A COGNITIVE MODEL FOR HUMAN BEHAVIOR SIMULATION IN
EBDI VIRTUAL HUMANS
H´ector Orozco, F´elix Ramos, Victor Fern´andez, Octavio Guti´errez
CINVESTAV del I.P.N., Unidad Guadalajara, Av. Cient´ıfica 1145, Col. El Baj´ıo 45015 Zapopan, Jal., M´exico
Marco Ramos
Universidad Aut´onoma del Estado de M´exico, Cerro de Coatepec, s/n Ciudad Universitaria 50130 Toluca, M´exico
Daniel Thalmann
´
Ecole Polytechnique F´ed´erale de Lausanne, EPFL IC ISIM VRLAB Station 14 CH-1015 Lausanne, Switzerland
Keywords:
Human behavior simulation, Virtual humans, Personality, MMPI, EBDI agents.
Abstract:
In this paper, we present a new cognitive model based on Psychology for simulating human behavior in
realistic virtual humans. To do this, we use the Minnesota Multiphasic Personality Inventory (MMPI), taking
into account the personality scales defined in it to endow the virtual humans with a real personality and form a
set of fuzzy rules used to obtain the emotional influences that modify virtual humans’ affective state according
to their personality and the events they perceive from their environment. We also implemented an EBDI-based
action selection by using an event calculus definition. This action selection mechanism allows virtual humans
to perform actions based on their current emotional state, their beliefs, their desires and their intentions. These
actions define virtual humans’ behavior for each situation they experience in the environment. As case study,
we present an scenario where a male virtual human with a psychopathic personality and a female virtual human
with a hysteric personality are interacting in a real way.
1 INTRODUCTION
As human beings, our daily life is influenced at each
time by stimuli received from the environment. These
stimuli generate emotional influences that change our
internal affective states and behavior, which give us
the ability to generate an almost immediate emotional
response to the different situations experienced in the
real world. In Psychology, one of the most studied
human characteristics that influences individuals’ be-
havior is Personality. Personality is often defined as
a set of psychological traits and mechanisms within
the individual, which are organized and relatively sta-
ble over time. These psychological traits and mech-
anisms influence individual’s interactions and adap-
tations to the environment intraphysical, social and
physical. Psychological or personality traits are de-
fined as forms of persistent patterns for perceiving,
relating and thinking about the environment and one-
self. These features distinguish a person from another,
and they are reflected in individuals’ behavior in a
wide range of contexts such as social and personal.
In this paper, we present a behavior model based
on Personality that endows virtual humans with the
ability to interact and behave in their environment ac-
cording to their emotional and mood states. In our
approach, we use the Minnesota Multiphasic Person-
ality Inventory (MMPI) (Tellegen et al., 2003) (Sell-
bom et al., 2008). This inventory is one of the most
widely used personality test in the field of Psychol-
ogy. The main objective of applying this test is to
identifying with a high degree of reliability and accu-
racy both the individual’s personality profile (person-
ality traits) and the detection of his/her possible psy-
chopathologies (emotional disorders). The 10 person-
ality scales evaluated by the MMPI are: hypochondri-
asis (Hs), depression (D), hysteria (Hy), psychopathic
deviate (Pd), masculinity-femininity (MF), paranoia
(Pa), psychasthenia (Pt), schizophrenia (Sc), hypo-
mania (Ma), and finally, social introversion (Si). We
take into account these scales to form a set of fuzzy
rules. This fuzzy set is used to obtain the emotional
104
Orozco H., Ramos F., Fernández V., Gutiérrez O., Ramos M. and Thalmann D. (2010).
A COGNITIVE MODEL FOR HUMAN BEHAVIOR SIMULATION IN EBDI VIRTUAL HUMANS.
In Proceedings of the 2nd International Conference on Agents and Artificial Intelligence - Agents, pages 104-111
DOI: 10.5220/0002738001040111
Copyright
c
SciTePress
influences that modify virtual humans’ affective state
according to their personality and the events they per-
ceive from their environment. We also implement an
EBDI-based action selection to endow the virtual hu-
mans with the ability to behave and perform actions
based on their current emotional state, beliefs, desires
and intentions. These actions define virtual humans’
behavior for each situation they experience in the en-
vironment.
The present work is organized in the following
way: Next section presents an overview of the most
important models of personality and emotion applied
to intelligent agents’ behavior. In third section we
propose a new cognitive model based on Personality
to simulate human behavior in virtual humans. Fi-
nally, in last section we will give our obtained con-
clusions from this work.
2 MODELS OF PERSONALITY
AND EMOTION APPLIED TO
BEHAVIOR OF AGENTS
Traditionally, the OCC model (Ortony et al., 1988)
has been considered as the standard model for emo-
tion synthesis and the best categorization of emotions
available. In this model emotions are interpreted as
reactions (positive or negative) to either consequence
of events, or actions of agents, or aspects of objects.
The OCC model explains human emotions and tries to
predict under certain situations, which emotions can
be investigated. Though this model is rather good, it
does not explain completely the origin of the emo-
tional processes and does not present how to filter
mixed emotions to obtain a coherent emotional state.
On the other hand, most of the proposed person-
ality models are based on trait theories, because the
conversion from trait dimensions to an efficient com-
putational model is very easy. These models con-
sist of a set of dimensions, where each dimension
represents a set of personality traits. The OCEAN
model (Costa and McCrae, 1992) is one of the most
widespread. This model groups personality traits in
ve dimensions: Openness, Conscientiousness, Ex-
traversion, Agreeableness and Neuroticism. Each di-
mension represents a set of specific personality traits
that correlate together. Although the OCEAN model
is widely accepted, it has many criticisms, because it
does not exactly indicate how Personality affects the
human behavior based on the obtained stimuli and ex-
perienced situations.
Searching for a better model of emotion, FLAME
(Fuzzy Logic Adaptive Model of Emotions) (El-Nasr
et al., 2000) is used to produce emotions and simu-
late the emotional intelligence process. This model
is based on fuzzy rules used to explore the capabil-
ity of fuzzy logic for modeling the emotional pro-
cess. These fuzzy rules are used for mapping from
events to emotions and from emotions to behaviors.
In (Liu and Lu, 2008) is presented a computer model
of motivation. This model integrates personality, mo-
tivation, emotion, behavior and stimuli together. In
spite of the fact that this model shows how motiva-
tion and Personality drive a virtual character’s emo-
tion, it only gives a primary outline for a motivation
model and it is restricted to be tested by a 3D facial
animation system. A new framework based on Artifi-
cial Intelligence for decision making is introduced in
(Iglesias and Luengo, 2007). This framework is used
to produce animations of virtual avatars evolving au-
tonomously within a 3D environment. The exposed
animations in this framework are not very realistic,
because the avatars follow a behavior pattern from the
point of view of a human observer.
A model of individual spontaneous reactions for
virtual humans is proposed in (Garc´ıa-Rojas et al.,
2008). This model was defined by analyzing real
people reacting to unexpected events. This model
presents a semantic-based methodology to compose
reactive animation sequences using inverse kinemat-
ics and key frame interpolation animation techniques.
Nevertheless, this model was created in a subjective
way in according to authors’ personal judgment. In
addition, the reaction types and animation sequences
virtual humans perform are not validated from the
point of view of psychology, but the obtained results
are satisfactory. A different mechanism to add a dy-
namic personality and personality trait openness into
agents is presented in (Ghasem-Aghaee and Oren,
2007). This work is based on the fact that personal-
ity trait openness has implications on cognitive com-
plexity and the decision making ability of agents in
problem solving. In this paper also it is implemented
a fuzzy agent to show personality descriptors, per-
sonality factors, personality style and problem solv-
ing success consequently. In addition, it is showed
a prototype system to demonstrate how personality
trait openness affects agents’ problem solving ability.
Other computational model of personality with per-
sonality change is explained in (Poznanski and Tha-
gard, 2005). This model uses a neural network for
simulating personality over time and intends to be
used as an application in a Sim-type video game. But,
for this work it is necessary to establish a set of psy-
chologically inspired rules to determine which situa-
tions change personality and in what ways.
Due to the direct correspondence between emo-
A COGNITIVE MODEL FOR HUMAN BEHAVIOR SIMULATION IN EBDI VIRTUAL HUMANS
105
tions and facial expressions (Hong, 2008), many re-
searchers prefer to employ Ekmans six basic emo-
tions (Ekman, 1994) (anger, disgust, fear, happiness,
sadness and surprise) for facial expression classifica-
tion and the OCEAN model, or else the OCC model
in combination with the OCEAN model. The mutual
dependence between emotions and personality is of-
ten represented by Bayesian belief networks (Ball and
Breese, 2000) (Kshirsagar and Magnenat-Thalmann,
2002). Another interesting approach that describes
emotions, mood, personality and their interdependen-
cies using vector algebra is proposed in (Egges et al.,
2004). Many of the revised works, which address
the use of personality in the behavior of virtual crea-
tures, make mistakes when assigning random values
to the different basic personality traits. The alloca-
tion of these values cannot be supported, because the
used theoretical framework does not make sense from
a psychological point of view, indicating that so far
none of the existing work provides an accurate and
reliable mechanism for modeling human behavior. In
this paper, our approach is more different because we
are inspired by psychological studies abouthuman be-
ing’s personality and we use the obtained results from
these studies to endow virtual humans with a realistic
personality, which allows them to behave in a more
intelligent way in their environment in according to
their affective state, beliefs, desires and intentions.
3 COGNITIVE MODEL FOR
HUMAN BEHAVIOR
SIMULATION
In order to generate a better behavior model for vir-
tual humans, in this work we apply the main ideas ex-
pressed by Paul D. MacLean in his model triune brain
(MacLean, 1973). Thus, we propose a new model of
three interrelated layers to generate behaviors influ-
enced by Personality. A brief view of the operating
cycle of our model is the following (see figure 1): the
virtual human obtains information and stimuli from
its environment through sensors. Thus, based on its
personality and its perception, it interprets them to de-
cide whether it has perceived an event that catches its
attention. Thus, once an event has been perceived,
this is processed unconsciously generating an emo-
tional response and an instinctive reaction that can be
of two types: the first one is a reflex reaction (for ex-
ample when we touch a hot object with the hand our
reaction is immediately withdraw it without having
previously thought) and the second one is an instinc-
tive reaction of protection (for example, when some-
body suddenly throws us an object, our reaction is to
avoid the blow that can cause us the object). Immedi-
ately in parallel to the generated emotional response
and instinctive reaction, the virtual human becomes
aware of the perceived event and it searches for an
explanation by looking for information from its be-
liefs and its long and short term memory. Once the
virtual human collects the necessary information, it
shows a behavior consistent with its personality. In
this way and based on its desires and intentions, the
virtual human executes actions that it deems the most
appropriate to the situation. Finally, the virtual human
is capable of evaluating the obtained results from its
exhibited behavior and learns about them.
Figure 1: Cognitive model for human behavior simulation
in EBDI virtual humans.
3.1 Fuzzy Rules for Generating an
Emotional Influence
We use a fuzzy rule-based model composed by a set
of if-then rules. This set of rules is used to gener-
ate the emotional influence that changes virtual hu-
mans’ emotional and mood states according to their
personality and the events they perceive from the en-
vironment. We use fuzzy logic, because the perceived
events, the emotional influence (a six-dimensional
vector with corresponding values for each of the six
basic emotions) and the personality have fuzzy lim-
its. This permits changing virtual humans’ emotional
and mood states in a more natural manner, gener-
ating more realistic behaviors. We use a set of in-
put variables corresponding to each personality scale,
for these scales we define ve fuzzy sets: very low,
low, medium, high, and very high. An event is repre-
sented with an input variable, which has seven defined
fuzzy sets: negative high, negative medium, nega-
tive low, neutral, positive low, positive medium, and,
positive high. The impact of an event over each emo-
tion in the emotional influence vector is described
by the following fuzzy sets negative very high,
negative high, negative medium, negative low, neg-
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
106
ative very low, neutral, positive very low, posi-
tive low, positive medium, positive high, and, posi-
tive very high.
In order to obtain the emotional influence vector,
we used jFuzzyLogic (Cingolani, 2009), which is a
java package that offers a complete fuzzy inference
system (FIS). This package implements a fuzzy con-
trol language specification according to IEC 1131-7
(IEC 1131, 1997). We defined in this language the
input and output variables and the set of fuzzy rules
used to obtain the emotional influence. For exam-
ple, some representative fuzzy rules written in FCL
for the Hy (hysteria) personality scale are: IF Hy
IS high AND event IS positive high THEN anger IS
negative high, disgust IS negative medium, fear IS
negative low, happiness IS positive low, sadness IS
negative low, surprise IS positive low and IF Hy IS
very high AND event IS negative low THEN anger IS
positive medium, disgust IS positive low, fear IS pos-
itive very low, happiness IS negative very low, sad-
ness IS positive low, surprise IS negative very low.
As example, figure 2 shows the defined fuzzy sets
for the variables Hy (hysteria), event, and anger, re-
spectively.
Figure 2: Defined fuzzy sets for the variables Hy (hysteria),
event and anger.
3.2 Affective State Update
We propose a process to update virtual humans’ emo-
tional and mood state by using the 10 personality
scales defined by the MMPI, the six Ekman’s basic
emotions (anger, disgust, fear, happiness, sadness and
surprise) and three basic moods (good, neutral and
bad). For this matter, we consider the virtual humans
as entities with a constant personality and a dynamic
behavior, which is constantly changing over time t.
Therefore, the virtual humans’ personality p is ini-
tialized with a set of constant values at time t = 0,
and their emotional and mood states, e
t
and m
t
respec-
tively, are dynamic over time and these are initialized
to 0 at time t = 0. We formalize these concepts as
follows:
p
T
= [Hs, D,Hy, Pd, MF, Pa, Pt, Sc, Ma, Si], (1)
Hs, D, Hy, Pd, MF, Pa, Pt, Sc, Ma, Si [0, 1]
Where Hs (hypochondriasis), D (depression), Hy
(hysteria), Pd (psychopathic deviate), MF (masculin-
ity/femininity), Pa (paranoia), Pt (psychasthenia), Sc
(schizophrenia), Ma (hypomania), and Si (social in-
troversion), are the 10 personality scales defined by
the MMPI.
Emotional state e
t
represents the intensities of the
six Ekman’s basic emotions at each time t. These
emotions are labeled as: an (anger), di (disgust), fe
(fear) ha (happiness), sa (sadness), and su (surprise)
in a 6-dimensional vector in the following way:
e
T
t
=
[an, di, fe, ha, sa, su] if t > 0
0 if t = 0
an, di, fe, ha, sa, su [1, 1]
(2)
In a similar way, the mood state m
t
represents the
intensities of three basic moods at each time t. These
moods are labeled as: gd (good), nl (neutral) and bd
(bad) in a 3-dimensional vector:
m
T
t
=
[gd, nl, bd],gd, nl, bd [1, 1] if t > 0
0 if t = 0
(3)
We also use an emotional history ω
t
and a mood
history σ
t
, which contain the emotional states e
0
un-
til e
t
and the mood states m
0
until m
t
, respectively.
The next step is to update the emotional and mood
states. To do this, we use an emotional influence vec-
tor a, which is obtained by applying the fuzzy rules
defined above. This vector contains a desired change
of intensity for each of the six basic emotions. Thus,
when a virtual human has assessed the emotional in-
fluence, the emotional and mood states are updated
A COGNITIVE MODEL FOR HUMAN BEHAVIOR SIMULATION IN EBDI VIRTUAL HUMANS
107
in two steps. The first step consists in updating the
emotional state. The second step consists of updating
the mood state. The emotional state is updated taking
into account the last mood as follows:
e
t+1
= e
t
+ Ψ
e
(p, σ
t
, a) +
e
(p, ω
t
, σ
t
) (4)
We define a 6× 10 Personality-Emotion Influence
Matrix P
0
(how personality influence emotions). This
matrix is defined once and it is multiplied with the
vector p to obtain a new vector u. We use this vec-
tor to construct a diagonal matrix P (how strong an
emotion can be given a personality). Thus, we com-
pute a 6× 3 Mood-Emotion Influence Matrix T (how
moods influence emotions) that is multiplied with the
current mood m
t
to obtain the mood influence on the
final emotional state. Thus, we obtain the following
definition for the function Ψ
e
:
Ψ
e
(p, σ
t
, a) = P· a+ T · m
t
(5)
Finally, the function for emotion decay
e
is de-
fined as a 6-dimensional vector. This vector con-
tains the amount of decrement or increment desired
for each of the six basic emotions. In our case, we use
0.03 to increase the intensity of an emotion or -0.03
to decrease its intensity. The mood state is updated by
a function that calculates the mood change based on
the new emotional state:
m
t+1
= m
t
+ Ψ
m
(p, ω
t+1
, σ
t
, a) +
m
(p, ω
t+1
, σ
t
)
(6)
We use a 3 × 6 Emotion-Mood Influence Matrix
Q, which defines the relation between emotions and
each mood dimension. The influence of vector a on
the mood is calculated by Q×a. Similarly to the emo-
tion update, using the personality, we now also define
a 3 × 10 Personality-Mood Influence Matrix R (how
personality influences each mood dimension). This
matrix is also defined once and it is multiplied with
the vector p to obtain a new vector v. We use this new
vector to construct a diagonal matrix R. So, we give
the following definition for the function Ψ
m
:
Ψ
m
(p, σ
t
, a) = R
0
· Q· a (7)
Finally, the function for emotion decay
m
is de-
fined as a 3-dimensional vector. This vector con-
tains the amount of increment or decrement desired
for each of the three moods. In our case, the desired
values to increase the good mood normally are 0.03
for the good value, 0.1 for the neutral value, and -0.3
for the bad value. The desired values to increase the
bad mood are -0.03 for the good value, 0.1 for the
neutral value, and 0.3 for the bad value.
3.3 EBDI-based Action Selection
We use an EBDI (Emotion-Belief-Desire-Intention)
architecture similar to those presented in (Georgeff
et al., 1999) (Jiang et al., 2007) (Pereira et al., 2008),
but we add time constraints to choose the actions a
virtual human should perform according to external
and internal events, and how these change virtual hu-
man’s affective state considering time. For this mat-
ter, we use Event Calculus, which is a temporal for-
malism that allows reasoning about events and the
time when these occur (Kowalski and Sergot, 1986).
We selected this formalism by its intuitive definition
of events. Next, we present the list of the event calcu-
lus predicates used to formalized our approach:
Initiates(e(), f, t): Fluent f holds, after event e() is
perceived at time t.
Terminates(e(), f, t): Fluent f does not hold, after
event e() is perceived at time t.
HoldsAt(f, t): Fluent f holds at time t.
Happens(e(), t): Event e() is perceived at time t.
InitiallyP(f): Fluent f holds from t = 0.
For a detailed definition of the axioms that rule
these predicates see (Shanahan, 1999).
A fluent is a variable that can change over time.
We use boolean fluents. We consider the following
three sets of boolean fluents:
1. A set of emotions, which consists of the six emo-
tions: anger (an), disgust (di), fear (fe), happiness
(ha), sadness (sa), and surprise (su);
2. A set of beliefs, which are agents’ assumptions
about the state of the world and agents possible
skills;
3. A set of desires, which are agents’ goals.
The above is formalized as follows:
E = {an, di, fe, ha, sa, su} (8)
B = {b
1
, b
2
, ··· , b
i
} (9)
D =
d
1
, d
2
, ··· , d
j
(10)
Fluents = E BD (11)
We assume that agents have a set of initial beliefs
(IB) and initial desires (ID), which are treated in the
following way:
IB B, f IB : InitiallyP( f) (12)
ID D, f ID : InitiallyP( f) (13)
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
108
The set of intentions I =
{intention
1
, intention
2
, ··· , intention
n
} represents all
the possible plans agents select in order to follow a
given course of action to achieve a goal, according to
agents’ emotional state and beliefs.
An intention is defined as follows:
intention = {Happens(e
1
(),t
1
) Happens(e
2
(),t
2
)
·· · Happens(e
m
(),t
n
) t
1
< t
2
< · ·· < t
n
} (14)
The events that happen in the virtual humans’ en-
vironment can be either internal events (IE) and ex-
ternal events (EE); internal events when they refers
to actions perform by virtual humans in the achieve-
ment of goals; and external events when the virtual
human perceives a change in the environment. These
are denoted as follows:
IE = {ie
1
(), ie
2
(), ·· · , ie
r
()} (15)
EE = {ee
1
(), ee
2
(), ·· · , ee
s
()} (16)
Events = IE EE (17)
In addition, we consider that agent’s beliefs with
respect to the actions it can perform, cannot be ac-
complished, for example, an agent believes that it can
swim in certain conditions, however when it attempts
swimming, it realizes that such belief is not truth, and
then its beliefs must be updated. This update process
is achieved in the following way:
HoldsAt(b
1
,t) HoldsAt(b
2
,t)
··· HoldsAt(b
i
,t)
Initiates(e(), d
j
,t)
Happens(e(),t) HoldsAt(emotion
1
,t)
HoldsAt(emotion
2
,t)
··· HoldsAt(emotion
k
,t) (18)
Happens(e
1
(),t
1
) Happens(e
2
(),t
2
)
··· Happens(e
m
(),t
n
)
t
1
< t
2
< ·· · < t
n
HoldsAt(d
j
,t) (19)
Happens(e
1
(),t
1
) Happens(e
2
(),t
2
)
··· Happens(e
m
(),t
n
) t
1
< t
2
< ·· · < t
n
Terminates(e
H
fails(), b
i
,t
w
)
Happens(e
H
fails(), t
w
)
HoldsAt(emotion
1
,t
w
)
HoldsAt(emotion
2
,t
w
)
··· HoldsAt(emotion
k
,t
w
) (20)
Terminates(e
m
(), d
j
,t
n
) Happens(e
m
(),t
n
) (21)
Where, emotion
k
E, b
i
B, e
m
() Events, and
d
j
D.
In equation 18, certain emotions hold and an event
is perceived at time t; then a certain desire d
j
is initi-
ated if a set of beliefs holds at the same time t. The
initiation of desire d
j
activates a set of possible course
of actions in order to achieve d
j
, this is expressed in
equation 19. Next, if there is failure in the execu-
tion of the plan (denoted by Happens(e
H
fails(), t
w
)),
then another plan is selected according to the current
predominant emotion emotion
K
(see equation 20).
Now, whenever the last action of any plan is success-
fully performed, desire d
j
is removed from agent’s de-
sires, because it was achieved (see equation 21).
4 CASE STUDY
We present as case study a situation where there are
two virtual humans, a woman and a man interacting in
a virtual environment. Woman has a hysteric person-
ality and man has a psychopathic personality. Figure
3 shows the study case. We assume that the virtual
humans have the following beliefs and desires:
Male virtual human’s beliefs: IamASilentWalker
and WomanIsDistracted.
Female virtual humans beliefs: IamCalm and
IHaveNothingToDo.
In the male virtual human side, we have the fol-
lowing plan definition:
HoldsAt(WomanIsDistracted, t)
HoldsAt(IamASilentWalker, t)
Initiates(WalkToWoman(), ScareWoman,t)
Happens(WalkToWoman(),t)
HoldsAt(happiness,t) (22)
Once the male virtual human adopts a desire, it
computes a plan to release the desire (see figure 3.a,
3.b and 3.c).
Happens(WalkToWoman, t
1
)
Happens(ScreamToWoman, t
2
) t
1
< t
2
HoldsAt(ScareWoman, t) (23)
Now, in the female virtual human side (see figure
3.a, 3.b and 3.c), the behavior is specified as follows:
A COGNITIVE MODEL FOR HUMAN BEHAVIOR SIMULATION IN EBDI VIRTUAL HUMANS
109
Figure 3: Exposed behavior by virtual humans according to
their personality and the events they perceive.
HoldsAt(IhaveNothingToDo, t)
HoldsAt(IamCalm, t)
Initiates(Wandering(), FeelTheWind, t)
Happens(Wandering(),t)
HoldsAt(Happiness,t) (24)
Once the male virtual human frustrates female vir-
tual human desires of feel the wind, the female creates
a new plan and updates some beliefs (see figure 3.d).
This is shown next:
Happens(Jump(),t
1
) Happens(Scream(),t
2
)
Happens(GetAway(), t
3
) t
1
< t
2
< t
3
Terminates(MaleScareHer(), IamCalm, t)
Happens(MaleScareHer(), t)
HoldsAt(Happiness,t) (25)
Then, the male reacts according to the events per-
ceived (see figure 3.d, 3.e and 3.f):
Happens(LaughtAtHer(), t
1
)
Happens(SayGoodByeHer(),t
2
) t
1
< t
2
< t
3
Terminates(GetAway(),WomanIsDistracted, t)
Happens(GetAway(),t)
HoldsAt(Happiness,t) (26)
This simple scenario showed how simple is to
model virtual human behavior based on EBDI by us-
ing event calculus.
5 CONCLUSIONS
In this paper, we presented a behavior model for re-
alistic virtual humans. This model is valid from a
psychological point of view, because it is supported
by studies on Personality and the resources provided
by them. With the use and interpretation of MMPI,
we implemented a real behavior model for virtual hu-
mans, because we do not simulate predefined actions,
but we consider all possible reactions for each person-
ality scale defined by MMPI, according to perceived
events from the environment. With this, virtual hu-
mans are able to behave in different ways to the same
perceived events based on their personality and affec-
tive state. We also have implemented an EBDI-based
intention selection using the Event Calculus formal-
ism. This intention selection mechanism allows vir-
tual humans to perform actions based on their current
emotional state, beliefs, desires and intentions. Thus,
these intentions are used to define virtual humans’ be-
havior for each situation they experience in the envi-
ronment according to their personality and the events
they perceive.
6 FUTURE WORK
As continuation of our work we are creating and im-
plementing a complete fuzzy model that will be used
to update virtual humans’ mood and emotional state
in a more natural and efficient way. We are work-
ing on defining different fuzzy sets and fuzzy rules
to update and regulate virtual humans’ affective state
according to their personality, emotional history and
mood history, taking into account the level of inten-
sity of perceived events from their environment. With
this new fuzzy model, we will completely replace the
matrix model that was described above. Finally, we
also are working on applying the principles of Emo-
tional Intelligence Model (Goleman, 1995) to endow
virtual humans with an emotional intelligence that al-
lows them to perceive and express emotions, assimi-
late emotion-related feelings, understand and reason
with emotion, and regulate emotions in themselves
and other virtual entities. To implement this intelli-
gence, we intent to use the Emotional Competence
Framework (Mayer et al., 2000) defined in the Emo-
tional Intelligence Model. With this framework, we
will develop virtual humans’ personal and social com-
petencies on the basis of the following characteristics:
Personal Competencies such as self-consciousness,
self-regulation and self-motivation, and Social Com-
petencies such as social awareness and social skills.
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
110
ACKNOWLEDGEMENTS
This research is partially supported by CoECyT-Jal
project no. 2008-05-97094, whilst authors H´ector
Orozco, Victor Fern´andez and Octavio Guti´errez are
supported by CONACYT grants no. 203060, 212753,
and 191493, respectively.
REFERENCES
Ball, G. and Breese, J. (2000). Emotion and personality
in a conversational agent. Embodied Conversational
Agents, pages 189–219.
Cingolani, P. (2009). jfuzzylogic: An open source
fuzzy logic library and fcl language implementation.
http://jfuzzylogic.sourceforge.net/html/index.html.
Costa, P. J. and McCrae, R. (1992). Normal personality
assessment in clinical practice: The neo personality
inventory. Psychological Assessment: A Journal of
Consulting and Clinical Psychology, 4:5–13.
Egges, A., Kshirsagar, S., and Magnenat-Thalmann, N.
(2004). Generic personality and emotion simulation
for conversational agents. Journal of Visualization
and Computer Animation, 15(1):1–13.
Ekman, P. (1994). Moods, Emotions, and Traits, the Nature
of Emotion: Fundamental Questions. Oxford Univer-
sity Press, New York, NY, USA.
El-Nasr, M., Yen, J., and Ioerger, T. (2000). Flame: Fuzzy
logic adaptive model of emotions. Autonomous Agents
and Multi-Agent Systems, 3(3):219–257.
Garc´ıa-Rojas, A., Guti´errez, M., and Thalmann, D. (2008).
Simulation of individual spontaneous reactive behav-
ior. In AAMAS ’08: Proceedings of the 7th Interna-
tional Joint Conference on Autonomous Agents and
Multiagent Systems, pages 143–150, Richland, SC.
International Foundation for Autonomous Agents and
Multiagent Systems.
Georgeff, M. P., Pell, B., Pollack, M. E., Tambe, M., and
Wooldridge, M. (1999). The belief-desire-intention
model of agency. In ATAL ’98: Proceedings of the 5th
International Workshop on Intelligent Agents V, Agent
Theories, Architectures, and Languages, pages 1–10,
London, UK. Springer-Verlag.
Ghasem-Aghaee, N. and Oren, T. (2007). Cognitive com-
plexity and dynamic personality in agent simulation.
Computers in Human Behavior, 23(6):2983–2997.
Goleman, D. (1995). Emotional Intelligence. Bantam
Books, New York, NY, USA.
Hong, J. (2008). From Rational to Emotional Agents: A
Way to Design Emotional Agents. VDM Verlag, Saar-
brucken, Germany.
IEC 1131 (1997). Programmable controllers, part 7:
Fuzzy control programming, fuzzy control language.
http://www.fuzzytech.com/binaries/ieccd1.pdf.
Iglesias, A. and Luengo, F. (2007). Ai framework for
decision modeling in behavioral animation of virtual
avatars. In ICCS ’07: Proceedings of the 7th interna-
tional conference on Computational Science, Part II,
pages 89–96, Berlin, Heidelberg. Springer-Verlag.
Jiang, H., Vidal, J. M., and Huhns, M. N. (2007). Ebdi:
An architecture for emotional agents. In AAMAS ’07:
Proceedings of the 6th International Joint Conference
on Autonomous Agents and Multiagent Systems, pages
1–3, New York, NY, USA. ACM.
Kowalski, R. and Sergot, M. (1986). A logic-based calculus
of events. New Generation Computing, 4(1):67–95.
Kshirsagar, S. and Magnenat-Thalmann, N. (2002). A mul-
tilayer personality model. In SMARTGRAPH ’02:
Proceedings of the 2nd International Symposium on
Smart Graphics, pages 107–115, New York, NY,
USA. ACM.
Liu, Z. and Lu, Y.-S. (2008). A motivation model for vir-
tual characters. In Proceedings of the Seventh Inter-
national Conference on Machine Learning and Cyber-
netics, volume 5, pages 2712 –2717.
MacLean, P. (1973). A Triune Concept of the Brain and
Behaviour. University of Toronto Press, Toronto.
Mayer, J. D., Salovey, P., and Caruso, D. R. (2000). Mod-
els of emotional intelligence. Cambridge University
Press, Cambridge, UK.
Ortony, A., Clore, G., and Collins, A. (1988). The Cognitive
Structure of Emotions. Cambridge University Press,
Cambridge, UK.
Pereira, D., Oliveira, E., and Moreira, N. (2008). Formal
modelling of emotions in bdi agents. In CLIMA ’08:
Proceedings of the 8th International Workshop on
Computational Logic in Multi-Agent Systems, pages
62–81, Berlin, Heidelberg. Springer-Verlag.
Poznanski, M. and Thagard, P. (2005). Changing personal-
ities: Towards realistic virtual characters. Journal of
Experimental and Theoretical Artificial Intelligence,
17(3):221–241.
Sellbom, M., Ben-Porath, Y., and Bagby, R. (2008). Per-
sonality and psychopathology: Mapping the mmpi-
2 restructured clinical (rc) scales onto the five factor
model of personality. Journal of Personality Disor-
ders.
Shanahan, M. (1999). The event calculus explained. LNAI:
Artificial Intelligence Today: Recent Trends and De-
velopments, 1600:409–430.
Tellegen, A., Ben-Porath, Y., Arbisi, P., Graham, J., and
Kaemmer, B. (2003). The MMPI-2 Restructured Clin-
ical Scales: Development, Validation, and Interpre-
tation. University of Minnesota Press, Minneapolis,
MN, USA.
A COGNITIVE MODEL FOR HUMAN BEHAVIOR SIMULATION IN EBDI VIRTUAL HUMANS
111