Page 111 - Modern Robotics Building Versatile Macines

THOUGHTFUL ROBOTS 91

is this young man throwing away his career?”) Scientists who
may have spent a lifetime designing and programming robots to
act intelligently did not know what to make of a robot whose
behavior seemed to emerge mysteriously from simple circuits and
subroutines.
In a way, Brooks’s robot marked a parting of ways between
artificial intelligence and artificial life (AL). AI researchers focused
on simulating cognition (reasoning), but AL researchers would
concentrate on building layers of sensation and reaction more like
that found in the nervous system. Intelligent behavior would not be
programmed so much as emerge spontaneously from the interaction
of the components.

reinforced by giving a higher value. This “trains” the system to perform
the task more efficiently. Today neural networks are used in a variety of
applications, including image processing, robot navigation, speech recog-
nition, and even credit and security screening.
The field of artificial life (AL) has extended the modeling of natu-
ral life processes to mimicking the way organic life reproduces and
evolves. An early example was John Conway’s Game of Life, a form
of “cellular automata” where patterns are manipulated by applying a
simple set of rules to each element. Researchers have made interesting
applications of this principle, for example, to simulate the behavior of
flocks of birds.
The other main thrust of artificial life is genetic programming. Here
programs are tested as to how well they can perform a task (such as
sorting). This results in a form of natural selection where the success-
ful programs have their code reproduced while the less successful are
erased. Like neural networks, genetic algorithms have shown consider-
able promise in developing applications “from the bottom up.”
Most artificial life consists entirely of software simulations inside a
computer. However, robotics researchers such as Hans Moravec at the
Stanford Artificial Intelligence Laboratory, or SAIL, have built “flocks”
of robots that use simple behaviors that, like cellular automation, can
result in complex interactions. There has even been some attempt to
couple a genetic simulation with a fabrication process to create robots
that can “evolve” in form from one generation to the next.