Information Technology Reference
In-Depth Information
2
Agent Systems
The kernel of a particular agent-based architecture is usually based upon some abstract
model of agency. This is executed on an agent platform and captured by an agent-
programming language. A brief discussion of these key elements is essential to provide
an overview of agent-based adaptive software solutions.
2.1
Agent Platforms
Although there is usually a degree of abstraction between language and architectural
issues, a language will inevitably pose some constraints upon the underlying execu-
tion layer implementing the semantic of the language. The most commonly adopted
approach in this regard is to structure the execution layer by distinguishing between
agents and agent platforms, which then provide the functional bases upon which agents
in a MAS can operate in their environment and interact with each other. In this way,
an agent can be seen as an active software entity using the agent platform as a mid-
dleware to gain access to standardized services and infrastructure, such as life-cycle
management, inter-agent communication, directory facilitators, coordination, security
management, and mobility (migration).
Agent platforms do not only free the developer from low-level details but they also
promote a basic level of modularity in the construction of the MAS as the platform
services are re-used in each agent. By adhering to FIPA reference specifications, agent
platforms such as JADE [1] and Agent Factory (AF) [2] also guarantee an important
level of cross-platform interoperability. In addition, some platforms, such as Jade and
AF, increase interoperability by not being tightly coupled to specific programming lan-
guages.
In systems such as Jade and Jack, the agent language is defined directly in terms
of Java classes. These classes are extensions of a basic agent class and have direct
access to the platform API. In AF, this pure-java option coexists with a number of
interpreted languages, including AF-APL and AgentSpeak. These are defined by using
a core library, which provides support for generic agent interpreters and for resolution
based logic.
2.2
The BDI Agent Model
The Belief, Desire, Intention (BDI) is undoubtedly the most popular agent model, with
many implementations directly related to Rao & Georgeff's abstract BDI architecture
[13] and its Procedural Reasoning System (PRS) implementation [5].
Kinny et al. [4], describes the design of a BDI agent in terms of three components:
-
A
Belief Model
, describing the information about the environment and internal
state that an agent may hold, together with the actions it may perform.
-
A
Goal Model
, describing the desires that an agent may possibly intend, and the
events to which it can respond
-
A
Plan Model
, describing the set of plans available to the agent for the achievement
of its goals
