Professional Documents
Culture Documents
1
environment. This is the focus of Torres-Jara, Na- action architecture for imprinting – the establish-
tale and Fitzpatrick’s contribution. They adopt a ment of strong attachment links between robot and
bottom-up approach to adapting and learning about caregiver. Imprinting allows adaptation as a result of
the environment. Rather than providing the system reward-based learning in the context of a history of
with abilities that would enable the robot to accom- affective interactions between the robot and the hu-
plish specific tasks, they provide it with a generic man. This certainly is an ability an epigenetic robot
ability: tapping. The repetitive, redundant, cross- will need in order to proceed further in its social de-
modal nature of tapping gives the robot an oppor- velopment.
tunity to reliably identify when the sound of contact
with the object occurs, and to collect samples of that 2.2 Short papers
sound. The authors demonstrate the utility of this
exploratory procedure for a simple object recogni- Building on their previous work on homeostatic be-
tion scenario. Now, how would a robot understand haviors, Andry, Gaussier, and Nadel propose a
whether it is touching an object or itself? That’s neural architecture for sequence learning, an impor-
the question Edsinger-Gonzales tries to answer. tant component for an imitating system. In their
His robot learns to disciminate between ego and exo architecture, demonstration and reproduction stages
forces purely on the basis of its proprioceptive sense are fused, with learning occuring when the system is
of force, through a series of semi-autonomous devel- not in an equilibrium.
opmental stages. Vitay proposes a computational model for a sys-
tem to learn to count objects. An interesting par-
Two papers round up this series of motor-related ticularity of his model is that this ability is realized
papers. In the first one, Konczak continues on last from the merging, via a common reward signal, of
year’s keynote with a review of two reflexive motor two independent computational models: a model for
patterns in humans: Primitive reflexes and motor switching spatial visual attention, and a model for
primitives. He argues that an understanding (and learning the ordinal sequence of phonetical numbers.
modelling?) of the human motor system that encom- Olsson, Nehaniv, and Polani continue with
passes both primitive reflexes and motor primitives their research on the issue of how many sensory sys-
as well as the interaction with supraspinal motor cen- tems adapt to the environment. This year, they have
ters is critical to the design of epigenetic robots. a robot learn how correlations over time in the sen-
Indeed, understanding the mechanisms (change in sors are related to the actuators’ activities. With this
connectivity? internal clock mechanism?) by which approach, a robot can develop from unknown sensors
the primitive motor system is being integrated with and actuators to being able to perceive motion.
later maturing supraspinal motor centers to give rise Van Dartel and Postma investigate whether
to complex motor behaviors should be most benefi- symbol manipulation, the hallmark of traditional
cial for robots to acquire complex motor skills in a cognitive science, can be achieved by robots with the
bottom-up approach. In the second paper, Veskos ability to simulate perception and behaviour inter-
and Demiris revisit a study published by Berthouze nally. As a case-study they investigate how and why
and Lungarella in a previous edition of the workshop internal simulation in a Situated Tower of London
on swinging and the degrees of freedom problem. Us- task improves performance in the Tower of London
ing a different formalism of central pattern genera- task, and compare system and human performance.
tors, the authors successfully apply the same prin- In an interesting attempt towards achieving on-
ciple of entrainment to acquiring two motor skills: going emergence, Clowes and Morse look at lan-
swinging and walking. guage as a way to not only speed up the acquisi-
Finally, two papers deal with much higher cog- tion of normal behaviours, but also to further enable
nitive tasks but still within the context of the sys- successful operation at tasks beyond the original ca-
tem’s sensory inputs. Lacerda et al. attempt to pabilities of the agent. Investigating the Vygotskian
interpret the infant’s first steps in the acquisition of idea that language plays a role in intimately structur-
the ambient language as a consequence of the in- ing the learning environment, allowing the construc-
fant’s general capacity to represent sensory input tion of more complex cognitive activities, the authors
and the infant’s interaction with other actors in its describe experiments with evolved agents that use
immediate ecological environment. The authors ar- self-directed speech and show that they achieve high
gue that their Ecological Theory of Language Ac- fitness more rapidly than agents that don’t.
quisition (ETLA), which they presented last year, Huang and Weng describe a reinforcement-
offers a productive alternative to traditional descrip- based model for a robot to develop its covert per-
tive views of the language acquisition process by pre- ceptual capability (the selection of an action by a
senting an operative model of how early linguistic value system). The method is tested in a non-trivial
function may emerge through interaction. Finally, vision-based navigation task.
Blanchard and Cañamero discuss a perception- Finally, in a contribution that nicely complements
2
that of Prince et al., Metta, Vernon, and Sandini inary experimental results showing the efficiency of
outline an agenda for the RobotCub project which this approach compared to standard fitness criteria.
could very well be an agenda for our entire commu- By sharing environmental sounds, humans and
nity. robots can share information on the environment.
Unfortunately, it is difficult for humans and robots
2.3 Posters to communicate such sounds. Hattori et al. sug-
gest to acquire a multi-modal mapping from sound
Understanding the development of joint attention is to motion (the motion performed by the object that
a crucial issue for research in epigenetic robotics. produced the sound) by learning from temporal con-
Prior studies in developmental psychology argue that tingencies. The robot can then imitate the motion
skills underlying joint attention do not emerge until using its body.
9 month of age. Stahl and Striano present a longi- Aryananda placed her humanoid head robot with
tudinal experimental study with sixteen infants from two basic in-built behaviors (tracking of faces and
7 to 10 months of age. Their results seem to indicate bright objects, mimicking of phoneme sequences) in
that such skills gradually unfold before 9 months of an open-ended environment and tried to find out
age. These findings that go in favor of a gradual de- what the robot could see, and what it could hear.
velopment give encouraging support for robotic mod- Results show that there was enough ”interesting” in-
els that try to reenact the developmental trajectory formation for the robot to start unsupervised lexical
towards joint attention. To achieve joint attention acquisition for example.
between machines using artificial vision systems, seg- Mirza, Nehaniv, te Boekhorst, and Daut-
mentation stability is often a key issue. Baillie and enhahn show that information theoretical methods
Nottale describe a way to measure the stability of a can be used for a robot to characterize its inter-
segmentation algorithm and present experimental re- actions and interaction history. In particular, the
sults obtained with several algorithms under various authors show that fractal dimension of the sensory-
forms of image distortion. motor phase plot is a useful measure.
How can robots and computational models help Prepin, Gaussier, Revel, and Nadel briefly
understanding developmental disorders like autism? outline the basic principles of a formal approach –
Björne and Balkenius have worked for a few years the Cognitive System Formalism – to represent, ana-
on a computational theory of autism. This year, they lyze and compare cognitive systems of developmental
investigate the role of context in motor impairments robotics and psychology.
for children with autism. They present a specula- Finally, Oka and Ozaki describe RobotPHONE,
tive view arguing that an inability to take contex- a simple robotic platform designed for educational
tual factors into account could explain some charac- purposes. By using a software SDK, students can ex-
teristic motor impairment in autism. From another periment with associative learning systems and build
perspective, Miyamoto, Lee, Fujii and Okada in- easily simple epigenetic systems.
vestigate how robots can facilitate interaction with
autistic children. They present results of a longi- Thanks
tudinal study with five children interacting with a
speaking robot in two situations. They discuss in We thank the National Institute of Communication
particular the cases of two children who showed de- Technology (NICT) of Japan for their generous sup-
velopmental change in interacting with the robot. port of this workshop, and Hideki Kozima in partic-
How does an infant develop the ani- ular for dealing with the local organization and with
mate/inanimate distinction? What underlie the administrative side of the organization. Finally,
the feeling of animacy? Kuwamura, Yamamoto we also thank the program committee members for
and Hashimoto discuss the relationship between their efforts in reviewing submissions.
movement and animacy with a psychophysical
experiment where dynamical systems are designed
using an interactive evolution method. Their results
suggest that oscillations and stroke movements are
key perceptual factors underlying the feeling of
animacy.
Bredeche and Hugues investigate the issue of
open development from an evolutionary robotics per-
spective. They suggest to define stages of increasing
complexity in order to evolve increasingly complex
controllers. For this purpose, they define an environ-
ment generator that automatically produces environ-
ments of adapted complexity. They present prelim-