Page 306 - Introduction to AI Robotics

289
7.11 Exercises
rule of thumb is that functions which operate on symbolic information go in
the deliberative layer, or deliberator, while functions which transform sensor
data into actuator commands go in the reactive layer, or reactor. The reactive
component is organized by behaviors, but the definition of behavior is wider
than the purely reflexive behaviors in the Reactive Paradigm. Behaviors may
be called skills or by other names to eliminate confusion with reflexive be-
haviors. Hybrid architectures also exhibit a wide variety of mechanisms for
combining behaviors, including subsumption, potential field combination,
filtering, voting, and fuzzy logic. The reactive portion tends to make more
use of assemblages of behaviors than in the Reactive Paradigm.
The deliberative component is often subdivided into layers. The layering
may reflect scope in managerial responsibilities or in time horizon. Layers
often consist of modules implemented as software agents. These agents may
have agent-specific world models or share a single global world model. The
world model can also serve as a virtual sensor for use by behaviors.
Managerial styles of architectures favor a bottom-up organization, concen-
trating on layering planning on top of behaviors and behavior management.
Deliberation is based on whether symbols are being manipulated. The over-
all behavior emerges as a function of the behavioral layer. State-hierarchies
divide deliberation and reaction by the state, or scope of knowledge, avail-
able to the modules or agents operating at that layer. The three states are
Past, Present, Future. Responsibilities are organized into planning and se-
quencing components. The overall emergent behavior is more due to the se-
quencing of behaviors, or skills, rather than concurrency. Model-based styles
favor a top-down organization, focusing on the creation and maintenance of
a global world model. The world model then serves as a virtual sensor to
the behaviors and as a data structure for planning. Often the software agents
within the deliberative layer in all three architectural styles exhibit self mon-
itoring and problem solving abilities, and fail upwards if the mission cannot
be accomplished. As such, they distribute planning and problem-solving.

7.11 Exercises

Exercise 7.1
Describe the Hybrid paradigm in terms of:

a. sensing, acting, and planning, and
b. sensing organization.