The visual simulation displays:
• A specific agents traversability or visibility map upon
• A background image unique to that scenario
• Agents moving around the workspace area
• Current agent’s actions
The simulation’s primary goal is to confirm the author’s
claim that implementing the method outlined in Failure An-
ticipation in Pursuit-Evasion will provide the ability to get
assistance to the primary pursuer within the visibility criteria.
This claim is tested using all three scenarios mentioned above.
The secondary goal is to measure the amount of time it takes
to discover potential failures as the time horizon increases. A
regression analysis is performed on the data to determine the
rate of change. The rate of chance is then compared to the
author’s results. The goal is to have the rate of change less
than the O(h3) limit cited by the authors.
Searching problems can be studied using many different
constraints on the problem (Fig. 2). Failure Anticipation in
Pursuit-Evasion mentions about some the constraints the au-
thors decided to use but not all of their decisions. The authors
specified heterogeneous search group, use of a finite discrete
graph and number of targets. This leaves a lot of implemen-
tation details such as target’s motion, pursuer’s motion and
pursuer’s sensory model that are not explicitly stated. The lack
of details on those topics is expected because the algorithm
discussed in the paper does not focus on these items. But,
the lack of these details add uncertainty when attempting to
replicate and evaluate their results.
There are also a few important difference between this
paper’s implementation and the methods described in Failure
Anticipation in Pursuit-Evasion because of a lack of clarity in
the paper and the need to reduce complexity of the project.
First, the authors used large time steps of two seconds.
Using time steps of this size will detract from the presentation
experience of the simulation. The simulation in this paper runs
with a time step of one second which is half the two seconds
that the authors were using. To comfortably perform all of
the computations within one time step the number of time
steps, or time horizon, that the pursuer will look ahead had
to be reduced. A ten second time horizon was used in the
simulation because it’s the farthest look ahead which will fit
within a single time step.
Second, the simulation environments are in 2D and do not
use elevation maps. The paper’s farm environment uses 2.5D
elevation maps in their simulation. Using a two dimensional
map reduces the complexity of collision and point detection
compared to performing these tasks in three dimensions.
Third, the simulation does not implement the full task
bundle auction that is described in the paper. Instead, the
simulation uses single task implementation which auctions off
the single task bundles. Multiple item task bundles have no
value when there will only be one possible task (following