Page 282 - Introduction to AI Robotics

Managerial Architectures
7.4 7.4 Managerial Architectures 265
Managerial styles of Hybrid architectures are recognizable by their decom-
position of responsibilities similar to business management. At the top are
agents which do high level planning, then pass off the plan to subordinates,
who refine the plan and gather resources, and pass those down to the lowest
level workers, the reactive behaviors. Higher level agents can see the results
of their subordinate lower level agents (essentially eavesdrop on them), and
can give them directions. As with subsumption, a layer can only modify the
layer below it. In Managerial styles, each layer attempts to carry out its direc-
tive, identify problems and correct them locally. Only when an agent cannot
solve its own problem does it ask for help from a superior agent; the agent is
FAIL UPWARDS said to fail upwards in this case.

7.4.1 Autonomous Robot Architecture (AuRA)

AUTONOMOUS ROBOT Autonomous Robot Architecture (AuRA) is the oldest of the Hybrids. It was
ARCHITECTURE actually designed and implemented by Arkin at the same time Brooks was
(AURA)
beginning to publish his work with subsumption. AuRA is based on schema
theory, and consists of five subsystems, equivalent to object-oriented classes.
Two of the subsystems comprise the deliberative portion: the Planner, and
the Cartographer. The Planner is responsible for mission and task planning.
It is subdivided into three components, equivalent to the Nested Hierarchical
Controller 93 discussed in Ch. 2. The Cartographer encapsulates all the map
making and reading functions needed for navigation. The Cartographer can
also be given an a priori map. For example, a human operator might load in
a map file for the Cartographer to use. The three components of the Planner
would interact with the Cartographer through methods to obtain a path to
follow, broken down into subsegments.
The Planner subsystem is divided into the Mission Planner, Navigator, and
Pilot. The Mission Planner serves as the interface with the human, and the
current implementation of AuRA has one of the most extensive and user
friendly robot interfaces available. The Navigator works with the Cartogra-
pher to compute a path for the robot and break it into subtasks (go over the
mountain to the water tower, follow the road along the ridge to the camp).
The Pilot takes the first subtask (go over the mountain to the water tower)
and gets relevant information (terrain, foliage types, what the water tower
looks like) to generate behaviors. The Pilot portion of the Planning subsys-
tem interacts with the Motor Schema Manager in the Motor subsystem, giv-