Page 179 - Designing Sociable Robots

breazeal-79017 book March 18, 2002 14:11

160 Chapter 10

both ears, or knitting both brows, or blinking both lids.) These are simply constructed from
those primitives controlling each individual facial feature.
The Motor Server Layer

The motor server layer arbitrates the requests for facial expression, facial display, or lip
synchronization. Requests originating from these three functions involve moving ensembles
of facial features in a coordinated manner. These requests are often made concurrently.
Hence, this layer is responsible for blending and or sequencing these incoming requests so
that the observed behavior is coherent and synchronized with the other motor modalities
(voice, eyes, and head).
In some cases, there is blending across orthogonal sets of facial features when subsystems
serving different facial functions control different groups of facial features. For instance,
when issuing a verbal greeting the lip synchronization process controls the lips and jaw
while a facial display process wiggles the ears. However, often there is blending across the
samesetoffacialfeatures.Forinstance,whenvocalizingina“sad”affectivestate,thecontrol
for lip synchronization with facial emphasis competes for the same facial features needed
to convey sadness. Here, blending must take place to maintain a consistent expression of
affective state.
Figure 10.2 illustrates how the facial feature arbitration is implemented. It is a priority-
based scheme, where higher-level subsystems bid for each facial feature that they want to
control. The bids are broken down into each observable movement of the facial feature.
Instead of bidding for the left ear as a whole, separate bids are made for left ear elevation
and left ear rotation. To promote coherency, the bids for each component movement of a
facial feature by a given subsystem are generally set to be the same. The flexibility is present
to have different subsystems control them independently, should it be appropriate to do so.
The highest bid wins the competition and gets to forward its request to the underlying facial
feature primitive. The request includes the target position, velocity, and type of movement
(postural or rhythmic).
The priorities are defined by hand, although the bid for each facial feature changes
dynamically depending on the current motor skill. There are general rules of thumb that are
followed. For a low to moderate “emotive” intensity level, the facial expression subsystem
sets the expression baseline and has the lowest priority. It is always active when no other
facial function is to be performed. The “emotive” baseline can be over-ridden by “voluntary”
movements (e.g., facial gestures) as well as behavioral responses (such as “sleeping”). If an
emotional response is evoked (due to a highly active emotion process), however, the facial
expression will be given a higher priority so that it will be expressed. Lip synchronization has
thehighestpriorityoverthelipsandmouthwheneverarequesttospeakhasbeenmade.Thus,
whenever the robot says something, the lips and jaw coordinate with the vocal modality.