November 6, 2007:
For the
third time since 2004, the U.S. Department of Defense has sponsored a race for
robotic vehicles. For several decades, the U.S. Department of Defense has been
trying to build a robotic vehicle. But
in early 2004, the Department of Defense decided to try something
different, and give enterprising
civilian organizations a chance to show what they could do. DARPA (Defense
Advanced Research Projects Agency) held the DARPA Grand Challenge. Put simply,
the first robotic vehicle (moving completely under software control, with no
human intervention) that could complete a 240 kilometer course, would get a
million dollars for its designers. No one even came close. But a second
Challenge, held in late 2005, yielded several finishers, and the first one
picked up the million dollar prize for navigating a 212 kilometers cross
country course in just under seven hours. All vehicles operated under software
control, as true robots.
The third "Challenge" race was
held in late 2007, and had a two million dollar prize for the first vehicle to
complete a 60 kilometer course through an urban environment (an abandoned air
force base) in under six hours.
While much progress has been
made, the basic problem is, and always has been, that there are a lot more
obstacles for a robotic land vehicles to deal with on it's own. At sea, and in
the air, it's a much different, and much simpler, situation. Over a century
ago, naval torpedoes were built that could make sufficient adjustments, while
under way, to reach their intended target. Guided missiles came along half a
century ago and achieved the same thing in the air.
The DARPA contest has
convinced developers of robotic vehicles that they have to give their creations
a large amount of basic knowledge of obstacles, and how to deal with them, to
consistently succeed. Until now, robotic vehicles depended on TV cameras (linked
to computers that could detect traversable paths), laser rangefinders and the
like to "learn on the go." But for a
robotic vehicle to succeed, it needs some basic knowledge of the world. There
is sufficient cheap computing power available to provide that, and robotic
vehicles make use this approach. This is
also creating the kind of "knowledgeable robots" that have for so long been
popular in Science Fiction literature.
One of the goals of all this
is a robotic "infantry packbot" (a low slung vehicle that brings supplies to
infantry deep in a combat zone) with a speech recognition and voice synthesizer
module so that, when the troops wondered aloud who the packbot took so long to
get the stuff to them, the vehicle could respond, "there was a lot of mud down
the hill today and I had to go around it." Equipping a robotic vehicle with
sensors that can detect water, mud, and the depth of both, is the sort of thing
a successful "packbot" will require to survive on a battlefield. Being able to
respond to audible commands is another feature the troops have already
requested for such a vehicle. So the effort is not just to build a robotic
vehicle, but a robot in the classic sense. That's how much computing power is
required to enable a machine to go for a cross country trip over unfamiliar
terrain, and succeed. Eventually, this will lead to robotic combat vehicles.
But first, the military wants
to create a robotic truck that can move supplies over roads, or cross country,
with only a few troops supervising a dozen or more robotic vehicles. This means
you need fewer troops in the combat zone, and fewer troops will become
casualties.
The DARPA Challenge races have
been a bonanza in terms of advancing the state of the art for robotic vehicles.
For less than $10 million in prize money and expenses, the Department of
Defense has created new technology that would have otherwise cost more than
$100 million, and taken a lot longer to perfect.