Page 157 - Designing Sociable Robots

breazeal-79017 book March 18, 2002 14:7

138 Chapter 9

dynamics of behavioral transitions must be well-matched to natural human interaction
speeds. For instance, the transition from the call-to-person behavior (to bring a distant
person near) to the activation of the greet-person response (when the person closes to
face-to-face interaction distance) to the transition to the vocal-play behavior (when the
person says his/her first utterance) must occur at a pace that the human feels comfortable
with. Each of these involves showing the right amount of responsiveness to the new stim-
ulus situation, the right amount of persistence of the active behavior (the motor act must
have enough time to be displayed and witnessed), and the right amount of delay before
the next behavior becomes active (so that each display is presented as a purposeful and
distinct act).
Temporal dynamics between levels A similar issue holds for the dynamics between
different levels of the hierarchy. If a child behavior is successfully addressing the goal of
its parent, then the parent should remain active longer to support the favorable progress of
its child. For instance, if the robot is having a good interaction with a person, then the time
spent doing so should be extended—rather than rigidly following a fixed schedule where the
robot must switch to look for a toy after a certain amount of time. Good quality interactions
should not be needlessly interrupted; the timing to address the robot’s various needs should
be flexible and opportunistic. To accomplish this, the parent behaviors are made aware of
the progress of their children. The container node of the child passes activation energy up
the hierarchy to its parent, and the parent’s activation is a combination of its own measure
of relevance and that of its child.

Affective influence Another important set of parameters adjust how strongly the active
behaviors influence the net affective state. The amount of valence, arousal, and stance
sent to the emotion system can vary from behavior to behavior. Currently, only the leaf
behaviors of the hierarchy influence the emotion system. Their magnitude and growth
rate determine how quickly the robot displays frustration, how strongly it displays pleasure
upon success, etc. The timing of affective expression is important, since it often occurs
during the transition between different behaviors. Because these affective expressions are
social cues, they must occur at the right time to signal the appropriate event that elicited the
expression.
For instance, consider the period of time between successfully finding a toy during the
seek-toy behavior, and the transition to the engage-toy behavior. During this time span,
the seek-toy behavior signals its success to the emotion system by sending it a positively
valenced signal. This increase in net positive valence is usually sufficient to cause joy to
become active, and the robot smiles. The smile is a social cue to the caregiver that the robot
has successfully found what it was looking for.