Content uploaded by Silvestro Micera
Author content
All content in this area was uploaded by Silvestro Micera on Dec 15, 2013
Content may be subject to copyright.
Emotional Expression Humanoid Robot WE-4RII
-Evaluation of the perception of facial emotional expressions by
using fMRI-
M. Zecca1,2, T. Chaminade3, M.A. Umiltà4, K. Itoh2,5, M. Saito6, N. Endo6,
Y. Mizoguchi6, S. Blakemore3, C. Frith3, V. Gallese4, G. Rizzolatti4, S. Micera7,
P. Dario2,7, H. Takanobu2,8,9, A. Takanishi1,2,5,9,10
1. Consolidated Research Institute for Advanced Science and Medical Care, Waseda University, Tokyo,
Japan, email: zecca@aoni.waseda.jp; takanisi@waseda.jp
2. RoboCasa, Waseda University, Tokyo, Japan
3. Wellcome department of imaging neuroscience University College of London
4. Dipartimento di Neuroscienze, Sezione di Fisiologia, Università di Parma
5. Department of Mechanical Engineering, Waseda University, Tokyo, Japan
6. Graduate School of Science and Engineering, Waseda University, Tokyo, Japan
7. ARTS Lab, Scuola Superiore Sant’Anna, Pisa, Italy
8. Department of Mechanical Systems Engineering, Kogakuin University, Tokyo, Japan
9. Humanoid Robotics Institute (HRI), Waseda University, Tokyo, Japan
10. Advanced Research Institute for Science and Engineering, Waseda University, Tokyo, Japan
Personal robots and robot technology (RT)-based assistive devices are expected to play a major role in our
elderly-dominated society, with an active participation to joint works and community life with humans. In order to
achieve this smooth and natural integration between humans and robots, interaction also at emotional level is a
fundamental required.
Objective of this research, therefore, is to clarify how the emotions expressed by a humanoid robot are
perceived by humans. The preliminary results show several similarities but also several differences in perception.
Key Words: facial emotional expression, humanoid robot, fMRI
1 Introduction
Japan has the world's highest percentage of senior citizens
over 65 (21%) and the smallest percentage of children under
15 (13.6%) [1]. These figures show that Japanese society is
aging much faster than expected, and they underscore the
effects of a shrinking birthrate [2]. In this aging society, it is
expected that there will be a growing need for home, medical
and nursing care services, including those provided by robots,
to assist the elderly both on the physical and the
psychological levels [3]. In this regard, human-robot
communication and interaction are very important,
particularly in the case of home and personal assistance for
elderly and/or handicapped people. If a robot had a "mind"
(intelligence, emotion, and will) similar to the human one, it
would be much easier for the robot to achieve smooth and
natural adaptation and interaction with its human partners
and the environment [4].
Takanishi et al have been developing developed the WE-3
(Waseda Eye No.3) series since 1995. So far they have
achieved coordinated head-eye motion with V.O.R.
(Vestibular-Ocular Reflex), depth perception using the angle
of convergence between the two eyes, adjustment to the
brightness of an object with the eyelids and four senses,
visual, auditory, cutaneous and olfactory sensations. In
addition, they obtained the expression of emotions by using
not only the face, but also the upper-half of the body with the
Emotion Expression Humanoid Robot WE-4 (Waseda Eye
No.4) series with the waist, 9-DOFs emotion expression
humanoids arms and humanoid robot hands RCH-1 (Robo
Casa Hand No.1) [5-7].
WE-4RII transmission of emotions was evaluated by
showing the movies of its six basic emotional expressions
exhibited to many subjects. The users chose the emotion they
thought the robot expressed. The averaged recognition rate of
all emotional expressions of WE-4RII was 93.5 [%], which
proved that WE-4RII can effectively convey its emotions
using its upper-half bodily expressions [7].
However, this kind of analysis lacks of objectivity. In order
to obtain more objective data about the user perception of the
emotions, a different approach should be pursued.
The mirror neuron system [8] is an area of our brain whose
neurons fire both when we perform an action and when we
observe the same action performed by someone else. The
function of the mirror system is a subject of much
speculation. These neurons may be important for
understanding the actions of other people, and for learning
new skills by imitation. It is also considered that the Mirror
Neuron System plays an important role in the recognition of
emotions.
Objective of this research, therefore, is to clarify how the
emotions expressed by a humanoid robot are perceived by
humans. 2 Material and Methods
2.1 Emotion Expression Humanoid Robot WE-4RII
The Emotion Expression Humanoid Robot WE-4RII (see Fig.
1) developed in Takanishi lab is capable of expressing 6
different emotions (Happyness, Anger, Surprise, Sadness,
Disgust, Fear) by using facial expressions and movements of
the neck, the arms and the hands [7].
Fig. 1 The Emotion Expression Humanoid Robot WE-4RII.
In Fig 2 the details about the mechanisms used to obtain
different facial expressions are presented. Eyebrows are
realized by sponge. Each eyebrow is actuated by 4 DC
motors connected by clear wire. Lips are obtained by 2
spinlde shaped springs. Their movement is realized by 4 DC
motors. Eyelids have 6 DOFs.
Fig. 2 Details of the mechanisms for facial expressions.
2.2 Experimental Paradigm
The experimental paradigm contained 16 conditions defined
by a 2x2x4 factorial design with factors:
• 2 Agents: human or humanoid robot WE-4RII;
• 2 Identities of each agent: bald / hairy version;
• 4 facial motions depicted by the agent: silent speech
(articulating a syllable e.g. "ba ba") or emotion
(happiness, anger or disgust).
Silent speech was selected because it is supposed to activate
the motor areas while not activating the emotional area of the
Mirror Neuron System.
Each stimulus consisted of 1.5 seconds greyscale videoclips
(i.e. 38 frames at 25 frames per second). One example for
happyness for one human actor and for the robot with the
wig is presented in Figure 3.
Fig.3 Example of Happyness for one human actor (top) and for the
robot WE-4RII with the wig (bottom)
Two different actors were recorded to prepare the human
stimuli while two versions of the robotic face were prepared
by the addition of a wig (Figure 4).
Fig. 4 2x2 Factorial Plan
Four different versions of each type of stimuli were used (see
fig 5 for happiness). All stimuli started from a neutral pose
and stopped with the emotional expression. Great care was
taken to match the dynamics of the human and robot stimuli
pairwise as much as possible, to minimize the false responses.
The greyscale was digitally modified to match the
background colour and the overall contrast between the
human and robot stimuli. The overall luminosity of the clips
was reduced to avoid too much visual fatigue to the subjects
under fMRI.
Fig. 5: four different samples for Happiness.
2.3 Behavioral Analysis
Participants were asked to recognize the emotional content of
the stimuli. After presentation of emotional stimuli, subjects
had to chose between 4 emotions in a forced choice paradigm
(Angry, Happy, Disgust, Neutral). There were 8 blocks of 8
stimuli for each participant reported here, and each stimulus
was shown once. All subjects experienced 1 to 4
experimental sessions before the one used for the present
analysis, ensuring they all experienced the robot and the
stimuli at the time of acquisition of the behavioral data
reported here.
2.4 fMRI acquisition
The fMRI acquisition consisted of 4 sessions, each one
composed by 8 blocks (4/emotional, 4/neutral tasks), with 8
stimuli per block (1 of each type).
Each presentation was followed by 1.5 seconds response
screen, with Stimulus-onset asynchrony (SOA) jittered
(normal distribution, 6±0.7 seconds), divided before and after
stimulus.
Reminder of task, continuous Likert scale [-200 200] with
target emotion and “None” for emotional task, “Lots” and
“None” for neutral task, side random
Standard SPM2 analysis with specific EPI sequence &
unwarp with fieldmaps was used.
fMRI of the whole brain, with a sequence optimized for
amygdala, orbitofrontal and ventrotemporal cortex,
composed by 48 slices, 3x3x3mm3, TR=4.32 secs, @1.5T
was used. Each subject worked for 1 hour.
Two different types of question were asked to the subjects:
1. Emotional: “How emotional was the face?” Rating from
“Neutral” to the target emotion (i.e. “Happy”);
2. Neutral: “How much movement did the face show?”
From “None” to “Lots”
3 Experimental Results
3.1 Behavioral Analysis
10 subjects participated after giving their informed consent,
independently of or in addition to the fMRI experiment. One
subject did not report any stimulus as "Neutral", and was
removed from the analysis, so that n=9 for the results
reported here.
Analysis of variance (factors of interest: Emotion, Agent;
random factors: version of the agent [bald vs hairy], subjects)
indicates a significant main effect of the agent used to
display the emotion on the ratio of correct answers (number
correct divided by total number for each condition), but no
significant effect of the emotion on the recognition or
interaction between the emotion and the agent.
Results of the behavioral analysis are presented in Table 1
and Figure 6. The difference between human and robot
agents is highly significant (t=4.512, p<0.001), with
emotions being better recognized for the human (98%) than
for the robot (85%) agents.
Table 1. Recognition ratio depending for each agend and emotion.
Fig 6. Recognition ratio depending on each agent and emotion.
3.2 fMRI analysis – behavioral analysys
13 subjects (Male: 4; Femail:9), all right handed, Average
age: 29.4 ±7, range 22.4 – 39.7, gave their informed
consent to the participation to this experiment.
One-way ANOVAs restricted to each emotion were used to
assess differences in ratings due to the agent used to depict
the emotion. Only emotional rating of the Angry stimulus
was significantly different (p<0.001).
Human more emotional and perceived as moving more than
robot across all conditions. It could be either subjective or
due to stimuli not being perfectly matched, though.
Only emotional ratings of Angry stimuli are significantly
affected by the agent (see fig. 7 and 8 for reference).
Therefore, angry stimuli will be excluded from the fMRI
analysis.
Fig 7. Emotion ratings (error bar: SD) depending to agent and
emotion. Number on top gives the effect of agent according to
ANOVA for each emotion and rating.
Fig. 8: Motion ratings (error bar: SD) depending to agent and
emotion. Number on top gives the effect of agent according to
ANOVA for each emotion and rating.
3.3 fMRI analysis – further analysys
At the time of the compilation of this paper, further analysis
of the data is still in progress. The results will be published in
a following paper.
4 Discussion and Conclusion
Objective of this work was to clarify how the emotions
expressed by a humanoid robot are perceived by humans. To
do this, we analyzed the response of the Mirror Neuron
System of several subjects looking at videos of the robot and
of human actors, and we compared the responses.
A pure behavioral analysis showed that the difference
between human and robot agents is highly significant, with
emotions being better recognized for the human (98%) than
for the robot (85%) agents (see Fig 6 and Table 1).
The analysis of the fMRI data is still preliminary. However,
some simple conclusion can be drawn. In a one-way
ANOVAs restricted to each emotion, only emotional rating
of the Angry stimulus was significantly different (p<0.001).
In the future, the analysis will be extended to different goups
of subjects, in order to clarify the dipendencies with age, sex,
and cultural backgound. The analysis will be also extended to
full body emotional expressions, which is believed to be even
more important than facial expressions [9] (i.e.. while a
fearful faces signal a threat, it does not provide information
about either the source of the threat or the best way to deal
with it. By contrast, fearful body positions signal a threat and
at the same time specify the action undertaken by the
individuals fearing for their safety).
ACKNOWLEDGMENT
Part of this research was conducted at the Humanoid Robotics
Institute (HRI), Waseda University. The authors would like to
express thanks to Okino Industries LTD, OSADA ELECTRIC CO.,
LTD, SHARP CORPORATION, Sony Corporation, Tomy Company
LTD and ZMP INC. for their financial support for HRI. And, the
authors would like to thank Italian Ministry of Foreign Affairs,
General Directorate for Cultural Promotion and Cooperation, for its
support to the establishment of the ROBOCASA laboratory. In
addition, this research was supported by a Grant-in-Aid for the
WABOT-HOUSE Project by Gifu Prefecture. Part of the research
has been supported by the EU FET NEUROBOTICS
FP6-IST-001917 “The fusion of Neuroscience and Robotics”.
Finally, the authors would also like to express thanks to ARTS Lab,
NTT Docomo, SolidWorks Corp., Consolidated Research Institute
for Advanced Science and Medical Care, Waseda University,
Advanced Research Institute for Science and Engineering, Waseda
University, Prof. Yutaka Kimura, Dr. Yuichiro Nagano and Dr.
Naoko Yoshida for their support for our research.
References
[1] Ministry of Internal Affairs and Communications of Japan, Preliminary
Report on the 2005 National Census, June 2006.
[2] NIPS. Population statistics of Japan 2003. Technical report, National
Institute of Population and Social Security Research, Japan, 2003
[3] JApan Robot Association, Summary Report on Technology Strategy
for Creating a Robot Society in the 21st Century, May 2001.
[4] S. Hashimoto, “KANSEI as the Third Target of Information Processing
and Related Topics in Japan.” Proceedings of the International
Workshop on KANSEI: The Technology of Emotion, pp.101-104.,
1997.
[5] H. Miwa, K. Itoh, et al.: “Design and Control of 9-DOFs Emotion
Expression Humanoid Arm”, Proceedings of the 2004 IEEE
International Conference on Robotics and Automation, pp.128-133,
2004.
[6] H. Miwa, K. Itoh, et al.: “Effective Emotional Expressions with
Emotion Expression Humanoid Robot WE-4RII”, Proceedings of the
2004 IEEE/RSJ International Conference on Intelligent Robots and
Systems, pp.2203-2208, 2004.
[7] M. Zecca, A. Takanishi, et al, “On the development of the Emotion
Expression Humanoid Robot WE-4RII with RCH-1”, HUMANOIDS
2004, v1, Page(s):235 - 252.
[8] G. Rizzolatti, L. Craighero, “The Mirror-Neuron System”,Annual
Review of Neuroscience, Vol. 27: 169-192, July 2004
[9] B. De Gelder, “Towards the neurobiology of emotional body language”,
Nature, v7:2006, pp 242-249