Artificial Mind for Virtual Characters - Advancing Artificial Intelligence Through Biological Process Applications - page 194

that the value d sent by an Alert Agent has to be
negative ; otherwise, it is not able to increase the
motivation intensity. So, it is not a deactivation
factor, but, on the contrary, an activating com-
ponent of the intensity equation. The i value may
decrease either if the stimulus intensity decreases
or the stimulus distance increases (by running
away from the scary stimulus, for example), both
causing the d absolute value to decrease.
Anyway, it is important to note that the
determination of d value results from an evalua-
tion made by a Cognition Agent, and not by the
Motivation Agent. Mixed Motivation Agents
can also exist, and may receive both negative
and positive values of d , each sent by a different
cognition agent. For instance, a hungry character
may be stimulated by the vision of food. There is
an internal component (in this example, a Hunger
Agent) that increases autonomously, but is affected
by environmental stimuli, which is perceived by
an Alert Agent.

prior motivations, which then pass to the active
state. Only active Motivation Agents can trigger
the cognitive processing by which Cognition
Agents perform action selection.
Competition could be just a special kind of
winner-take-all mechanism that does not restrict
the result to just a single Motivation Agent, but
selects a little set of them instead. Restricting
the competition result to just one Agent would
not allow a negotiation (as described in the next
topic) between prior motivations in the cognitive
structure. The importance of this negotiation is
that it could permit opportunism or compromise
actions, which are interesting behaviors, as already
mentioned. Besides, a single winner mechanism
overestimates little circumstantial differences
between motivation intensities. So, the chosen
mechanism establishes a proportional intensity
threshold above which all motivations are ac-
tivated. For instance, all agents included in the
competition process can send each other a message
like this for each competition cycle:

Inner Motivational Environment

IF α i k ≥ i n
THEN
ReceiverState = Inactive
// Temporary change of receiver state ,

The set of all Motivation Agents constitutes the
Inner Motivational Environment, which has no
direct contact with the Outer Environment (see
above), and represents the motivational state of
the character. Its purpose is to assure that only
motivations with high priority can trigger their
correspondent cognition agents. In other words,
this means that it is responsible for the filtering
process that decreases the computational complex-
ity of the whole action selection process.
A Motivation Agent can be in one of three
states: inactive, competitive and active. A mo-
tivation is considered inactive when it is in the
comfort zone of intensity value. This means that it
can't affect behavior because it is satisfied. From
the moment the intensity value increases to the
point of entering the tolerance zone, some action
is needed to discharge it. So, its state becomes
competitive , which means that it is now part of
a Competition process. This process selects the

where:

0 < α <1, α being an arbitrary value chosen at
design time.
i k : sender intensity value;
i n : receiver intensity value.

After sending all messages, the remaining
agents that are still in the competitive state are
set to active stat e, and are allowed to trigger their
corresponding Cognition Agents. At this point,
computation of intensities of all other Motivation
Agents resets their states based on their motiva-
tion zone.
All those Motivation Agents that have inten-
sity values above or equal α i max (where i max is