Biological Foundations of the Reactive Paradigm - Overview

Progress in robotics in the 1970’s was slow. The most influential robot was
the Stanford Cart developed by Hans Moravec, which used stereo vision to
see and avoid obstacles protruding from the ground. In the late 1970’s and
early 80’s, Michael Arbib began to investigate models of animal intelligence
from the biological and cognitive sciences in the hopes of gaining insight into
what was missing in robotics. While many roboticists had been fascinated
by animals, and many, including Moravec, had used some artifact of animal
behavior for motivation, no one approached the field with the seriousness
and dedication of Arbib.

At nearly the same time, a slender volume by Valentino Braitenberg, called
Vehicles: Experiments in Synthetic Psychology,25 appeared. It was a series of
gedanken or entirely hypothetical thought experiments, speculating as to
how machine intelligence could evolve. Braitenberg started with simple
thermal sensor-motor actuator pair (Vehicle 1) that could move faster in warm
areas and slower in cold areas. The next,more complex vehicle had two thermal
sensor-motor pairs, one on each side of the vehicle. As a result of the
differential drive effect, Vehicle 2 could turn around to go back to cold areas.
Throughout the book, each vehicle added more complexity. This layering
was intuitive and seemed to mimic the principles of evolution in primates.
Vehicles became a cult tract among roboticists, especially in Europe.
Soon a new generation of AI researchers answered the siren’s call of biological
intelligence. They began exploring the biological sciences in search
of new organizing principles and methods of obtaining intelligence. As will
be seen in the next chapter, this would lead to the Reactive Paradigm. This
chapter attempts to set the stage for the Reactive Paradigm by recapping influential
studies and discoveries, and attempting to cast them in light of how
they can contribute to robotic intelligence.
The chapter first covers animal behaviors as the fundamental primitive
for sensing and acting. Next, it covers the work of Lorenz and Tinbergen in
defining how concurrent simple animal behaviors interact to produce complex
behaviors through Innate Releasing Mechanisms (IRMs). A key aspect
of an animal behavior is that perception is needed to support the behavior.
The previous chapter on the Hierarchical Paradigm showed how early
roboticists attempted to fuse all sensing into a global world map, supplemented
with a knowledge base. This chapter covers how the work of cognitive
psychologists Ulrich Neisser109 and J.J. Gibson59 provides a foundation
for thinking about robotic perception. Gibson refuted the necessity of global
world models, a direct contradiction to the way perception was handled in
the hierarchical paradigm. Gibson’s use of affordances, also called direct perception,
is an important key to the success of the Reactive Paradigm. Later
work by Neisser attempts to define when global models are appropriate and
when an affordance is more elegant.
Many readers find the coverage ETHOLOGY on ethology (study of animal behavior) and
COGNITIVE cognitive psychology (study of how humans think and represent knowledge)
PSYCHOLOGY to be interesting, but too remote from robotics. In order to address this concern,
the chapter discusses specific principles and how they can be applied
to robotics. It also raises issues in transferring animal models of behavior
to robots. Finally, the chapter covers schema theory, an attempt in cognitive

psychology to formalize aspects of behavior. Schema theory has been used
successfully by Arbib to represent both animal and robot behavior. It is implicitly
object-oriented and so will serve as the foundation of discussions
through out the remainder of this book.