Information Technology Reference
In-Depth Information
Roles (see 6) are responsible for specific goals and require capabilities from
the agents performing the goals that allow the agents to actually successfully
achieve the goals the role is responsible for. Roles can be specialized and can
be in conflict with each other, meaning that an agent that performs one role
might not be allowed to perform a specific other role. DomainRoles and Actors
are specialization of the more general role concept.
Figure 7 displays the Behavior aspect of the PIM4Agents metamodel. A be-
havior for an agent is specified by a plan. Each plan is composed of a trigger
event, a pre-condition, a post-condition, and a set of activities. Activities can
be complex patterns like for examples loops or simple tasks. Flows provide a
sequencing among the activities. Special tasks (i.e. BeginTask and EndTask)
mark for a diagram where the execution of the plan starts and where it ends.
The intended semantics is that when an event arises within the agent's body
that matches the trigger event, the preconditionObject is checked. If this Ob-
ject returns with true, the execution of the body (i.e. the set of activities) is
started at the BeginTask. Execution of the plan body ends when an EndTask is
encountered. It is assumed that the postconditionObject evaluates to true when
the execution of the plan body terminates.
4 Taking a Closer Look at Interactions
Regarding the design of agent interactions we take for the discussion in the
article a restricted point of view. We purely concentrate on what we call contract-
based interactions. This means that we assume that for the interactions which
we want to take into account a predefined contract exists which the agents
use when performing the interaction. The definition and analysis of interaction
protocols like the Contract Net Protocol (see below) is complex and therefore
it is beneficial if system engineers can flexibly adopt interaction protocols that
are well understood. The approach presented in this article allows to set up
repositories for model fragments (e.g. interaction protocols) and adopt them in
separate designs of multiagent systems by transforming the model fragments into
representations that can be directly included into the local design (e.g. capability
specifications of individual agents).
To make the discussion in this article not too complicate we assume that all
models are defined at design time of the system and then purely used at run time.
System dynamics is purely restricted to instance level and does not include type
level. However, this still allows dynamic assignment of agent instances to roles.
How this role assignment is actually done is out of the scope of this article. We
assume that an agent that was assigned to take a specific role on the one hand will
always try its best to fulfil the obligations the role is asking for and also provides
in principle all capabilities the role is asking for. At run time no explicit checks
are done on whether such requirements are actually met. We assume that if such
checks would be done they would be performed at type level at design time.
