Information Technology Reference
In-Depth Information
which were found to be consistent with behavior as observed in real soccer games.
For instance, players that had scored a number of goals were trusted more by their
teammates, and received more passes from them. Also, players with negative
emotions committed more fouls than other agents.
To further evaluate the application, 10 participants (experienced video gamers
between 25 and 32 years old) were asked to judge the behavior of the virtual players,
by answering questions like “do you think the players behave in a realistic manner?”,
“do you think the players show appropriate emotions?”. The initial results of this
evaluation were promising: the participants very much appreciated the players'
abilities to show emotions and trust. Overall, they had the idea that the presented
model made the agents more believable, and that it enhanced the experienced fun
when watching the soccer games. Nevertheless, in a later stage, a more elaborated
evaluation experiment will be performed, in cooperation with colleagues from Social
Sciences. In this experiment, we plan to compare virtual players that use different
variants of the presented model with players that do not have emotions and trust.
5 Discussion
To enhance believability of virtual soccer players' affective behavior, this paper
presented a generic model for decision making of virtual agents in relation to
emotions and trust. The backbone of the presented model is the BDI framework,
which was extended with mechanisms to represent the dynamics of emotions and
trust. After testing the model by means of simulation experiments, it has been
incorporated into the virtual agents within the RoboCup 2D soccer environment.
Although preliminary, an initial evaluation pointed out that the model has the
potential to enhance believability of virtual soccer players.
Related work in the area of believable virtual soccer players is scarce. Since the
initiation of the RoboCup competition in 1997, a number of participants focused on
the cognitive aspects of the players (e.g., [3, 11]), but emotional aspects were mostly
ignored. A welcome exception is presented by Willis [16], who proposes an
architecture to endow the robots in the physical league with emotional intelligence.
Unlike the model put forward in the current paper, this architecture has not been
implemented yet, nor does it address the concept of trust. Finally, some papers (e.g.,
[1]) address implementation of emotions within the spectators and commentators of
virtual soccer games; however, to the best of our knowledge, these approaches have
never been applied to the player agents.
As mentioned in the introduction, endowing virtual soccer players with more
human-like behavior can be useful for several reasons (see [16] for an extensive
overview): for instance, the games become 1) more fun to watch and 2) more faithful
to reality, which makes it possible to use them as an analytical tool for coaches. A
third reason put forward in [16] is that soccer teams with emotional intelligence may
have a competitive advantage over other teams. Although this was not the main focus
of the current paper, future research may investigate how well our agents perform
against similar agents without emotions and trust.
Other future work will include a further elaboration of the model, among others, by
focusing on other emotions such as anger and fear, and the interaction between
Search WWH ::
Custom Search