Page 130 - Introduction to AI Robotics

113
4.3 Subsumption Architecture
Higher cohesion is associated with modules that do one thing well, like the
SQRT function in C. The examples in Sec. 4.3 and 4.4 attempt to illustrate the
choices a designer has in engineering the behavioral software of a robot.

4.2.3 Representative architectures
In order to implement a reactive system, the designer must identify the set
of behaviors necessary for the task. The behaviors can either be new or use
existing behaviors. The overall action of the robot emerges from multiple,
concurrent behaviors. Therefore a reactive architecture must provide mecha-
nisms for 1) triggering behaviors and 2) for determining what happens when
multiple behaviors are active at the same time. Another distinguishing fea-
ture between reactive architectures is how they define a behavior and any
special use of terminology. Keep in mind that the definitions presented in
Sec. 4.2 are a generalization of the trends in reactive systems, and do not
necessarily have counterparts in all architectures.
There are many architectures which fit in the Reactive Paradigm. The two
best known and most formalized are the subsumption and potential field
methodologies. Subsumption refers to how behaviors are combined. Poten-
tial Field Methodologies require behaviors to be implemented as potential
fields, and the behaviors are combined by summation of the fields. A third
RULE ENCODING style of reactive architecture which is popular in Europe and Japan is rule
encoding, where the motor schema component of behaviors and the com-
bination mechanism are implemented as logical rules. The rules for com-
bining behaviors are often ad hoc, and so will not be covered in this book.
Other methods for combining behaviors exist, including fuzzy methods and
winner-take-all voting, but these tend to be implementation details rather
than an over-arching architecture.

4.3 Subsumption Architecture

Rodney Brooks’ subsumption architecture is the most influential of the purely
Reactive Paradigm systems. Part of the influence stems from the publicity
surrounding the very naturalistic robots built with subsumption. As seen
in Fig. 4.5, these robots actually looked like shoe-box sized insects, with
six legs and antennae. In many implementations, the behaviors are em-
bedded directly in the hardware or on small micro-processors, allowing the
robots to have all on-board computing (this was unheard of in the processor-
impoverished mid-1980’s). Furthermore, the robots were the first to be able