Information Technology Reference
In-Depth Information
Meta-level controls
The third requirement of narrative comprehensibility is that behaviors should
be structured into a coherent sequence. Instead of jumping around between ap-
parently independent actions, the agent's activities should express some com-
mon threads. But these relationships between behaviors are difficult to express
in most behavior-based systems because they treat individual behaviors as dis-
tinct entities which do not have access to each other. Conflicts and influences
between behaviors are not handled by behaviors themselves but by underlying
mechanisms within the architecture. Because the mechanisms that handle re-
lationships between behaviors are part of the implicit architecture of the agent,
they are not directly expressible to the user.
The Expressivator deals with this problem by giving behaviors
meta-level
controls
, special powers to sense and influence each other. Because meta-level
controls are explicitly intended for communication and coordination between
behaviors, they are in some sense a violation of the behavior-based principle of
minimal behavioral interaction. Nevertheless, meta-level controls are so useful
for coordinating behavior that several have already found a home in behavior-
based architectures. An example is Hamsterdam's meta-level commands, which
allow non-active behaviors to suggest actions for the currently dominant be-
havior to do on the side (Blumberg 1996). In the Expressivator, behaviors can
(1)
query
which other behaviors have recently happened or are currently ac-
tive; (2)
delete
other behaviors; (3)
add
new behaviors, not as subbehaviors,
butatthetopleveloftheagent;(4)
add new sub-behaviors
to
other
behaviors;
(5)
change the internal variables
that affect the way in which other behaviors
are processed; (6)
turn off
a behavior's ability to send motor commands, and
(7)
move running subbehaviors
from one behavior to another.
The most important function for these meta-level controls in the Expres-
sivator is to allow for the implementation of transitions. Transitions, at a mini-
mum, need to be able to find out when an old behavior needs to be terminated,
delete the old behavior, engage in some action, and then start a new behavior.
This means that transition behaviors need to have all the abilities of a regular
behavior, and a few more: (1) they need to be able to know what other behav-
iors are running; (2) they need to be able to delete an old behavior; and (3)
they need to be able to begin a new behavior. Ideally, they should also be able
to alter the new behavior's processing to reflect how it relates to what the agent
was doing before. In the Expressivator, transitions can do all these things with
meta-level controls.
