Professional Documents
Culture Documents
net/publication/254926056
CITATIONS READS
24 265
3 authors:
Stelarc Stelarc
Brunel University London
10 PUBLICATIONS 100 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Damith C. Herath on 02 December 2016.
Access provided by Curtin University of Technology (13 May 2014 21:15 GMT)
A R T I S T S ’ A R T I C L E
© 2012 ISAST LEONARDO, Vol. 45, No. 5, pp. 401–407, 2012 401
Fig. 1. Schematic of the hardware and software system of the Articulated Head. (© Damith C. Herath. Design: Powerhouse Museum Design
Studio, Sydney.)
(a) The Prosthetic Head was a 5-m-high (d) Use of an industrial robot arm en- motion tracking, it will be exposed as
projection on a wall. Although im- sures that the system performs reli- such quickly; if the behavior does not
pressive in scale, it was essentially a ably and robustly. appear to be connected with the actions
screen-based installation and thus of the user, or only insufficiently so, the
purely virtual. In contrast, the Ar- On the scientific and engineering side, system will be considered faulty or viewed
ticulated Head is an actual-virtual the concept of the Articulated Head posed as random (Waytz et al. [9] discuss some
system with an LCD screen imaging unique challenges. The context of a work of the properties of human-robot inter-
the Head mounted to the end of an of art meant that (a) there would be no actions that influence the attribution of
industrial robot arm that becomes clearly defined task, (b) there would be intentionality).
an articulated 6-degrees-of-freedom few boundary conditions constraining As described above, we assume that
neck. The fusion between physical the interaction with the visitor and (c) the solution is to be found in an action-
and virtual elements is reinforced expectations with which visitors would perception control system with a tight
by the syncing of the physical be- approach the Articulated Head would vary coupling between action and perception
havior of the robot and the facial widely. and an attention model at its core.
behavior of the Prosthetic Head. The Most importantly, however, the Articu- Human attention is typically investi-
robot arm, mounted on a steel lated Head would have to be perceived gated in controlled psychological experi-
base, gives it an anthropomorphic as an intentional agent based on its ments focusing on specific aspects of the
scale and feel. The system is mini- motor behavior alone, in order not to overall phenomenon, e.g. shifts in visual
mal and clean in its aesthetics. undermine its conversational skill real- attention triggered by priming stimuli. In
(b) The robot system has a physical, ized via the A.L.I.C.E. chatbot [8] inte- thousands of studies, many insights have
sculptural presence that allowed grated into the Prosthetic Head. Perceived been gained, yet a general definition of
us to actualize and evaluate sound agency might not be difficult to evoke, attention has remained elusive. Rather
location, vision tracking and a face- as humans ascribe agency quickly, but broad primary characteristics have been
tracking system in a 3D space. the illusion breaks down quickly, too, found to be selection (of sensory infor-
(c) The hybrid robot system bypasses and evaluation might be unwieldy. If the mation), binding and limited capacity
any “uncanny valley” [7] issues, behavior of the robot appears to be the [10]. For attention systems in machines,
as it does not resemble a human mechanical consequence of whatever however, an important distinction be-
but clearly announces its machine the human user does (or a particular as- tween two different types of attention
character. pect of it), e.g. pursuit based on simple emerged: saliency in the perceptual in-
its lifetime and the spatial location in the after the event that caused it has dis- to determine whether a new perceptual
real world it is referring to. Its weight is appeared, but that at the same time its event encountered is—per definition—
originally determined using a base weight strength fades even if registration of the identical to one of the already existing
assigned to the type of perceptual event perceptual event is sustained. We used a attention foci. Locations of existing at-
that is the source of the attention focus. specific exponential function called the tention foci are matched with the loca-
Thus, for instance, a face-detection event Kohlrausch function, which is known to tions of new candidates. If an incoming
will receive a higher base weight than an be able to describe a wide range of physi- event and one of the attention foci are
acoustic localization event, as the Articu- cal and biological phenomena [24]. The indeed found to be identical, the old
lated Head is set up to be geared toward free parameters of the function are ini- focus is maintained but its location and
face-to-face interactions with humans. A tialized dependent on the type of percep- weight are updated. The combination
factor dependent on a chosen property tual event, but, again, they are modified of the new and old weights is modeled
of the perceptual event, e.g. a confidence dynamically during run time. Adjusting supra-additively—that is, the resulting
value returned by a sensing routine, is a stretching/compressing parameter of value is smaller than the sum of the
then multiplied by the base weight to de- the decay function toward shorter decays two original values. The decay func-
termine the final attentional weight. This can cause the Articulated Head to appear tion, however, will not be reset in the
is used to increase the stability of certain very nervous, constantly switching to fusion of attention foci. Thus, even if
behaviors, e.g. when the Articulated Head new attention foci, whereas adjusting it new events are constantly reinforcing an
is engaged in a face-to-face interaction. toward longer decay times will make it old attention focus—for instance, a per-
If the face-tracking routine returns a appear slow and unresponsive. son standing still within the visual field
high confidence value, signaling de- The attention foci are in general spa- of the Articulated Head—the focus will
tection of a face with the monovision tially organized—that is, they are defined eventually reach very low weight values
camera, the Articulated Head becomes via a segment of 3D space centered on and will be removed (modeling adapta-
difficult to distract from this interaction the location of the event that attracted tion). More generally, the above settings
and may start to mimic the user’s head attention (compare space- versus object- enable the Articulated Head to strike a bal-
orientation. based attention in models of human at- ance between focusing on a single source
The decay function ensures that an tention [25]). This becomes particularly and distributing attention over several
attention focus has a certain lifetime important when the attention system has sources. In particular, if there is a crowd