Information Technology Reference
In-Depth Information
As a single attribute b i may be arbitrary complex, the plan library
of a BDI-agent may also be represented within this knowledge base.
For simplicity and clarity, the mapping illustrated in Figure 6.11
simplifies the deliberation process and shows only the two actions
Update desires and Update intentions . It is of course possible to
break these actions down into more specific actions which are then
executed sequentially. This refinement may be applied to improve
modularisation or reusability of single components.
Furthermore, each possible action of a BDI-agent is represented by
a corresponding effector. As mentioned previously for reactive agents,
the Update intentions -action has to generate specific events in order
to trigger the desired effectors.
Reflective agent architectures
In addition to deliberative agents, reflective agents are capable of
reflecting their choices and learning from the outcomes of previous
actions. In other words, reflective agents are capable of adapting their
behavior according to the results of previous actions and their percep-
tions of the effects of specific actions. Referring to Sloman, reflective
agents are among the most sophisticated agent architectures [123].
Representing a reflective agent architecture within the GRAMS
reference model may be achieved (at least) in the following way:
Building upon the BDI-agent described previously, an agent must
additionally be capable of altering its knowledge base (including its
plan library) and to modify its sensor-effector-chains accordingly.
In the most simple case, the functionality to alter the plan library
may be part of an already existing reason -action. Instead of only
choosing the next action, an agent would also have to evaluate his
former actions and thus reflect on previous choices. Once an agent
has decided to change its behavior, the agent's sensor-effector-chains
have to be altered accordingly. As these chains are not defined in a
permanent way but represent the logical order of activating sensors
and effectors by means of (endogenous) events, this modification just
requires creating appropriate events.
Search WWH ::




Custom Search