Emotional
expression of robots was examined in a recent study participated by BME Faculty
of Mechanical Engineering, Department of Mechatronics, Optics, and Mechanical
Engineering Informatics (BME-GPK MOGI). The result was made in international
cooperation of researchers, institutes and research groups on robotics and
ethology. The examination is going to be tested with a waiter robot. The first
author of the paper, published in Scientific Reports, published by the
prestigious Nature, was Beáta Korcsok (left), assistant lecturer of the BME-GPK
MOGI Department. Interview with professor Péter Korondi (right), a member of the
project.
How did you find this topic of
research?
Péter
Korondi: The Department of Mechatronics, Optics and Mechanical Engineering Informatics
(MOGI) takes part in researches on ethorobotics, which field of study was born
in this form due to a cooperation of domestic scientific institutions. In the
researches on ethorobotics, the BME-GPK MOGI takes the mechanical tasks, while
the ELTE Department of Ethology, and the MTA-ELTE Comparative Ethology Research
Group takes the ethology-wise ones. The Miskolc University Institute of
Information Science, and as a foreign partner, the Tokyo-based Chuo University
also participated in the research.
 |
| MOGI Robi (left), and MOGI Ethon (right) |
The research was triggered by a conversation, when
Ádám Miklósi ethologist told, that how they rewrote mother-child attachment
studies into dog-owner attachment tests. That time we decided that if we
replace dogs with robots, the robot-owner attachment can be examined, too. Of
course, with this, the attachment behaviour examination of robots should be
needed. Ádám Miklósi mentioned some rules, that they have found, and they
agreed, that these verbal ethology rules should be described with fuzzy logics.
That was the particular moment when the Hungarian ethorobotics was founded. The
MOGI Robi, already retired, was created that time.
Ethorobotics, which term is a merge of the terms
’ethology’ and ’robotics’, seek opportunities of developing such social robots,
whose behaviour and communication can integrate them into the human
environment, and are more acceptable for humans. This field of study came to
life, although the usage of social robots is increasing, the communication
towards humans is still an unsolved problem. The communication of the robot
must be understandable for anyone, and also should not be disturbing in the
long term.
The most wide-spread direction of the research is
creating humanoid robots, which encounters some problems. The humanoid robots,
which can manage tasks, are still not going to be produced technologically for
a long time; secondly, their human look can trigger fear or dismay in humans.
The ethorobotical approach, on the other hand, robots should be handled as an
individual species, and their communication should be set to their function and
abilities.
During our research, we examined such models, mainly
based on the behaviour of social animals, enabling us to design proper
behaviour and communication signals. In our study, we examined such signals and
voice effects.
What kind of tasks did you have
to deal with?
Péter Korondi: Within this research series, we have been focusing on
the emotional expression, for instance, with the help of visual signals, when
we designed the emotional expressions. For example, with the help of an
artificial agent appearing on the monitor, the agent was built of a circle and
a square, maze-like abstract shape. With motion and changes in its colour and
size, we created emotional expressions along with such patterns, which are
present in the emotional expressing behaviour of some animal species.
In the recent period, we examined, that in case of
artificially-created voices, whether can we observe or not such patterns, which
present in the animal and human vocalisations. We also pointed connections
between the acoustical parameters of the vocalisation and the emotional state
of the individual.
Humans can not only understand emotional expressions
of individuals having different mother tongue, or culture, with non-verbal
vocalisations (e.g. crying, screaming, laughing or sighing), but it is true to
many land mammals, for example, dogs and pigs.
This phenomenon is enabled by the similar vocal tracts,
and mode of vocalisation of the mammals, thus, expression of the similar
emotional states. Such a pattern is, for instance, that the high fundamental
frequency voices are perceived as more intensive, while the short ones are
perceived as more positive ones.
During our research, with the help of Praat acoustics
software, we generated 600 voices, with the systematic change of the pitch, and
the length of the sounds, besides, we added multiple acoustic parameters to the
sounds, creating seven voice categories. The artificial voices range from
simple sounds (which are the simplest and clearest ones) to complex sounds,
modelling mammal vocalisation. The sounds were examined by an online survey
with statistical data analysis.
What was the most significant
challenge you faced in the research?
Péter Korondi: In this research, the most significant challenge was
choosing the proper parameters and thresholds of the artificial sounds, as
there was a wide range of opportunities, due to the complexity of the animal
vocalisation. The long history of bioacoustical researches helped us, not only
examining mechanisms of animal and human vocalisations, but also emotional
expressions between and inside species, and which are studied at the ELTE
Department of Ethology.
 |
| Felirat Illustration: when filling the questionnaire, the voices had to be located in the coordination system above, depending on their intensity and emotional valence |
The study about the voice-triggered
emotions was conducted with an online questionnaire. How was the research
conducted? How many individuals were in the sample, with what kind of selection
method?
Péter Korondi: The research was done with an online interactive
questionnaire, which we were seeking participants, advertised on social media
platforms. In the study, voluntary adults were participating. The poll was
available in English and Hungarian as well. After the demographic questions,
and some voices showing the users how to handle the platform, the participants
had to listen to ten of each of the seven sample categories. After hearing the
sound, the participants had to locate it in a coordinate system, depending on
its intensity and emotional valence. The answers were filtered by the response
time, and we ruled out responses coming after 20 seconds. We processed the
response of 237 participants, 95 responding in Hungarian, 142 in English.
Which animal had a sound kit the closest to the optimal?
Péter Korondi: The participants evaluated artificial voices based on
the patterns, observed on animal voices, not depending on the biological
complexity, so the higher frequency sounds were perceived more intensive. In
contrast, the shorter sounds were perceived as more positive. As this result
suggests, that the examined emotional coding patterns should be very ancient.
Moreover, these rules work similarly regardless of the exact species. Besides
vocalisation, the sound-processing neural processes should be similar among
land mammals. So there were no optimal, or less optimal sound kit regarding
emotional expression, enabling further voice setting of these features.
 |
| Biscee, the waiter robot |
Do you plan further researches on
the topic?
Péter Korondi: Yes, the artificial voices created in the study,
should be examined in further research, that they trigger approach-avoidance,
which serves as the third, social dimension of this model. With the inclusion
of this dimension, such negative and intensive voices can be separated, which
are all negative and intense voices, based on the first questionnaire, but the
reaction given to such sounds can vary totally. As an everyday example, we could
think about sounds of anger, and fear or pain: we come less close to an angrily
barking, snarling dog, than a painfully whimpering one. However, both are
intensive voices with a negative resonation. The research results on
artificial-voice triggered approach-avoidance are going to be published soon.
We are going to study the created voices in real-life
situations, with the help of a social robot, Biscee, which we plan to launch in
a café environment, working as a waiter, enabling us to analyse voices, and
further communication signals.
László Benesóczky
No comments:
Post a Comment