Page 159 - Robots Androids and Animatrons : 12 Incredible Projects You Can Build

Multivalue threshold Typically, in neural networks individual neu-
rons have a singular threshold (positive or negative). Once the
threshold is exceeded, the output of the neuron is activated. In
our example the output is compared to multivalues, with the out-
put going to the best fit.
Instead of thinking of the output as numeric values, think of each
numeric range as a shape instead; a circle, square, and triangle will
suffice. When the neuron is summed, it outputs a shape block
(instead of a number). The receptor neurons (LEDs) have a
shaped receiver unit that can fit in a shape block. When a shape
block matches the receiver unit, the neuron becomes active
(LED turns on).

In our case each output neuron relates to a particular behavior,
sleeping, hunting, and feeding, all essential behaviors for survival
in a photovore-style robot. Each output shape represents the cur-
rent light level. At low light levels, the photovore stops hunting
and searching for food (light). It enters a sleep or hibernation
mode. At medium light level, the photovore hunts and searches for
the brightest light areas. At high light levels, the photovore stops
and feeds via solar cells to recharge its batteries.
138
Instead of building a photovore robot in this chapter, we will use
an LED to distinguish between each behavior state (see Fig. 6.37).
You can label the three LEDs sleeping, hunting, and feeding. Each
LED will become active depending upon the light level received by
the CdS cells.

6.37 Schematic of basic neural circuit

Team LRN
Chapter six