Page 190 - Introduction to AI Robotics

173
5.5 Assemblages of Behaviors
This example also illustrates the tendency among purely reactive motor
schema to assign one sensor per behavior.

5.5 Assemblages of Behaviors

The UGV competition case study illustrated the basic principles of the design
of reactive behaviors. In that case, there were a trivial number of behaviors.
What happens when there are several behaviors, some of which must run
concurrently and some that run in a sequence? Clearly there are releasers
somewhere in the system which determine the sequence. The issue is how
to formally represent the releasers and their interactions into some sort of
sequencing logic. If a set of behaviors form a prototypical pattern of action,
they can be considered a “meta” or “macro” behavior, where a behavior is
assembled from several other primitive behaviors into an abstract behavior.
This raises the issue of how to encapsulate the set of behaviors and their
sequencing logic into a separate module.
The latter issue of encapsulation is straightforward. The same OOP schema
structure used to collect a perceptual schema and a motor schema into a be-
havior can be used to collect behaviors into an abstract behavior, as shown
by a behavior being composed of behaviors in Fig. 5.1. The coordinated con-
trol program member of the abstract behavior expresses the releasers for the
component behaviors.
This leaves the issue of how to formally represent the releasers in a way
that both the robot can execute and the human designer can visualize and
troubleshoot. There are three common ways of representing how a sequence
SKILLS of behaviors should unfold: finite state automata, scripts and skills. Finite state
automata and scripts are logically equivalent, but result in slightly differ-
ent ways about thinking about the implementation. Skills collect behavior-
like primitives called Reaction-Action Packages (RAPs) into a “sketchy plan”
which can be filled in as the robot executes. FSA-type of behavioral coordina-
tion and control were used successfully by the winning Georgia Tech team 19
in the 1994 AAAI Pick Up the Trash event, and the winning LOLA team in the
1995 IJCAI competition for the Pick Up the Trash event. Scripts were used
by the Colorado School of Mines team in the 1995 competition; that entry
performed behaviorally as well as the winning teams but did not place due
to a penalty for not having a manipulator. Those three teams used at most
eight behaviors, even though LOLA had a more sophisticated gripper than
the Georgia Tech team. In contrast, CHIP the second place team in the IJCAI