Page 156 - Modern Robotics Building Versatile Macines

136 Modern Robotics

out the boundaries between the road and surrounding terrain. It
gradually improved its ability to stay on the road at faster speeds, up
to about 20 miles (32 km) per hour.
Navlab 2, built in 1990 and converted from a military Humvee,
introduced a new navigation system. A neural net was “trained” to
drive by being shown simulated scenes or video footage from human
road trips.
In 1995, Navlab 5 drove across the country from Washington,
D.C., to San Diego, California, at an average speed of over 62
miles (100 km) per hour. By now all the computing power needed
could be provided by an ordinary laptop computer. (The need for
computation was also reduced by having an extensive library of
road types and vectors representing angles and curves in the road.)
The accompanying human driver had to take control less than 2
percent of the time.
A tougher driving challenge for robots has been provided by the
annual races sponsored by the Defense Advanced Research Projects
Agency (DARPA). The 132-mile course is a twisting path through
Mojave desert and mountain passes similar to the terrain that might
be encountered by military vehicles in Afghanistan or Iraq.
In the 2004 race, none of the robotic contestants managed to
finish the course, but in October 2005, five of the 23 autonomous
vehicles reached the finish line. The winner, with a time of six hours
and 53 minutes, was “Stanley,” a Volkswagen SUV modified by a
Stanford University team with an array of lasers, cameras, and other
sensors controlled by an onboard computer.

Moore’s Law and the Quest
for Robot Intelligence

Moravec is not only a leading robotics researcher but also a writer
whose popular books Mind Children: The Future of Robot and Human
Intelligence (1988) and Robot: Mere Machine to Transcendent Mind
(1999) offer a provocative look at the possible future of robotics. To
understand Moravec’s predictions, it is first necessary to look at how
computing power has increased over time—and when it may reach the
point where robots transcend human capabilities.