Strategies for Multi-Asset Surveillance

UW Computer Science DepartmentUW Computer Science Department

Strategies for Multi-Asset Surveillance

Dr. William M. Spears

Dimitri Zarzhitsky

Suranga Hettiarachchi

Wesley Kerr

University of Wyoming


Scenario

Foliage detector

Target detector

Maximize the number of T targets found by α assets.


Forest Generator

L x L environmentwith T targetsand foliage.


Asset Control

• Behavior-based asset controllers.– Straight Line (SL)

• Assets “bounce” off boundary walls. Ignores foliage.

– Straight Line Avoid Forest (SLAF)• Like SL but also reverse course if encounter foliage.

– Super Straight Line Avoid Forest (SSLAF)• Like SLAF but move opposite to center of mass of

foliage (a more sophisticated foliage sensor).


Target Control

• Stationary targets for baseline study.

• “Hiding Gollum” target controller:– Targets try to cross from left to right through

environment while hiding in foliage.


Stationary Targets

Why is SLAF so poor and SSLAF so good?

0

20

40

60

% Targets Found

10 20 30 40 50 60 70

% Foliage

Performance on Stationary Targets

SL

SLAF

SSLAF


Asset Coverage Maps

SL SLAF SSLAF

SL provides uniform coverage of the space. SSLAF provides increaseduniform coverage of the non-foliage space. But SLAF misses entire regions.


Gedanken Experiment

What if the targets move slowly from left to right? Will the prior results change?


Gollum Targets

Why is SLAF so good?

0

20

40

60

80

% Targets Found

10 20 30 40 50 60 70

% Foliage

Performance on Gollum Targets

SL

SLAF

SSLAF


Probabilistic AnalysisController 1:Uniformly coverwhole area (like SL).

Controller 4:Uniformly coverone row (worst case SLAF).

Controller 2:Uniformly coverone column (bestcase SLAF).

Controller 3:Uniformly coverone diagonal (average case SLAF).


Area Controller

t

t

t

tt

t

S

rv

r

v

v

LS

r

LS

STE

t

2cityasset velo

asseton detector target of radius

ocitytarget vel

assets ofnumber

111found] targets[

2

2

Expected number of timesteps for asset to cover area.

Visibility timeof target.


Column Controller

t

t

t

t

tt

S

rd

rv

r

v

v

LS

d

LS

STE

t

2thcolumn wid

2cityasset velo


ocitytarget vel

assets ofnumber

111found] targets[


Diagonal Controller

t

t

t

t

tt

S

rd

rv

r

v

v

LS

d

LS

STE

t

2thcolumn wid

2cityasset velo


ocitytarget vel

assets ofnumber

22111found] targets[


Row Controller

height row2

2found] targets[

t

t

rL

TrE


Comparison of Controllers

SLAF works well on moving targetsbecause diagonal controller performance is like column controller performance.

Comparison of Controllers

0

0.2

0.4

0.6

0.8

1

1.2

0 .2 .4 .6 .8 1.0 1.2 1.4 1.6 1.8

target velocity

% t

arg

ets

fo

un

d Area Controller

Colum n/DiagonalController

Row Controller


Summary

• Developing predictive mathematical theory for multiple assets performing surveillance.– Currently includes number of assets, their speed, target

speed, and environment size.

– Working on including probability of detection (a noisy sensor), percentage of foliage, and time limits on mission length.

• Goal is to provide mathematical tools to yield an optimal strategy for a surveillance mission.

Strategies for Multi-Asset Surveillance

Documents