Pursuit Curves

Katy Steiner, Jonah Franchi

5/10/2011
Abstract
The purpose of this paper is to demonstrate and explain how pursuit
curves are derived and modeled, give examples of dierent types of pursuit
curves, and nally show a more complex pursuit curve in 3-D space.
1
1 Introduction & Assumptions
A pursuit curve is, as its name implies, a curve showing the path an object takes
as it pursues another object. The velocity vector of the pursuer is always going
directly towards the prey, which excluding the idea of bending space time, is
a straight line. In order to model pursuit curves a few assumptions must be
made. First, all prey are following a set path at a constant speed. Second, there
is nothing in the way of either prey or predator and they have unlimited energy.
2
2 A Basic Pursuit Curve
(All of the proceeding in the section was found and derived by
http://online.redwoods.cc.ca.us/instruct/darnold/deproj/sp08/mseverdia/pursuit.pdf)
For our rst example of a pursuit curve, consider our predator being a missile
and our prey, a ship.
The ship starts at (x
0
, 0) at t 0 and is traveling at V
s
along the vertical
line x = x
0
. The missile starts at (0, 0) at time t = 0 and travels at a constant
speed of V
m
along a curved path so that its velocity vector is pointing directly
at the ship.
If the missile at time t 0 is the point (x, y) and the position of the ship is
(x
0
, V
s
t) then the slope of the tangent to the pursuit curve at (x, y) is
dy
dx
=
y V
s
t
x x
0
Solve for t to get
t =
y
V
s

x x
0
V
s
dy
dx
(1)
The missile has always traveled a distance of V
m
t along the pursuit curve so
using the arc length formula we get
V
m
t =
_
x
0

1 +
_
dy
dx
_
2
dz
Then we solve this for t and get
t =
1
V
m
_
x
0

1 +
_
dy
dx
_
2
dz (2)
Combining equation (1) and (2) gives
1
V
m
_
x
0

1 +
_
dy
dx
_
2
dz =
y
V
s

x x
0
V
s
dy
dx
3
Let p(x) = dy/dx
1
V
m
_
x
0
_
1 + [p(z)]
2
dz =
y
V
s

x x
0
V
s
p(x)
Dierentiate both sides
1
V
m
_
1 + [p(x)]
2
=
1
V
s
dy
dx

x x
0
V
s
dp
dx

1
V
s
p(x)
Replace dy/dx with p(x) and reduce
(x x
0
)
dp
dx
= n
_
1 + [p(x)]
2
(3)
Where n = V
s
/V
m
, which is our relative speed.
Next we solve this rst by separating the variables, letting p=p(x)
dp
_
1 + p
2
=
ndx
x x
0
Integrate both sides with a table of integrals
ln (p +
_
1 + p
2
) + C = nln (x
0
x) (4)
Since p = dy/dx, and the missiles velocity is always pointed directly toward
the ship, at time t = 0 the missiles velocity is dy/dx = 0. Solving for C, we nd
that C = nln (x
0
).
Insert this result into equation (4) to get
ln (p +
_
1 + p
2
) + ln (x
0
x)
n
ln (x
0
)
n
= 0
Using the properties of logarithms we can say
0 = ln (p +
_
1 + p
2
) + ln
_
x
0
x
x
0
_
n
0 = ln
_
(p +
_
1 + p
2
)
_
1
x
x
0
_
n
_
Since ln(1) = 0
(p +
_
1 + p
2
)
_
1
x
x
0
_
n
= 1
Now we try to solve for p
p +
_
1 + p
2
=
_
1
x
x
0
_
n
Let q = (1 x/x
0
)
n
4
p +
_
1 + p
2
= q
1 + p
2
= (q p)
2
1 + p
2
= q
2
2qp + p
2
2qp = q
2
1
p =
1
2
_
q
1
q
_
Substitute back in for p and q
dy
dx
=
1
2
_
_
1
x
x
0
_
n

_
1
x
x
0
_
n
_
Now integrate with respect to x
y(x) + C =
1
2
_ _
1
x
x
0
_
n
dx
1
2
_ _
1
x
x
0
_
n
dx
Let u = 1 x/x
0
, du = (1/x
0
)dx
y(x) + C =
1
2
_
x
0
u
n
du
1
2
_
x
0
u
n
du
y(x) + C =
1
2
x
0
_
u
n+1
n + 1
_
+
1
2
x
0
_
u
n+1
n + 1
_
Simplify
y(x) + C =
1
2
x
0
_
u
n+1
n + 1

u
n+1
n + 1
_
y(x) + C =
1
2
x
0
u
_
u
n
n + 1

u
n
n + 1
_
Substitute back in for u
y(x) + C =
1
2
x
0
_
1
x
x
0
_
_

_
_
1
x
x0
_
n
n + 1

_
1
x
x0
_
n
n + 1
_

_
y(x) + C =
1
2
(x
0
x)
_

_
_
1
x
x0
_
n
n + 1

_
1
x
x0
_
n
n + 1
_

_
5
Since the missile starts at the origin y(0) = 0 we can solve for C
C =
1
2
x
0
_
1
1 + n

1
1 n
_
C =
1
2
x
0
_
2n
1 n
2
_
C =
1
2
x
0
_
2n
1 n
2
_
We now have our solution to (3)
y(x) =
1
2
(x
0
x)
_
(1 x/x
0
)
n
1 + n

(1 x/x
0
)
n
1 n
_
+
n
1 n
2
x
0
where n = V
s
/V
m
Solving and graphing in Matlab produces the following graph:
The above method just shown uses hand calculations in the most basic case
to solve for the path of the predator. This method proves to be a bit lengthy for
6
our purposes, so we will use a dierent method in the next sections. This method
will allow for us to enter our equations into our numerical solver, MATLAB,
and produce graphs of the pursuit curves.
3 A Circular Pursuit Curve
(All of the proceeding in the section was found and derived by
http://online.redwoods.cc.ca.us/instruct/darnold/deproj/sp08/mseverdia/presentation.pdf)
Pursuit curves can take on many forms. Apart from a linear or parabolic
form, pursuit curves can also be circular. For our example, consider our predator
to be a bully and our prey to be a nerd. The nerd runs in a circular path in a unit
circle and is chased by a bully from the inside. Our bully will be represented by
B(x(t), y(t)), and our nerd by N(p(t), q(t)). The path of the nerd is represented
by a vector equation:
N(t) = p(t)i + q(t)j
The pursuit curve of the bully is also represented by a vector equation:
B(t) = x(t)i + y(t)j
The path of the nerd is described by p and q as functions of time. For
the pursuit curve of the bully, x and y are unknown because they depend on
the path of the nerd. We nd x and y by developing a system of dierential
equations using two unit vectors.
A unit vector is a vector whose magnitude is 1. Its formed by dividing any
one vector, n by its magnitude.
u =
n
|n|
The rst unit vector used is the unit tangent vector, T, or the unit vector
tangent to B. V
B
will represent the velocity of the bully and V
N
is the velocity
of the nerd. We know that |V
B
| = k|V
N
| because the velocities are proportional
(the bully is running at k relative to the speed of the nerd). Using this knowledge
we nd T.
Note that k is the same as n in our previous section, but as this was taken
from a dierent source we left the values as the originals were.
T =
V
B
|V
B
|
=
V
B
k|V
N
|
The second unit vector is the unit dierence vector, D, in the direction of
BN:
D =
NB
|NB|
The unit vectors T and D are equal because the bully is always heading
toward the nerd.
T = D
7
Using T and D, we can solve for V
B
:
V
B
k|V
N
|
=
NB
|NB|
V
B
= k|V
N
|
NB
|NB|
Converting V
B
into vector notation will allow us to equate coecients and
develop a system of dierential equations.
V
B
= k|V
N
|
NB
|NB|
= k

_
dp
dt
_
2
+
_
dq
dt
_
2
(p x)

i + (q y)

j
_
(p x)
2
+ (q y)
2
Equating coecients gives:
dx
dt
= k

_
dp
dt
_
2
+
_
dq
dt
_
2
(p x)
_
(p x)
2
+ (q y)
2
dy
dt
= k

_
dp
dt
_
2
+
_
dq
dt
_
2
(q y)
_
(p x)
2
+ (q y)
2
Now we can use these equations for any k, p, and q.
So for our example, we will let the bully chase the nerd at 75% of the nerds
speed, so k = .75. Since the nerd is running in a unit circle, our equations are:
p(t) = cos t
q(t) = sin t
dp
dt
= sin t
dq
dt
= cos t
When we substitute k, p, and q into our general dierential equations, we
get the following system of dierential equations
dx
dt
= (.75)
_
(sin t)
2
+ (cos t)
2
(cos t x)
_
(cos t x)
2
+ (sin t y)
2
dy
dt
= (.75)
_
(sin t)
2
+ (cos t)
2
(sin t y)
_
(cos t x)
2
+ (sin t y)
2
which can be simplied.
dx
dt
= (.75)
(cos t x)
_
(cos t x)
2
+ (sin t y)
2
8
dy
dt
= (.75)
(sin t y)
_
(cos t x)
2
+ (sin t y)
2
Solving and graphing in Matlab produces the following graph:
This shows that as the bully chases the nerd following a standard pursuit
path, as long as his speed isnt at least equal to the speed of the nerd he wont
even be on the same path as him.
4 Other Pursuit Curves
Our next set of equations show other variations of paths that these curves can
take.
4.1 Spider Curve
Given the path of the prey to be:
p(t) = t sin t
q(t) = 8 cos 6t
dp
dt
= t cos t + sint
dq
dt
= 48 sin 6t
9
We will let k = .7. Our dierential equations are as follows:
dx
dt
= (.7)
_
(t cos t + sint)
2
+ (48 sin 6t)
2
(t sin t x)
_
(t sin t x)
2
+ (8 cos 6t y)
2
dy
dt
= (.7)
_
(t cos t + sint)
2
+ (48 sin 6t)
2
(8 cos 6t y)
_
(t sin t x)
2
+ (8 cos 6t y)
2
Solving and graphing in Matlab produces the following graph:
Because the speed of the predator doesnt match the speed of the prey the
afterimage of the path is slightly below the prey at most times.
4.2 Lion and Gazelle
Our next example will be a lion chasing a gazelle with the following equations:
p(t) = cos t
q(t) = sin 3t
dp
dt
= sin t
dq
dt
= 3 cos 3t
Solving and graphing in Matlab produces the following graph:
10
5 A 3D Pursuit Curve
Now that we have delved a bit into pursuit curves in 2 dimensional space, we will
now go into how pursuit curves work in 3 dimensional space. A third variable
is needed for pursuit curves in 3 dimensions so we will have three dierential
equations. Thus the dierential equations we will use look like the following:
dx
dt
= k

_
dp
dt
_
2
+
_
dq
dt
_
2
+
_
dr
dt
_
2
(p x)
_
(p x)
2
+ (q y)
2
+ (r z)
2
dy
dt
= k

_
dp
dt
_
2
+
_
dq
dt
_
2
+
_
dr
dt
_
2
(q y)
_
(p x)
2
+ (q y)
2
+ (r z)
2
dz
dt
= k

_
dp
dt
_
2
+
_
dq
dt
_
2
+
_
dr
dt
_
2
(r z)
_
(p x)
2
+ (q y)
2
+ (r z)
2
Since three equations are needed to graph a pursuit curve in 3D, we will let
the third equation equal t in order to produce a graph of our curve verses time
for our lion verses gazelle example. In this example, since it is our rst in 3-D,
we are simply projecting our graph into the t plane and not really making a
new curve.
Our equations are as follows:
p(t) = cos t
11
q(t) = sin 3t
r(t) = t
dp
dt
= sin t
dq
dt
= 3 cos 3t
dr
dt
= 1
Solving and graphing in Matlab produces the following graph:
Solving and graphing our Spider equation using the same r equation pro-
duces the following graph:
12
6 A Pursuit Curve on a 3D Surface
Graphing pursuit curves in 3D opens up the possibility of graphing a pursuit
curve on a 3D surface. In order to do this the surface must be dened as a
function of x and y. That way the pursuit curve will be directly graphed on the
surface. In this example, the lion verses gazelle model will be used. The lion is
pursuing the gazelle at 30% of the gazelles speed and the surface equation is
Z = 20 sin xcos y.
Our equations are as follows:
p(t) = cos t
q(t) = sin 3t
r(t) = 20 sin xcos y
Solving and graphing in Matlab produces the following graph:
13
7 Varying Speeds
Another way to change up the pursuit curve equations is to vary the speed at
which the predator chases the prey.
7.1 Bully and Nerd
The rst example will be variation of a 2-Dimensional circular pursuit curve.
Because our coordinate plane is divided into four quadrants, the velocity will
vary depending on which quadrant the predator is in. For quadrants I, II,
III, and IV, the predator will be chasing the prey at 30%, 100%, 30%, 100%
respectively. This is accomplished in Matlab by dening k to equal .3, 1, .3, 1
for the specied quadrants.
Solving and graphing in Matlab produces the following graph:
14
This gives us a path that is a smashed circle. In the end, as with the basic
circular pursuit curve, the bully ends up following the same path innitely.
7.2 Lion and Gazelle
Going on to vary the velocity for a 3-Dimensional curve follows the same prin-
ciple. The velocity variable, k, will be dened for certain values.
Solving and graphing in Matlab produces the following graph:
15
Although the graph shows that the path of the two cross we cant actually
say that the lion catches the gazelle. Just because their paths cross doesnt
mean that they were there at the same time. In order to determine this we
would need to be able to plot two comet plots on the same graph at the same
time. Unfortunately our knowledge of Matlab is insucient to allow us to do
that.
8 Conclusion
In real life pursuit curves are used to model the path that a predator follows
prey. This doesnt limit the applications to simply animals, but can be used in
any instance where one object is following another, such as our idea of a missile
following a ship. In a more realistic scenario the prey would not just follow a set
path, but would change its course depending on where the predator was. There
would also be dead zones where neither of the objects could travels, such as a
tree being in the way, which would change the path. This doesnt degrade their
usefulness though seeing as the can be and are used to model guided missile.
16
9 M-Files
These are the general codes we used from which we derived all the others.
General 2-D codes-
1)
function Yprime=general(t,y,flag,k)
Yprime = zeros(2,1);
P=t*sin(t);
Q=8*cos(6*t);
dP=t*cos(t)+sin(t);
dQ=-48*sin(6*t);
Yprime(1)=k*(sqrt((dP)^2+(dQ)^2)*(P-y(1)))/(sqrt((P-y(1))^2+(Q-y(2))^2));
Yprime(2)=k*(sqrt((dP)^2+(dQ)^2)*(Q-y(2)))/(sqrt((P-y(1))^2+(Q-y(2))^2));
2)
t=linspace(0,8*pi,500);
x=t.*sin(t);
y=8*cos(6*t);
[t,Y]=ode45(general,[0,8*pi],[0;0],[],0.75);
h=plot(x,y,b,Y(:,1),Y(:,2),r);
legend(h, Prey,Predator);
title(Spider Pursuit Curve)
xlabel(x)
ylabel(y)
axis equal;
shg
General 3-D Codes-
1)
Adding these 3 lines adds a 3rd parameter to our original pursuit equations.
R=t;
dR=1;
Yprime(3)=k*(sqrt((dP)^2+(dQ)^2+(dR)^2)*(R-y(3)))/(sqrt((P-y(1))^2+(Q-y(2))^2+(R-y(3))^2));
2)
This code graphs the new 3rd parameter.
t=linspace(0,2*pi,500);
x=cos(t);
y=sin(3*t);
z=t;
[t,Y]=ode45(general3d,[0,2*pi],[0;0;0],[],0.3);
h=plot3(x,y,z,b,Y(:,1),Y(:,2),Y(:,3),r);
legend(h, Prey,Predator);
title(Predator vs. Prey 3D)
17
xlabel(x)
ylabel(y)
zlabel(z)
grid on
General 3-D Surface Codes
By adding these 4 lines to our previous code we can add a surface to our 3-D graphs.
Z changes depending on our z value in the previous codes.
hold on
[a,b] = meshgrid(-2:.2:2, -2:.2:2);
Z = exp(cos(a)+sin(b));
mesh(a,b,Z)
hidden off
General Varying Speed-
To get this we simply changed our k from a constant value and made it like this...
k=.3*(P<0)+.9*(P>0);
References
http://online.redwoods.cc.ca.us/instruct/darnold/deproj/Sp98/PeterG/index.htm
http://online.redwoods.cc.ca.us/instruct/darnold/deproj/sp08/mseverdia/presentation.pdf
http://online.redwoods.cc.ca.us/instruct/darnold/deproj/sp08/mseverdia/pursuit.pdf
18