and groups of inventors and tinkerers. Although over 100 teams had applied, only 23 were chosen to compete, and only 15 of those made it past the final cut to the starting line.

One of the vehicles, dubbed Cliff, hailed from Virginia Tech, and was built by a team headed by Charles Reinholtz, professor of mechanical engineering. They became involved after winning last year's Intelligent Ground Vehicle Competition (IGC) in Michigan. The faculty and team viewed Darpa's Grand Challenge as an extension of their efforts in the IGVC and another opportunity for an interesting and challenging senior design project.

The team started with a utility vehicle made by Club Car, a div. of Ingersoll-Rand. It had some of the characteristics the team needed to compete in the rugged terrain of the Mojave Desert. For example, it sported 23x10.5/12 heavy-duty all-terrain tires and shift-on-the-fly, electronically activated four-wheel drive. Conventional rack-and-pinion front steering made Cliff very maneuverable and gave it a turning radius of about 10 ft, a major asset for avoiding obstacles.

Autonomous steering was handled by a Bodine permanent-magnet right-angle gearmotor and an encoder connected directly to the pinion through a universal joint and spider coupling. The vehicle was also equipped with independent A-arms and coil-over shocks in front and a semi-independent rear suspension with leaf springs and coil-over shocks.

There were other factors that made Cliff a good choice, such as its lightweight powerplant: a 20-hp Honda GX620 four-stroke gas engine. It was mated to a continuously variable transmission that ranged from 6:1 to 1:1. The combination yielded good mileage and enough power to get the vehicle up to 35 mph. Hydraulic disc brakes on all four tires provided stopping power and an independent pneumatic-brake system was used as a fail-safe backup.

Overall, the vehicle was light, weighing in at about 1,250 lb. It was also relatively sturdy and powerful, capable of carrying a 1,000-lb payload. This was important because all the onboard computers, sensors, and other electrical equipment would be run off a heavy, 2,800-W, gaspowered generator. Adding to the weight would be a 20-gallon, high-density, polyethylene gas tank holding fuel for the vehicle and a 6-gallon tank for the generator.

Although fairly tall at about 8 ft, including roll cage and GPS antenna, Cliff is only 4.6-ft wide and 10.5-ft long, much smaller than the converted SUVs that had entered the race.

Cliff also carried some sophisticated hardware. The computers had to take 200 waypoints along with assigned widths and speed limits released by Darpa just 2 hr before the race started and chart a safe course across the desert. The computing systems consisted of a global mapping computer, a local mapping/path-planning computer, and a system-status/motion-control computer, four main navigation systems (radar, laser rangefinder, differential GPS/INS, and infrared camera) and four sensor interfaces and associated RS-232 connectors and cables.

From the interfaces, information went to a dedicated status/motion-control computer using RJ45 connectors and Ethernet. Cliff also used circular DIN connectors (donated by Lumberg Inc.) to link the sensor suite, motor controls, and drive-by-wire hardware to the computer and controllers inside an environmentally isolated compartment. This was necessary for several reasons. Because of the density of electronic equipment on board, the DIN connectors' 360° metal housings had to shield out any potential electromagnetic interference. Plus, the connector provides an IP68 rating, which means it's watertight and effective in sealing out dust, an important factor in an open vehicle driving across the Mojave Desert.

Cliff had some minor problems making it through the qualifying rounds. For example, it started off by running into a barrier it was supposed to avoid. The team quickly found and repaired the glitch, an inadvertent mistake in the longitude inputs. Cliff, in keeping with the programming, had blithely set off for the first way-point, which, according to the original inputs, was slightly west of Nanjing, China.

Two hours before the race began, when Darpa revealed the course, Virginia Tech discovered its global-mapping computer could not provide the resolution needed to detect obstacles on the course. This meant the mapping computer could only be used to maintain the bigger picture. It had to be used with sensors reading local terrain data.

For navigation, the Virginia team used a Honeywell TALIN DGPS/INS system for global-position information with a forward-looking IR camera, an Eaton VORAD radar, and three Sick laser rangefinders, each with a 100° field of view. During the competition, only the laser rangefinders were needed. Two rangefinders mounted atop the roll cage and angled down gathered data used to create a 3D map of the terrain. The third was fixed to the brush guard and aimed horizontally, and was used for obstacle avoidance.

Rangefinder data was processed by the local mapping and path-planning computer, a National Instruments PXI 8186 Controller housed in a PXI-1002 Chassis. The controller used LabView real-time software to create a local map and find the most desirable path. It transferred this data to the system status and motion-control computer, which was responsible for controlling the throttle, brake, and steering motors. But while onboard computers can take the 200 waypoints and plot the best path through the course, sensors determine what obstacles the vehicle needs to avoid.

In the end, no vehicle made it to the finish line. In fact, the farthest any contestant got was 7.4 miles, which was accomplished by the Carnegie Mellon team with a converted Humvee named Sandstorm. Cliff came to an abrupt stop about 100 yards into the course due to a "control conflict," according to Reinholtz. More specifically, the Virginia Team had adjusted the brakes to compensate for the cold desert nights and ended up overcompensating, which led to the failure.

"Despite the fact no vehicle finished the course, we fully expect someone to do it in the future," says Goodwin. "Whether that will be in the next race remains to be seen, and we do expect dramatic technological improvements for the next race."


Last March, Darpa (Defense Advanced Research Projects Agency, a research arm of the DoD), invested $13 million in The Grand Challenge, a cross-country race between unmanned vehicles. Starting in Barstow, Calif., and finishing over 140 miles away in Primm, Nev., the first vehicles to cross the finish line first in less than 10 hr would collect the first-place prize of $1 million. Thomas Goodwin, a spokesperson for The Grand Challenge, defined the race as "a field test of autonomous ground vehicles over rugged terrain for the purpose of advancing robotics technology that will assist U.S. armed forces on the battlefield and save lives."

Darpa sponsors the race to encourage the academic and private sector to develop technologies that will let robots replace battlefield soldiers, thus saving lives. Although that seems like a lofty goal, it's also urgent, because DoD wants this kind of combat readiness fielded by 2015. Congress has mandated that at least one-third of all military vehicles be manless by then.


Darpa has already announced that this year's Grand Challenge is scheduled for October 8. The race should take only one day, but if it is not finished within daylight hours, they will shut down vehicles on the course and let them resume racing the following morning. Another new twist will be obstacles intentionally set out on the course by Darpa. They will be rigging the already challenging and rugged course with "tank traps," large, steel X-shaped structures designed to stop tanks. But as extra incentive for this year's race, Darpa is doubling the prize money to $2 million.

Darpa recently reported that through October of last year, they had received 79 applications for this year's race including three from high-school teams, 10 teams from American colleges and universities, and 33 teams that had applied for last year's Grand Challenge.

Virginia Tech, meanwhile, is going to retire Cliff and enter a newcomer to represent them in the next Grand Challenge race, a robotic vehicle called Rocky. Cliff will live on in spirit, even though he is retiring, because the team will be building Rocky on the same type of utility vehicle. And what worked well with Cliff will be incorporated into Rocky. Of course, some changes are being considered. For example, Rocky might use more wireless technology. And the Virginia Tech team will be changing the motor-control scheme and the sensing suite.

Sponsored Recommendations

The Digital Thread: End-to-End Data-Driven Manufacturing

May 1, 2024
Creating a Digital Thread by harnessing end-to-end manufacturing data is providing unprecedented opportunities to create efficiencies in the world of manufacturing.

Medical Device Manufacturing and Biocompatible Materials

May 1, 2024
Learn about the critical importance of biocompatible materials in medical device manufacturing, emphasizing the stringent regulations and complex considerations involved in ensuring...

VICIS Case Study

May 1, 2024
The team at VICIS turned to SyBridge and Carbon in order to design and manufacture protective helmet pads, leveraging the digitization and customization expertise of Toolkit3D...

What's Next for Additive Manufacturing?

May 1, 2024
From larger, faster 3D printers to more sustainable materials, discover several of the top additive manufacturing trends for 2023 and beyond.

Voice your opinion!

To join the conversation, and become an exclusive member of Machine Design, create an account today!