Page 189 - Designing Sociable Robots

breazeal-79017 book March 18, 2002 14:11

170 Chapter 10

). For each P i , the weighting function
the displacement from (a,v, s) to each (A P i , V P i , S P i
). The weight is bounded
f i (A, V, S, N) decays linearly with distance from (A P i , V P i , S P i
). The
between 0 ≤ f i (A, V, S, N) ≤ 1, where the maximum value occurs at (A P i , V P i , S P i
argument N defines the radius of influence which is kept fairly small so that the contribution
for each specialized prototype posture does not overlap with the others.
Comparison to Componential Approaches
It is interesting to note the similarity of this scheme with the affect dimensions viewpoint
of emotion (Russell, 1997; Smith & Scott, 1997). Instead of viewing emotions in terms of
categories (happiness, anger, fear, etc.), this viewpoint conceptualizes the dimensions that
could span the relationship between different emotions (arousal and valence, for instance).
Instead of taking a production-based approach to facial expression (how do emotions gen-
erate facial expressions), Russell (1997) takes a perceptual stance (what information can an
observer read from a facial expression). For the purposes of Kismet, this perspective makes
a lot of sense, given the issue of readability and understandability.
Psychologists of this view posit that facial expressions have a systematic, coherent, and
meaningful structure that can be mapped to affective dimensions (Russell, 1997; Lazarus,
1991; Plutchik, 1984; Smith, 1989; Woodworth, 1938). (See figure 10.6 for an example.)
Hence, by considering the individual facial action components that contribute to that struc-
ture, it is possible to reveal much about the underlying properties of the emotion being
expressed. It follows that some of the individual features of expression have inherent signal
value. This promotes a signaling system that is robust, flexible, and resilient (Smith & Scott,
1997). It allows for the mixing of these components to convey a wide range of affective
messages, instead of being restricted to a fixed pattern for each emotion. This variation al-
lows fine-tuning of the expression, as features can be emphasized, de-emphasized, added, or
omitted as appropriate. Furthermore, it is well-accepted that any emotion can be conveyed
equally well by a range of expressions, as long as those expressions share a family resem-
blance. The resemblance exists because the expressions share common facial action units. It
is also known that different expressions for different emotions share some of the same face
action components (the raised brows of fear and surprise, for instance). It is hypothesized
by Smith and Scott that those features held in common assign a shared affective meaning to
each facial expression. The raised brows, for instance, convey attentional activity for both
fear and surprise.
Russell (1997) argues the human observer perceives two broad affective categories on the
face, arousal and pleasantness. As shown in figure 10.6, Russell maps several emotions and
corresponding expressions to these two dimensions. This scheme, however, seems fairly
limiting for Kismet. First, it is not clear how all the primary emotions are represented with
this scheme (disgust is not accounted for). It also does not account for positively valenced