The driverless vehicles churning up the dust as they attempted to speed the 142 miles from Barstow, CA to Primm, NV across the desert this past March were not what seemed like another episode of a robot reality television show. In fact these autonomous vehicles were contenders in DARPA's Grand Challenge (http://rbi.ims.ca/XXXX-XXX deep link is www.darpa.mil/grandchallenge). That program tests the abilities of students and engineers to design vehicles that can maneuver through a Mojave Desert obstacle course in less than 10 hours, with the winner taking home a $1 million cash prize.

The task
What such teams were up against and how they surmounted the design challenges is typified by "The Blue Team," assembled by graduate student Anthony Levandowski at the University of California at Berkeley. Instead of modifying a car or truck, Levandowski's group chose a motorcycle, an inherently unstable vehicle. They reasoned a motorcycle would better remain within the race's narrow course since only one "axis" needed to conform to terrain. But as if desert conditions aren't bad enough, the trek included an underpass, an overpass, and water hazards. Levandowski funded most of the project out of pocket, although sponsors such as Raytheon, National Instruments, and Vicor chipped in.

The team used the Virtual Engineering Environment (VEE) from Agilent Technologies (Palo Alto, CA) to develop control software. According to Levandowski, "Coding in VEE makes it easy to modify routines and add modules as needed." The team has moved some control software to C, though, because of its speed advantage over VEE.

First efforts used VEE to simulate the motorcycle, named Ghost Rider (http://rbi.ims.ca/XXXX-XXX, deep link is www.ghostriderrobot.com), and the controllers. Programmers could observe the program pass data and control between various function blocks. That high-level view of operations simplified testing and debugging.

How it works
The autonomous motorcycle relies on a GPS receiver, a MEMS gyroscope, radar, several cameras, and other sensors and actuators to put it at the right place at the right time. "We use fast algorithms to control steering and to keep the vehicle upright," notes Levandowski. To stop and stay upright, the front wheel turns 90 degrees and the front wheel motor "rocks" the motorcycle in a direction opposite a pending tip over.

Sensors also help the motorcycle avoid obstacles. "Say a camera sees a big rock, but the millimeter radar sees nothing," Levandowski says. "Does the PC choose a conservative approach and guide the vehicle around the 'rock,' or does it just continue on? That's not an easy decision to make." This sort of situation highlights the need for good software and careful programming.

The motorcycle and other robots receive their racecourse coordinates-GPS readings-only two hours before the race. The vehicles then must operate on their own. Teams cannot guide vehicles by remote control.

In early versions, the Ghost Rider proceeded for a short distance and then lost control. Some problems centered on the mechanics. "When we built the full-scale motorcycle, we purchased a motor for $250, but we should have bought a motor that has better specifications," says Levandowski. "That motor would have cost about $3,000, but it would have been worth it. It's difficult to keep the inexpensive motor in control." Specifically he was referring to the use of neodymium magnets in the steering dc servo motor, which do not demagnetize under shock loads encountered by the bike. So, cautions Levandowski, don't skimp on hardware in a critical portion of a design.

The electronics portion of the Ghost Rider relies on an embedded PC with peripheral devices communicating through standard USB- and serial-port connections. According to Levandowski, two additional lessons relate to software. First, the visual environment provided by VEE greatly simplified program development and it let programmers quickly make changes during field tests. Second, the use of standard interfaces reduced software overhead.

Mixed Results
Did the Ghost Rider reach the finish line? No-but neither did any other robot claim the prize. During launch, a less-experienced crew member failed to activate the guidance system properly and the bike crashed beyond repair. The most successful vehicle went through many turns to 7.4 miles before hanging up on a berm in mountainous switchbacks.

"The bar was high," says DARPA Spokesman Jan Walker. "It was the first opportunity for such vehicles to 'get out of the company parking lot' and into a trail environment that DARPA is interested in." The agency is planning the next Challenge to take place at least a year after the first and has sweetened the pot to a $2 million prize.