Content uploaded by Tom Giraud
Author content
All content in this area was uploaded by Tom Giraud on Mar 17, 2014
Content may be subject to copyright.
“It” + “I”: Virtual Embodiments as Hybrid Experiences
T. Giraud
1
, A. Paljic
2
and L. Leroy
3
1
LIMSI-CNRS, France
2
Ecoles des Mines, Paritech, France
3
Université Paris 8, France
Abstract
A dichotomy exists in the way virtual embodiments are currently studied: embodied entities are considered by
conversational approaches as other selves whereas avatar approaches study them as users’ hosts. Virtual
reality applications such as in our case study often propose a different, in between embodiment experience. In
the context of a virtual house for sale visit, this paper aims at examining the user’s self-reported embodiment
perception resulting from such a hybrid experience. To induce variability in this embodiment experience, we
manipulated avatar representations (high versus low anthropomorphism) and frame of reference (egocentric
versus exocentric). Results show the importance of the entity humanness to foster both experiences. When
controlled by humanness, having a conversational experience appears uncorrelated to an avatar experience.
This highlights the need to study these hybrid experiences as a combination of both approaches.
Categories and Subject Descriptors (according to ACM CCS): I3.7 [Computer Graphics]: Three-
Dimensional Graphics and Realism - Virtual reality / Animation.
1. Introduction
To embody is the act of giving a body to an agent, a person or
a system. Use of embodiments in interactive applications has
a long history in Virtual Reality. But as one tries to trace it,
two distinct approaches stand out despite early unified
frameworks [Tha96]. Works on autonomous virtual agents
represent a first line of research [CSX04]. In this perspective,
embodiment aims to provide a natural interface with agent
functionalities by using human-human interaction routines
[CSX04]. The other line of research deals with user
embodiment, for example in collaborative interfaces
[BBF*95]. These incarnations of user presence in virtual
environments are called Avatars. One line of research
questions how users can communicate through non verbal
behaviours [GCP*99]. This academic dichotomy seems
pertinent as it appears transposable from everyday life: we are
used to own our body, and interact with others’ bodies. But
with embodied entities, combinations of those experiences
are possible. As an example, many computer games propose
gameplaies combining autonomous and controlled aspects.
The focus of this paper is on exploring the user’s embodiment
experience resulting from the interaction with a hybrid
embodied entity. This bodily entity is a parallel combination
of user controlled aspects and autonomous behaviours. An
experiment is presented to evaluate the hybrid construction of
the embodiment perception. To induce variability in this
embodiment perception, we manipulated avatar
representations (high versus low anthropomorphism) and
frame of reference (egocentric versus exocentric).
2. Background on the two paradigms
2.1 Conversational approach
Embodied entities used as autonomous partners are called
Embodied Conversational Agents (ECAs). Today, numerous
approaches exist to design and evaluate ECAs [RP04]. In this
paper, we focus on the user perception of the agent
embodiment. For conversational agent, this experience is
firstly characterized by its naturalness. As they are designed
to benefit from human to human interaction routines,
naturalness is often linked to anthropomorphism. For
example, users’ social responses increase with human-
likeness computer representations [Gon08]. Biological
movements are also crucial for anthropomorphism [DMT12].
To complete the ECA believability, it should respond in a
social coherent manner in the task context. These are the form
and behavioural aspects of ECA realism [BYMS06]. A social
presence feeling results from this overall conversational
naturalness. Biocca and colleagues defined it as “the feeling
of being with another” [BHB03]. It induces the automatic
generation of models of the intentionality of others. In a
larger perspective, Vugt and colleagues propose a model
which describes users’ engagement as a result of aesthetics,
ethics, realism, similarity, relevance and valence [VKH*07].
2.2 Avatar approach
Avatars are digital body representations controlled by the
user. They personify presence, location, identity, activity, etc
[BBF*95]. This experience of controlling an embodied entity
has been studied in different ways. For example the Proteus
effect is described as a user behaviour adaptation to conform
to the avatar persona [YB09]. Users will express more
dominance if acting through a tall avatar. Such influence is
mediated by immersion variables such as presence [FBB09].
Also, Mohler and colleagues observe that distance judgments
are more accurate when acting through an avatar body
[MCTB10]. Displacements synchrony, body movements’
synchrony and body representation increased accuracy
independently. Extreme embodiment experience can lead to
body ownership experience. Initially done with a rubber
hand, the experience of virtual body ownership has been
replicated on whole body ownership [PE08]. This was done
by altering the normal association between touch and its
visual correlate. A recent work by Kalckert and Ehrsson
make clear the distinction between ownership and agency
[KE12]. Overall, body representation, sensory and action
synchrony as well as person perspective contribute to
different components of the embodiment experience.
2.3 Hybrid works
These two paradigms are well defined and their research field
very active. Works on embodiment take one or the other
perspective, but rarely cross them. Few studies deal with this
question of hybridization. Gerhard and colleagues propose a
hybrid avatar/agent model to enable a continuous
embodiment presence [GMH04]. This model is sequential:
autonomous agents are used when users are away. As a
sequential approach, the hybrid embodiment experience is not
questioned. To our knowledge, no work on the study of this
possible parallel combination of experiences exists.
3. Experiment
This study aims at exploring the embodiment experience
from both paradigms in a use case which includes some
hybrid embodiment aspects. Those aspects of conversational
and avatar experiences are present at the same time inducing
a possible mixed experience.
3.1 Application and materials
The experiment takes place in the context of a specific use
case proposed by the industrial Saint Gobain Recherche: the
exploration of thermal comfort in a virtual house. A character
is used to represent the comfort by its autonomous reactions
to the environment. The user controls the character
displacements. This combination of controlled and
autonomous aspects makes this use case relevant for our
research. The application is composed of several components:
•
A virtual house with a living room meshed by a set of
temperature data. Temperatures result from a simulation
to ensure their realism (e.g. colder near a window).
•
A 3D virtual character which reacts to the environment.
He has five waiting/idle behaviours triggered depending
on the location: freezing idle if on cold temperatures to
sweating idle if on hot temperatures.
•
An immersive setup: users stand in front of a wall
(2m*3m) with active stereoscopy and head tracking
(Figure 1 (a)).
•
A directional control of character displacements: users
point with a Wiimote at a 2D mini-map of the sitting
room to direct the character to a position (Figure 1 (d)).
3.2 Procedure and evaluation
To vary the embodiment experience, we designed four
conditions (2*2 between subjects): Human (high
anthropomorphic) / Sphere (low anthropomorphic) (Figure 1.
(b)) * Egocentric reference frame / Exocentric reference
frame (Figure 1. (c)) (Hu-Sp / Eg-Ex). In the egocentric
reference frame condition, the user viewpoint is not
collocated with the avatar body (1 meter behind) but is
tethered to its displacements. Each participant is explained
the use case scenario of the task: he visits a house for sale and
has to explore the living room thanks to the 3D character in
order to make an idea of its thermal comfort. Before starting
the task, participants are invited to test the immersive system
on a neutral scene (outside the virtual house) without the
Figure 1. (a) User in the immersive environment. (b) Left: High anthropomorphic character, Right:Low anthropomorphic
character. (c) Top: Exocentric point of view, Down: Egocentric point of view. (d) Ray to point on the 2D mini-map to displace
the character.
character. Then the user begins the task, exploring the virtual
environment by moving the character. This exploration
session lasts 3 minutes. Once finished, participants are invited
to fill in the questionnaire with all affirmations in a
randomized order. This part takes approximately 10 minutes.
This paper focuses on the self-report analysis measures.
3.2 Perceived embodiments
Perceived embodiments refer to the two ways an embodied
entity can be traditionally perceived: as an avatar or an ECA.
Two scales are proposed to measure each type of experience.
One consequence of a conversational experience is the
simulations of “other minds” and the attribution of
intentionality. To evaluate this intentionality attribution three
items are proposed (Table 1). The resulting average is the
Intentionality variable (Int). An avatar experience results in a
better projection in place of the body in the virtual world:
better distance estimation and better immersion [LSK*08].
Three items are proposed to evaluate this projection feeling
(Table 1). The resulting average is the Projection variable
(Pro). Each item of these two dimensions is a Likert type
affirmation. Answers are ranging from 1 (I fully disagree) to
5 (I fully agree). How these two scales interact is of particular
interest.
3.3 Perceived humanness
As shown by Konijn and colleagues [VKH*07], the overall
embodiment perception is a result of concurrent intermediate
variables. In this paper, we focus on variables related to the
humanness of the embodied entity: Realism (Rea),
Anthropomorphism (Ant) and Similarity (Sim). Indeed these
variables are known to foster the experience from both
approaches [VKH*07]. Each scale is composed of four items,
and each item is a Likert type affirmation (Table 1). They are
ranging from 1 (I fully disagree) to 5 (I fully agree).
4. Results
36 subjects have realized our experiment ranging from 16 to
57 years old (M = 29, SD = 11.7) with 42% women and 58%
men. 39% of respondents were having a strong background in
informatics. All variables have normal distributions allowing
the use of parametric tests. Scale reliability is evaluated with
Cronbach’s α. All scales range from acceptable to good with
Cronbach’s α ranging from 0.61 to 0.81. Table 2 presents
Pearson and partial correlations between perceived
embodiment variables Int and Pro with Humanness. The three
variables correlate positively and significantly. When
controlled by Humanness, Int and Pro are no more
significantly correlated. Table 3. shows multivariate linear
regressions with humanness variables in predicting perceived
embodiment variables. Such analysis enables to understand
the unique influence of each predictor taken into account the
variance of the others.
A low level of multicollinearity was present among the six
predictors and the two perceived embodiment variables
(Variance Inflation Factor < 2.5). Multivariate linear
regressions are done twice: first without and then with the
three interaction terms (Rea*Ant, Rea*Sim, Ant*Sim).
Table 1. List of items and the reliability of each scale.
Scale Items
Realism
The avatar has a
natural appearance
α = 0,71
The avatar seems to be real
The avatar seems to be fake *
The avatar has an artificial appearance *
Anthropo
-
The avatar seems to be human
morphism
The avatar behave as a human being
α = 0,76
The avatar seems to be
different from a
human*
The avatar behave differently than a human
being *
Similarity
The avatar and me are alike inside
α = 0,71
The avatar and me have common
characteristics
The avatar and me are different *
The avatar and me are dissimilar */**
Humanness
The twelve items from Realism, Antrhopo
-
α = 0,80
m
orphism and Similarity
Intention
-
The avatar has intentions
ality
The avatar could want to do something
α = 0,61
The avatar thinks
Projection
The avatar reflects me
α = 0,81
The avatar represent my person
I am feeling at the avatar's place
* These items results were inverted
** This item was removed from the scale due to bad fit
Significant predictors are highlighted in Table 3. Model
1.2 explains 45,6% of the Intentionality variation and model
2.2 explains 61,4% of the Projection variation.
Table 2. Pearson correlations and partial correlations
between perceived embodiments and humanness.
Pearson
Hum
Int
Pro
Humanness
r
1
,478
*
,620
*
p
,003 ,000
Intentionality
r
1 ,492
*
p
,002
Projection
r
1
p
Partial
(controlled by Hum)
Int
Pro
Intentionality
r
1 ,284
p
,098
Table 3. Multivariate linear regressions.
Intentionality
Projection
model 1
.1
model 1.2
model
2.1
model 2
.2
R²
0,
341
0,
456
0,
458
0,
614
β
p
β
p
β
p
β
p
Rea
-,115
,46 -,083
,60 ,117
,41 ,234
,09
Ant
,425*
,02 ,375*
,04 ,165
,31 ,080
,72
Sim
,278
,11 ,294
,08 ,527*
,00 ,518*
,00
Rea
*
Ant
-,358
,12 -,036
,85
Rea*Sim
,470*
,05 ,395*
,05
Ant*Sim
-,281
,07 -,312*
,02
*. Significance at the 0.05 level
5. Discussion
In our case study, the Humanness of the embodied entity is
positively associated with Intentionality (the scale associated
with the perception of a conversational entity) as well as
Projection (the scale associated with the perception of an
avatar). Both are coherent results regarding the literature. The
positive association between Intentionality and Projection
appears to be mostly mediated by the humanness level. An
interesting finding is the absence of correlation between them
when controlled by humanness: attributing intentions to the
embodied entity is not contradictory with self projecting in it.
A question for future researches is to look at this pattern of
relations with a more extreme avatar experience, for example
by increasing the coupling between the user and the virtual
body. In this case, increasing the conversational approach by
autonomous behaviours might impair the avatar experience.
For both perceived embodiment scales, the three Humanness
subscales with their interaction terms explain a large part of
their variances in the regression analysis. This result confirms
the importance of these three dimensions as well as their
complementarities. Looking at subscales of Humanness, it
appears that both scales (Int and Pro) are differently predicted
by Humanness. Intentions attribution is associated with the
more explicitly related to Humanness scale (Ant). The
interaction term Rea*Sim shows the relative importance of
Similarity as a moderator, as shown by Vugt et al [VKH*07].
This means that giving a human shape and behaviors to the
entity alone is not a guarantee of the intentions attribution:
the perceived similarity should be high as well. The
Projection scale is associated with Similarity as well as two
interaction terms: Rea*Sim and Ant*Sim. To foster the avatar
perception, increasing the similarity level should go along
with increasing the entity realism and decreasing its human
attributes. To enable to operationalize these results as
guidelines for virtual embodiment design, the next step is to
identify the embodied entity attributes (shapes, behaviours,
controllers, etc.) influencing these self-reported dimensions.
Taken together, these results outline the necessity to
overcome the traditional dichotomy in embodiment studies.
Owning our body and seeing other’s bodies is a real life case
which is no longer the rule in immersive virtual words. This
is even true for robot in real life [RS07]. This study shows
that these two experiences are not contradictory (one is not
inhibiting the other). It highlights the need to study these
hybrid experiences as a combination of both approaches.
Acknowledgment
The authors wish to thank Saint Gobain Recherche which has
supported this research. They provided the industrial use case
as well as the 3D environment associated with. Special thanks
to Frédéric Achard who followed the project for his patience
and insightful feedbacks.
References
[BYMS06] B
AILENSON
J.N.,
Y
EE
N.,
M
ERGET
D.,
S
CHROEDER
R.:
The Effect of Behavioral Realism and Form Realism of Real-
Time Avatar Faces on Verbal Disclosure, Nonverbal Disclosure,
Emotion Recognition, and Copresence in Dyadic Interaction.
Presence: Teleoper. Virtual Environ 15, 4 (2006) 359:372.
doi:10.1162/pres.15.4.359
[BBF*95] B
ENFORD
S.,
B
OWERS
J.,
F
AHLÈN
L.E.,
G
REENHALGH
C.,
S
NOWDON
D.: User Embodiment in Collaborative Virtual
Environments. CHI'95, (1995) 242:248. Doi:
10.1145/223904.223935.
[BHB03] B
IOCCA
F.,
H
ARMS
C.,
B
URGOON
,
J.K.: Toward a More
Robust Theory and Measure of Social Presence: Review and
Suggested Criteria. Presence: Teleoper. Virtual Environ 12, 5
(2003) 456:480. doi:10.1162/105474603322761270.
[CSX04] C
ATRAMBONE
R.,
S
TASKO
J.,
X
IAO
J.: ECA as User
Interface Paradigm. In book From Brows to Trust (2004)
Springer.
[DMT12] D
ALIBARD
S.,
M
AGNENAT
-T
HALMANN
N.,
T
HALMANN
D.:
Anthropomorphism of artificial agents: a comparative survey of
expressive design and motion of virtual Characters and Social
Robots. Workshop on Autonomous Social Robots and Virtual
Humans at CASA (2012).
[FBB09] F
OX
J.,
B
AILENSON
J.,
B
INNEY
J. Virtual experiences,
physical behaviors: The effect of presence on imitation of an
eating avatar. Presence: Teleoper. Virtual Environ 18, 4 (2009)
294:303. doi:10.1162/pres.18.4.294.
[GMH04]
G
ERHARD
M.,
M
OORE
D.,
H
OBBS
D.: Embodiment and
copresence in collaborative interfaces. Journal of human-
computer studies 61, 4 (2004) 453:480.
doi:10.1016/j.ijhcs.2003.12.014.
[Gon08] G
ONG
L.: How social is social responses to computers? The
function of the degree of anthropomorphism in computer
representations. Journal Computers in Human Behavior 24, 4
(2008) 1494:1509. Doi:10.1016/j.chb.2007.05.007.
[GCP*99] G
UYE
-V
ILLÈME
A.,
C
APIN
T.K.,
P
ANDZIC
I.S.,
T
HALMANN
N.M.,
T
HALMANN
T.: Nonverbal Communication Interface for
Collaborative Virtual Environments. Virtual Reality 4, 1 (1999)
49:59. doi:10.1007/BF01434994.
[KE12]
K
ALCKERT
A.,
E
HRSSON
H.H.: Moving a Rubber Hand that
Feels Like Your Own: A Dissociation of Ownership and Agency.
Frontiers in Human Neuroscience 6, 40 (2012). doi:
10.3389/fnhum.2012.00040.
[LSK*08] L
ONGO
M.
R.,
S
CHÜÜR
F.,
K
AMMERS
M.P.,
T
SAKIRIS
M.,
H
AGGARD
P.: What is embodiment? A psychometric approach.
Cognition 107, 3 (2008). 978-98.
doi:
10.1016/j.cognition.2007.12.004
[MCTB10] M
OHLER
B.J.,
C
REEM
-R
EGEHR
S.H.,
T
HOMPSON
W.B.,
B
ÜLTHOFF
H.H.: The Effect of Viewing a Self-Avatar on
Distance Judgments in an HMD-Based Virtual Environment.
Presence: Teleoper. Virtual Environ 19, 3 (2010) 230:242.
doi:10.1162/pres.19.3.230
[PE08]
P
ETKOVA
V.I.,
E
HRSSON
H.H.: If I Were You: Perceptual
Illusion of Body Swapping. PLoS ONE 3, 12 (2008) :e3832. doi:
10.1371/journal.pone.0003832.
[RS07]
R
ICH
C.,
S
IDNER
C.L.: Robots and Avatars as Hosts,
Advisors, Companions, and Jesters. In AAAI 30, 1 (2007)
128:136.
[RP04] R
UTTKAY
Z.,
P
ELACHAUD
C.: From Brows to Trust:
Evaluating Embodied Conversational Agents (Human-Computer
Interaction Series). Springer (2004).
[Tha96] T
HALMANN
D. A New Generation of Synthetic Actors: the
Interactive Perceptive Actors, In Proc. Pacific Graphics ‘96
(1996), 200:219.
[VKH*07] V
UGT
H.C.,
K
ONIJN
E.A.,
H
OORN
J.F.,
K
EUR
I.,
E
LIENS
,
A.: Realism is not all! User engagement with task-related
interface characters. Interacting with Computers 19, 2 (2007)
267:280.
doi:10.1016/j.intcom.2006.08.005.
[YB09] Y
EE
N.,
B
AILENSON
J.N.: The Difference Between Being and
Seeing: The Relative Contribution of Self-Perception and Priming
to Behavioral Changes via Digital Self-Representation. Journal
Media Psychology 12, 2 (2009) 195:209. doi:
10.1080/15213260902849943.
