Bad Guys Are Sneaky
The Problem
Bad guys can be followed by UAVs, but what if they hide behind something?
The Solution
Develop an algorithm that allows one or more UAVs to follow bad guys even through brief periods of being hidden behind buildings, or garbage cans
Optimization
Maximize score / #UAVs
This penalizes for needing a lot of UAVs
The Algorithm
Generate probabilistic future paths
Possible actions
Wait for the bad guy to come out
Plan a path to where we think he's going
Let him go because he's not that important
The Algorithm - Watershed
Generate a watershed to get the viewable area from any given point
Use that to influence the score
http://www.redblobgames.com/articles/visibility/
The Algorithm - Likelihood
Scoring
Assign importance(p) to different targets
Store the amount of time each target was in frame
Multiply time in frame by importance
Normalize by the score if all targets were in frame for the duration of the test
score_{raw} = \sum_{n=1}^{E} t_n * p_n
s
c
o
r
e
r
a
w
=
∑
n
=
1
E
t
n
∗
p
n
score_{raw} = \sum_{n=1}^{E} t_n * p_n
s
c
o
r
e
r
a
w
=
∑
n
=
1
E
t
n
∗
p
n
score_{max} = \sum_{n=1}^{E} t_{tot} * p_n
s
c
o
r
e
m
a
x
=
∑
n
=
1
E
t
t
o
t
∗
p
n
score_{max} = \sum_{n=1}^{E} t_{tot} * p_n
s
c
o
r
e
m
a
x
=
∑
n
=
1
E
t
t
o
t
∗
p
n
score_{norm} = \dfrac{score_{raw}}{score_{max}}
s
c
o
r
e
n
o
r
m
=
s
c
o
r
e
r
a
w
s
c
o
r
e
m
a
x
score_{norm} = \dfrac{score_{raw}}{score_{max}}
s
c
o
r
e
n
o
r
m
=
s
c
o
r
e
m
a
x
s
c
o
r
e
r
a
w
Results
Results
Made with Slides.com