Page 108 - Modern Robotics Building Versatile Macines

88 Modern Robotics

update the internal map or representation that the robot needed for
successful navigation.
Brooks began to wonder whether the computation-intensive
approach to robot vision and navigation was a dead end. After all,
as he noted in his talk “The Deep Question”:

We had a very complex mathematical approach, but that couldn’t be
what was going on with animals moving their limbs about. Look at
an insect, it can fly around and navigate with just a hundred thousand
neurons. It can’t be doing this with very complex symbolic math-
ematical computations. There must be something different going on.

While visiting his wife’s family in Thailand, Brooks found him-
self with a lot of time on his hands. He began to think about how
the human brain assimilates the data of the senses. A “classic” AI
robot tries to create a “world model” based on the incoming data
and then plan and calculate actions.
This could not be how life evolved. The simplest organisms must
have developed a way to link quickly, for example, the sensing of a
shadow, a vibration, or a rustle to appropriate behavior. The organ-
isms that could make successful connections would survive, repro-
duce, and pass the genetic blueprint for such neural circuitry on to
the next generation.
Brooks realized that as organisms evolved into more complex
forms they could not start from scratch each time they added new
features. A mouse whose eyes were being completely “rewired”
for a sharper image would likely end up inside a cat. Rather, new
connections (and ways of processing them) would be added to the
existing structure. An eye might thus be able to see motion or
contrast better, giving the mouse a better chance of surviving and
reproducing.

A “Brainless” Robot

Brooks decided to rewire his robot (which was now called Allen).
Instead of connecting it to a computer that would calculate a map