Information Technology Reference
In-Depth Information
6.9.2 Case studies
For evaluating the practical applicability of the GRAMS reference
model, three applications were analyzed. The resulting simulation
models were developed according to the GRAMS reference model. The
complexity of the modeling task was increased continuously. The first
case study considered a simple warehouse scenario and was used for
basic evaluation purposes. The second case study considered a military
scenario with increased interaction of agents and was developed jointly
with Christian Gerstner [40, 41, 42]. A third application of the
GRAMS reference model considered coordination of rescue units due
to a massive trac accident and involved a wide variety of different
agent types and complex interaction patterns. This application was
developed jointly with Daniel Weyel [142].
As a first step within the evaluation of the GRAMS reference
model, a small but typical case study was analyzed using the GRAMS
reference model. This case study considers a simplified warehouse
scenario (2-dimensional environment) including three different types of
agents (worker, forklift, warehouse manager) with limited interaction.
A detailed description of the simulation model is given in appendix A.1.
The GRAMS reference model provided a valuable guideline through-
out the development process. Regarding this case study, the GRAMS
reference model could be applied very well and did not impose any
unnecessary or limiting constraints. Regarding the experiences gained
from the first case study, two issues are worth mentioning:
First, as defined by the GRAMS reference model, all sensor and
effector actions are triggered by corresponding events. Especially
in case of sensor-effector sequences which are repeatedly executed
(e. g., a typical main loop) or do not depend on exogenous events,
introducing (possibly a multitude of) events to control behavior
of an agent can be irritating at first. Once one gets used to this
kind of modeling, the simple but strict principle - each sensor and
effector action is triggered by an event - seems quite natural and
rewards model developers with a strict interface defining precisely
input and output of an agent.
