Page 292 - Introduction to AI Robotics

275
7.5 State-Hierarchy Architectures

Figure 7.7 Each of the 3T layers running on a separate computer, controlling the
EVAHR robot. (Photograph courtesy of the National Aeronautics and Space Admin-
istration, Johnson Space Center.)

The top layer of 3T is Planner. It fulfills the duties of the mission planner
and cartographer by setting goals and strategic plans. These goals are passed
to the middle layer, called the Sequencer. The Sequencer uses a reactive plan-
ning technique called RAPs to select a set of primitive behaviors from a li-
brary and develops a task network specifying the sequence of execution for
the behaviors for the particular subgoal. The Sequencer is responsible for the
sequencer and performance monitoring functions of a generic Hybrid archi-
tecture. The Sequencer layer instantiates a set of behaviors (called skills)to
carry out the plan. These behaviors form the bottom layer, called the Con-
troller or Skill Manager. In order to avoid confusion with the connotations of
purely reflexive behaviors left over from the Reactive Paradigm, 3T does not
call its behaviors “behaviors.” Its behaviors have the same broader scope as
AuRA and SFX, permitting sensor fusion and assemblages of primitive be-
haviors. The preferred term is “skill” to distinguish its behaviors from the
connotations of behaviors popularized by the subsumption architecture. A
skill is often an assemblage of primitive skills; indeed, one of the interesting
aspects of 3T is its foundation as a tool for learning assemblages.