Basic Idea

The enhanced emotional agent $ ANIMAT^{2}$ is an extension of the agent $ ANIMAT^{1}$. While agent $ ANIMAT^{1}$ did exploit the sensory information with a fixed 'filter' looking first for food, then for an empty space and then, if nothing is available, stops moving, agent $ ANIMAT^{1}$ did not 'learn' from its experience. Agent $ ANIMAT^{2}$ is different. He is able to learn from experience by exploiting not only its sensory input but also some feedback given from its 'body' having some built-in fitness values. His 'evaluation filter' of its sensory input is modified in a way that the behavior of $ ANIMAT^{2}$ 'in the long run' will improve to find faster and faster the needed food. And -important- the agent will learn all this without a special teacher!

The enhanced agent $ ANIMAT^{2}$ is based on agent $ ANIMAT^{1}$:


$\displaystyle WORLD$ $\displaystyle \in$ $\displaystyle ENV \times AGENT$ (4.68)
$\displaystyle ENV$ $\displaystyle \subseteq$ $\displaystyle POS \times PROP$ (4.69)
$\displaystyle POS$ $\displaystyle \subseteq$ $\displaystyle X \times Y$ (4.70)
$\displaystyle ainp$ $\displaystyle :$ $\displaystyle POS \times ENV \longmapsto PERC$ (4.71)
$\displaystyle PERC$ $\displaystyle \subseteq$ $\displaystyle PROP^{n}$ (4.72)

The main extension is the additional internal state VITAL which represents the source for possible built-in fitness values.


$\displaystyle AGENT$ $\displaystyle \in$ $\displaystyle PERC \times ISTATES$ (4.73)
$\displaystyle ISTATES$ $\displaystyle \in$ $\displaystyle VITAL \times MEM$ (4.74)

Assuming that the agent has -in analogy to biological systems- some inherited prewired reactive knowledge about basic live-supporting functions signaling basic states of the system. In this case we assume only one vital state representing the state of the system with regard to its basic goals.

To be able to use the information of this built-in vital state the system needs a specific feedback function.



Subsections
Gerd Doeben-Henisch 2012-03-31