Information Technology Reference
In-Depth Information
influencing these patterns. A personality profile representing the five personality traits
of the FFM is part of each simulated agent. To increase credibility, we proposed a
general model for mapping real personality profiles and studies to a consistent form
and number range in [2]. This model was used to map findings from a study in [3],
which linked personalities to general driving behavior, to our agent profiles. To inte-
grate a mapping into an application, task dependent parameters are derived from the
personality to influence the agents' decisions. This will yield consistent individual
behavior patterns for each agent.
Although, this recurring consistent behavior is desirable, it becomes implausible if
an agent is stuck in a certain situation due to its personality derived behavior. There-
fore, it is necessary that agents are able to adapt. Thus, the static personality-based
model is extended with a dynamic emotional state.
1
Emotions are integrated into the
agent architectures in three parts: experiencing, influencing, and fading. Experiencing
has been modeled through predefined incidents. When an agent experiences an inci-
dent, the negative and positive emotion of its emotional state are adjusted according
to the type of incident. Studies (e.g., [4]) have shown correlations between personality
of subjects and their perception of emotions. Thus, the perception depends on the
emotional state and personality of an agent. In our case, high neuroticism will intensi-
fy negative emotions and high extraversion or conscientiousness will intensify posi-
tive emotions [2].The influence of the emotional state is modeled as a temporary
modification of personality profiles [2]. In general, the personality of a person is a
constant attribute over extended periods of time. However, by modeling the influence
of emotions as a change of the personality, it can be added or removed without having
to change the existing system. Exponential and linear functions control how the influ-
ence of each emotion dimension fades over time. For details see [2].
3
Results
As a proof of concept the presented agent model was applied to a specific traffic sce-
nario in which the lane of an agent is blocked by an obstacle (e.g., a delivery truck).
To clear it, the agent needs to change onto an adjacent opposing lane with dense traf-
fic. The agent could pass the obstacle by waiting for an appropriate gap to not influ-
ence oncoming traffic (1), by forcing oncoming traffic to slow down or stop (2), or by
waiting for an oncoming agent to slow down for it (3). In reality, most drivers would
choose case (1). However, with increasing waiting time, drivers will accept increa-
singly smaller gaps and at some point may choose option (2). To simulate this beha-
vior, the lane change (LC) model MOBIL [5] was added to our agents. It includes a
politeness factor to control the aggressiveness of agents in LC decisions, which is
derived from an agent's personality in our case. We adapted MOBIL to include
changes to opposing lanes by weighting the advantage which agents gain from a LC
with their complementary politeness.
The test scenario was simulated with static personality-based drivers (PB) and dy-
namic emotion-based drivers (EB). In a first step, traffic flows were measured for the
1
Mood is not considered here, because it persists for prolonged time periods and agents in our
application are only observable for a limited time.
