Page 155 - Designing Sociable Robots

breazeal-79017 book March 18, 2002 14:7

136 Chapter 9

motiv m corresponds to the inputs from drives or emotions
gain is the weight for each contributing drive or emotion
m
success() is a function that returns 1 if the goal has not been achieved, and 0 otherwise
releaser goal,k is a releaser that is active when the goal state is true (i.e., a goal releaser)
LoI is the level of interest, LoI = LoI initial − decay(LoI, gain )
decayLoI
LoI initial is the default persistence
frustration increases linearly with time, frustration = frustration + (gain · t)
frust
bias is a constant that pre-potentiates the behavior
x
decay(x, g) = x − for g > 1 and x > 0, and 0 otherwise
g
When the behavior group is inactive, the activation level is updated by the equation:

A behavior = max A child , (releaser n · gain ), decay(A behavior , gain ) (9.2)
n decayBeh
n
Internal Measures
The goal of each behavior is defined as a particular relationship between the robot and
its environment (a goal releaser). The success condition can simply be represented as
another releaser for the behavior that fires when the desired relation is achieved within the
appropriate behavioral and motivational context. For instance, the goal condition for the
seek-person behavior is the found-person releaser, which only fires when people are
the desired stimulus (the social-drive is active), the robot is engaged in a person-finding
behavior, and there is a visible person (i.e., skin tone object) who is within face-to-face
interaction distance of the robot and is not moving in a threatening manner (no excessive
motion). Some behaviors, particularly those at the top level of the hierarchy, operate to
maintain a desired internal state (keeping its drive in homeostatic balance, for instance). A
releaser for this type of process measures the activation level of the affiliated drive.
The active behavior sends information to the high-level perceptual system that may be
needed to provide context for the incoming perceptual features. When a behavior is active,
it updates its own internal measures of success and progress to its goal. The behavior
sends positive valence to the emotion system upon success of the behavior. As time passes
with delayed success, an internal measure of frustration grows linearly with time. As
this grows, it sends negative valence and withdrawn-stance values to the emotion system
(however, the arousal and stance values may vary as a function of time for some behaviors).
The longer it takes the behavior to succeed, the more frustrated the robot appears. The
frustration level reduces the level-of-interest of the behavior. Eventually, the
behavior “gives up” and loses the competition to another.