In October 2005 teams watched their robots attempt to navigate the rugged Mojave Desert as part of a challenge sponsored by the Defense Advanced Research Projects Agency (DARPA). The previous year's challenge had ground to a halt when none of the competitors completed more than 5 percent of the 150-mile race. But last year everything changed. Four robots finished the race in fewer than 10 hours, and the winning Stanford Racing Team's robot, fondly named Stanley, clocked speeds as high as 38 miles per hour. This dramatic turn of fortune can be attributed to advances in software and sensors.
While onboard laser and radar systems scanned the terrain, machine-learning algorithms tracked and studied the images, allowing Stanley, a modified Volkswagen Touareg, to swerve around obstacles and negotiate turns. Probabilistic methods for analyzing the road ahead kept Stanley from a common pitfall for robotic vehicles: hallucinating imaginary obstacles.
While Stanley may have a human name, the two-legged robot RABBIT has a disarmingly human gait (left). Jessy W. Grizzle, a control theorist at the University of Michigan at Ann Arbor, has tested his new mathematical model of walking and running on RABBIT, whose lower legs taper to wheels rather than feet. Because this robot is not able to statically balance on one leg, the model incorporates the effects of gravity more fully than other models. As scientists endeavor to automate more human tasks, robots may exhibit pleasing form as well as function.