Page 154 - Designing Sociable Robots

breazeal-79017 book March 18, 2002 14:7

The Behavior System 135

Level of Interest,
Frustration Activation Level,
A
Behavior Group
Releasers
Behavior
Emotion, Bias
Drive A ma x
0
Threshold,
Goal Releaser T
[A, V, S]
Figure 9.3
The model of a behavior.

system, the observed behavior of the robot is the result of competition at the functional,
strategy, and task levels.

The Behavior Model
The individual behaviors within a group compete for activation based on their computed
relevance to the given situation. Each behavior determines its own relevance by taking into
account perceptual factors (as defined by its affiliated releaser and goal releaser) as well
as internal factors (see figure 9.3). The internal factors can either arise from an affiliated
emotion (or drive at the top level), from activity of the behavior group to which it belongs
(or the child behavior group, if present), or the behavior’s own internal state (such as its
frustration, current level of interest, or prepotentiated bias). Hence, as was the case with the
motivational system, there are many different types of factors that contribute to a behavior’s
relevance. These influences must be converted into a common currency and combined to
compute the activation level for the behavior. The activation level represents some measure
of the behavior’s “value” to the robot at that point in time.
Provided that the behavior group is active, each behavior within the group updates its
level of activation by the equation:

A update = (releaser n · gain ) + (motiv m · gain )
n m
n m

+ success releaser goal,k · (LoI − frustration) + bias (9.1)
k
A child is the activation level of the child behavior group, if present
n is the number of releaser inputs, releaser n
gain is the weight for each contributing releaser
n
m is the number of motivation inputs, motiv m