Environmental Engineering Reference
In-Depth Information
For the reasoning capabilities, a cognitive agent needs a representation
of itself, the part of the environment in which it operate and exists, but
also the agents it has to communicate with. Thus, the internal state given
by this world model will very likely influence the decision and current
actions of the agent. The BDI architecture by Rao and Georgeff (1992)
is the most classic example of a cognitive agent model, based on desires,
beliefs, and intentions, which will be described next.
For cognitive agents, their actions should not be seen as direct action of
the changes they perceive from the environment, but more as a result of
their reasoning on understanding the world in which they are situated.
•
Reactive agents
. For reactive agents, there is expected to be a matching rule
of action in the knowledge base for each of the inputs it perceives that will
lead to actions of the agent. Actions are a direct reaction on the inputs of an
agent. A reactive agent usually has no internal world model, as its actions
are fully described by rules or functions of the input. The knowledge base
for reactive agents could be a rule base known from expert systems, where
conditional rules map to a specific output, or physical or biological inspired
functions, such as physical forces — which again impact the environment
or other agents. Therefore, reactive agents are less social, and have only
little or no direct communication with other agents. Their communication
is more indirect through the environment, such as the concept of stigmercy
in swarm intelligence (Valckenaers et al. 2002).
8
AGENT ARCHITECTURES
The agent architecture is closely related to the agent type, as the architecture in
the internal organization of the agent, which describe how it reasons and reacts
to perceived input. The architecture presents a design model of the agent, where
the flow of information from input to actions are explicitly defined through basic
concepts of the agent, such as perception, goals, and desires.
A number of different architectures have been proposed for agents, and they are
often associated with the type of agents that participate in the systems. One of the
best examples of an architecture that support the reactive behavior of the agent is
the subsumption architecture by Brooks (1986). The principle of the architecture
is that an agent has a set of accomplishing behaviors, arranged in a subsumption
hierarchy. Each of the behavior maps a given set of input values directly to an
output value that affect the actuators of the agent. The behaviors in the hierarchy
is arranged, so lower layers represent low-level behaviors and are prioritized over
higher layers that represent more abstract behaviors. The subsumption architec-
ture has been found useful for controlling robots and other AGVs, where lower
layers represent the basic high prioritized tasks, such as obstacle avoidance, while
the upper layers focus on the general goals of the robots, such as going from A
to B or exploring an environment. The famous boid model of Craig Reynolds
Search WWH ::
Custom Search