Information Technology Reference
In-Depth Information
connect a given object with the middleware, thus, we obtain a separation be-
tween the objects functionality and the middleware itself. The systems are also
responsible for object creation, startup and shutdown of communication chan-
nels and for event processing. In Section 3.3, we describe some aspects of our
deployment system in detail.
Goal-Based Interaction.
Normally, we interact with objects and devices by in-
voking single methods. That is to bring about a certain state, we perform a
sequence of actions, hopefully leading to the goal state. In [12]
goal-based inter-
action
for dynamically changing heterogeneous environments has been discussed.
Instead of invoking actions, a goal state is specified. Afterwards, some planning
system computes an appropriate sequence of actions leading from the current
to the desired state of the world. Goal-based interaction is thus a promising
approach to decouple smart environments. As mentioned above, such environ-
ments are dynamic and heterogeneous. Therefore, it is not possible to hard-wire
all possible action sequences in advance. Depending on the current state of the
world, suitable actions need to be chosen. Furthermore, the goal-based interac-
tion allows to decouple the environment, because it suces to specify a goal to
be achieved, instead of knowing the internals of the remote object, which allow
to conclude the concrete sequence of actions.
Action Description Languages.
To describe the capabilities of devices and their
methods, action description languages are used. Here we use PDDL [10] as known
from planning and shown in Ex. 3. Such operators have successfully been used
in the area of automatic planning [10] as well as in the context of smart envi-
ronments [17].
3
From Objects to Agents
In this section, we describe the agentification of objects. After discussing some
general assumptions, we extend the simple event filtering presented above to
allow filtering with respect to the current state of the whole world. As argued
below, the perceived state of the world, coincides with the beliefs of the agent.
Afterwards, we discuss goal-based interaction with objects - instead of inter-
acting by method calls, we allow a new form interaction, namely by uttering
goal states to be achieved by the object. Then, we describe how to turn simple
objects into agents using the proposed mechanisms.
As already discussed in the introduction, the following approach is based on
some assumptions: (a) Components of the system can be executed on different
hosts. (b) Every component might encapsulate different functionality. (c) Com-
ponents may enter and leave the ensemble at any point in time. (d) Interaction
with objects / components should be goal-based. (e) Legacy code should be us-
able without (severe) modifications. Based on these assumptions we designed
a system in which simple Java objects can be deployed as agents. That is, we
agentify objects using our deployment system. To furthermore ease the devel-
opment we support goal-based interaction with objects. This allows to decouple
