Page 159 - Modern Robotics Building Versatile Macines

RADICAL ROBOTICIST 139

ISSUES: MORAVEC V. BROOKS

Sheer processing power may not be the decisive factor in achieving
humanlike robot intelligence. The way in which processing is organized
within the robot may be just as important, as may be the possibility of
using a group of cooperating robots instead of a single machine.
Both Rodney Brooks and Hans Moravec have made impressive
achievements in developing sophisticated robots that can navi-
gate within and interact with a complex, changing environment.
Moravec’s approach focuses on visual processing and the ability to
create rich three-dimensional models. For example, Moravec sees
his cleaning machines as being intelligent and able to analyze each
room and plan efficient routes. Gradually, new attachments and
capabilities would be added so the cleaning robot could pick up and
put away objects (such as children’s toys), as well as scrub, mop, or
polish as appropriate to the surface and type of dirt.
By comparison, Rodney Brooks’s “subsumption architecture,” as
embodied in the Roomba robotic vacuum cleaner, is based on identify-
ing and implementing those behaviors that, in combination, will accom-
plish the desired task. A Roomba has no map of its surroundings; it does
not calculate optimal paths. It combines long random sweeps and cer-
tain special modes (edge-following and spiraling) that when combined
do a good job of covering the entire areas while focusing on the spots
most in need of cleaning. Where Moravec sees a single, increasingly
sophisticated cleaning robot, Brooks (and his colleagues Colin Angle
and Helen Greiner) suggests in his book Flesh and Machines that “flocks”
or entire “ecologies” of simpler, Roomba-like robots could each perform
particular chores of mopping, scrubbing, and so on.
While these two approaches to robotics seem very different, they
could well be complementary in many ways. Many simpler tasks may
be suitable for the Roomba-style robots, while others might require
a single, sophisticated robot more like what Moravec envisages. For
example, suppose the army needs to secure and make safe an area
that is likely to have mines or improvised explosive devices. A flock
of simple robots might be able to find and identify the mines and
bombs. A sophisticated robot could then go to each device, deter-
mine which type of tool or manipulator to use, and disarm it safely.
As a practical matter, many robot designs use both top-down, cen-
tralized and bottom-up, distributed approaches.