Third, a deliberative layer must be able for deciding
which behavioral mode (search or track) should be
operating. This is a role provided by the executive
(Williams and Nayak 1997), a system for dispatching
actions, managing time and resources, monitoring
the execution of the plan, and initiating plan changes.
Execution models can be based on procedures or on
finite-state machines (Bohren et al. 2011).
Finally, a deliberative planning layer must be used
to devise a plan for search. Planning for search and
track (and variants such as search and rescue) is one of
the oldest problems in operations research. The
foundations of the theory of search are found in the
paper by Koopman (1957), which divides the problem
into two subproblems: optimal allocation of effort
(that is, what percentage of time to spend in a given
subregion) and optimal rescue track. Planning for
search is potentially challenging because it is assumed
that the target is in an area that is too large to search
exhaustively, the target’s location is represented as a
probability distribution over subregions of the search
area, and the target may or may not be moving. A typical
planning cycle involves the production of a probability
distribution for the object’s location at the time of the
next search. A trajectory uses this distribution along
with a list of assigned search assets to produce
operationally feasible search plans that maximize the
increase in probability of detecting the object. If the
search is unsuccessful, a posterior probability map for
object location that accounts for the unsuccessful search
and the possible motion of the object is generated,
providing the basis for planning the next increment of
search (Kratzke, Stone, and Frost 2010).
These four cognitive components occupy a wide
range of behaviors between the purely reactive and the
purely deliberative. They interact in complex ways that
require an effective and robust coordination mechanism, and they provide a good example of distributed,
hierarchical organization, as we now examine.
for Search and Track
We’ve noted that autonomy architectures are fundamentally hierarchical, distributed, and human
centered. In this section we frame these features in the
context of search and track.
Figure 3 proposes a hierarchical autonomous architecture for search and track. Mission goals are inputs
to the system. A mission controller acts as planner and
plan executive. High-level plans for searching are
generated and dispatched at this level. The ability to
switch between searching and tracking is represented
Figure 4. Testing Autonomous Search and Track Using the Morse Blender Simulation Environment.