Information Technology Reference
In-Depth Information
Given these characteristics, reactive agents can easily be modeled
within the frame of the GRAMS reference model. The agent
A
reactive
=
(
ATT
,
SEN
,
EFF
) is specified as follows (see Figure 6.10):
ATT
denotes the agents' properties (e. g., state of the agent).
SEN
=
{
s
1
,...,s
n
}
denotes the agents' set of sensors.
EFF
denotes the agents' set of effectors. The
Control
-effector is designated to choosing an appropriate effector
e
i
depending on the current sensor input. Each sensor action consists of
processing the actual sensor input, followed by generating a
Control
-
event in order to trigger the
Control
-effector. The
Control
-effector
in turn decides which effector
e
i
has to be triggered (depending on
the sensor input).
=
{
Control
,e
1
,...,e
k
}
Defined this way, the mapping reflects very closely the
Sense-Reason-
Act
cycle used in many agent architectures (cp. [30]).
Deliberative agent architectures
In contrast to reactive agent architectures, deliberative agents are
characterized by an explicit internal world model and (sophisticated)
reasoning and planning capabilities. Perhaps the most widely used
deliberative agent architecture is the BDI-architecture. The main
components of this architecture are
B
eliefs,
D
esires and
I
ntentions of
an agent.
Beliefs
represent an agents knowledge about the world which may
be incomplete or imperfect.
Desires
represent the long-term goals of
an agent. As such they are an integral part of the deliberative behavior
and heavily influence an agents actions. Once an agent has commited
itself to a goal, it has to select the next action. Usually, the agent
is equipped with a plan library from which the plan is chosen which
fits best to its current goal. The actually chosen plan which an agent
executes is its
intention
. Simply said, desires represent long-term
goals while intentions represent short-term goals which directly lead
to a sequence of actions.
