Page 112 - Introduction to AI Robotics

3.5 Schema Theory
Releaser 95
appearance of
small, moving object

Perceptual Motor
Schema Schema
toad’s vision toad’s legs
Get coordinates Turn to
of small, moving coordinates
Sensory object of small, Pattern of
moving object
Input Motor Actions

Feeding Behavior

Figure 3.9 Toad’s feeding behavior represented as a behavior with schema theory.

havior exists. According to the vector field implementation of the schema
model, the toad now receives two vectors, instead of one. What to do? Well,
rana computatrix summed the two vectors, resulting in a third vector in be-
tween the original two! The toad snaps at neither fly, but in the middle. The
unexpected interaction of the two independent instances probably isn’t that
much of a disadvantage for a toad, because if there are two flies in such close
proximity, eventually one of them will come back into range.
This example illustrates many important lessons for robotics. First, it val-
idates the idea of a computational theory, where functionality in an animal
and a computer can be equivalent. The concept of behaviors is Level 1 of the
computational theory, schema theory (especially the perceptual and motor
schemas) expresses Level 2, and Level 3 is the vector field implementation
of the motor action. It shows the property of emergent behavior, where the
agent appears to do something fairly complex, but is really just the result of
interaction between simple modules. The example also shows how behav-
iors correspond to object-orienting programming principles.
Another desirable aspect of schema theory is that it supports reflex be-
haviors. Recall that in reflex behaviors, the strength of the response is pro-
portional to the strength of the stimulus. In schema theory, the perceptual
schema is permitted to pass both the percept and a gain to the motor schema.
The motor schema can use the gain to compute a magnitude on the output
action. This is an example of how a particular schema can be tailored for a
behavior.