Available online at www.sciencedirect.

com

Expert Systems
with Applications
Expert Systems with Applications 35 (2008) 235244
www.elsevier.com/locate/eswa

A base function for generating contour traversal paths in

stereolithography apparatus applications
Shuo-Yan Chou
a

a,*

, Chang-Chien Chou

a,b

, Yu-Kumg Chen

Department of Industrial Management, National Taiwan University of Science and Technology, No. 43, Sec. 4, Keelung Rd., Taipei City 106, Taiwan, ROC
b
Department of Information Management, Lunghwa University of Science and Technology, No. 300, Sec. 1, Wanshou Rd., Guishan,
Taoyuan County 333, Taiwan, ROC
c
Department of Electronic Engineering, Huafan University, No. 1, Huafan Rd., Shihding Township, Taipei County 223, Taiwan, ROC

Abstract
The technique of layered manufacturing in rapid prototyping is to fabricate product prototype by scanning the cross-sectional contours of the product using a laser beam layer by layer. The outlines of geometrical objects on each layer are dierent, and each layer may
contain several geometrical objects. In order to simplify the problem, each geometrical object on the same plane is approximating by its
own minimum circumscribed circle. Therefore, the minimum traversal path of circles can be the based model of the minimum traversal
path of scanning geometrical objects. Furthermore, the minimum traversal path of three circles is the degenerated case of the minimum
traversal path of plural circles. And the problem of the minimum traversal path of three circles can be transferred to the problem of the
minimum traversal path of one circle and two points in this paper. By using the concepts of reection of light in physics and geometrical
mathematics, the equation of the minimum traversal path of three circles is derived in this paper. This equation can be easily implemented in many areas of application, including robotic motion planning and path planning for submarine, ship, and airplane.
2007 Published by Elsevier Ltd.
Keywords: Minimum traversal function; Stereolithography apparatus; Layered manufacturing; Path planning; Robotic motion planning; Analytic geometric function

1. Introduction
In order to provide fast and precise computation of the
shortest path for the moving robot (Khatib, Craig, & Lozano-Prez, 1989; Nehmzow & Owen, 2000), circles are used
to embrace obstacles in the robotic path planning. For
example, there is an application on the mission planning
software for the PHONIX autonomous underwater vehicle
(AUV) (Brutzman, 1994; Leonhardt, 1999) of the US navy.
It encloses the underwater obstacles by circles to avoid collisions that lead to the catastrophic loss of the vehicles. By
using the tangents among circles, the PHONIX system
derives the shortest path for the underwater vehicle rapidly

Corresponding author.
E-mail address: sychou@im.ntust.edu.tw (S.-Y. Chou).

0957-4174/$ - see front matter 2007 Published by Elsevier Ltd.

doi:10.1016/j.eswa.2007.06.025

as shown in Fig. 1. Quick deriving a precise shortest path is

critical and crucial on the battleeld.
In the eld of layered manufacturing (LM) of rapid prototyping, the well-known processes stereolithography
apparatus (Jacobs, 1992) and laminated object manufacturing (LOM) (Chiu & Liao, 2003) fabricate the prototype
of the product object with scanning the cross-sectional contours of the object layer by layer (Lan, Chou, Chen, &
Gemmill, 1997). In order to minimize the total time for
completing the fabrication, the minimization of the processing time for each scanning layer is thus essential. For
each scanning layer, a laser beam scans along the contours
in the layer to solidify the cross-section of the object. The
path planning (Murty, 1995) for the scanning on each layer
is therefore a traversal optimization problem with respect
to the geometric entities in a two dimensional plane (Majhi,
Janardan, Smid, & Gupta, 1999; Wah, Murty, Joneja, &

236

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

Planned path

Original path
Obstacle
Safe radius path

Fig. 1. Result of deriving a shortest path of PHONIX collision avoidance

system.

Leung, 2002). Fig. 2 illustrates the concepts of a LM example in rapid prototyping.

Since the shapes of objects on each scanning plane are
dierent, it is very dicult to gure out the minimum traversal path among these objects. By using some smallest
circumscribed circles to approximate these objects, the
problem can be simplied as nding the minimum traversal
path between circles (Tang & Pang, 2003). Therefore, nding the minimum traversal path between circles can be used
in the path planning of LM. Finding the minimum traversal path of three circles is the fundamental case of nding the minimum traversal path of plural circles. The

problem resolved in this paper is therefore can be dened

as follows. Given three disjoint circles, nd the path that
traverses all three circles in a predened sequence for which
the sum of the circumferences of the three circles and the
two connecting links is a minimum.
Fig. 3 shows three circles C1, C2, and C3 with their centers O1, O2, O3, and radii r1, r2, r3, respectively. The starting point is at point P1. Traverse the circumference of C1
and then return back to the point P1. Traverse the link
P 1 P 2 and then arrive the point P2 on C2. Traverse the circumference of C2 and then return back to the point P2.
Then traverse across the link P 2 P 3 to reach the point P3
on C3. Finally, traverse the circumference of C3 and come
back to the end point P3 to nish up this traverse. To get
the minimum length of this traversal path is the goal of this
paper.
In this paper, the problem of the minimum traversal
path of three circles can be further degenerated to the problem of the minimum traversal path of two points and one
circle. With applying the light reection (or refraction) phenomenon and geometrical mathematics, the minimum traversal function can be derived accordingly.
The fashion of using circles approximating objects is not
only applying in the robotic motion planning (de Berg, van
Kreveld, Overmars, & Schwarzkopf, 2004), the robotic
arms rotation (ORourke, 1997), but also welcome in using
disks to calculate the connectivity of objects (Guibas,
Hershberger, Suri, & Zhang, 2001) of kinetic data structure

Fig. 2. The processes of the layered manufacturing in rapid prototyping.

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

r2

O2

237

resents the distance between point U and point V. And the

function cirf(C) represents the length of the circumference
of circle C. Thus, we can write

C2

L03 minfdistP 1 P 2 distP 2 P 3 cirf C 1

cirf C 2 cirf C 3 g
P2
P3
r3

P1
r1 O 1

C3
O3

C1

Fig. 3. Traversal path of three circles.

in computational geometry. There are many studies concerning the connectivity of the static stationary disks (Preparata & Shamos, 1985), such as the base stations of mobile
phone communication. Also there are many complex studies in computing the connectivity of moving disks (Johnson,
1994; Toh, 1996), such as the Ad-Hoc mobile network. Due
to the characteristics of radio wave, the communication of
geodesy and the satellite constellation (Wood, 2001) are
usually representing the radio coverage by disks, too.
Besides, there are many related researches, such as circle
packing (Graham, Lagarias, Mallows, Wilks, & Yan,
2005) and the problem of circle visibility (Kaiser, 2002) that
makes planning the shortest path which walks through circles and can be seen by one of the circles in anytime. Versatile studies have shown the convenience and practicability
of the fashion of taking circles representing objects.
The rest of this paper is organized as follows. The problem transformation that converts three circles problem into
a two points and one circle problem is given in the next section. The derivation of minimum traversal function is then
presented in Section 3. The proposed algorithm of the problem rotation is presented in Section 4. Concluding remarks
and potential applications are provided in Section 5.

where P1, P2, and P3 locate on the circumference of C1, C2,

and C3 respectively. Since the radii of the given three circles
are constants, this property is to provide the following
lemma.
Lemma 1. Given three disjoint circles C1, C2, and C3 in the
two dimensional plane. The problem E03 can be reduced to the
problem E21 for one circle C2 with two points O1 and O3 by
adding a constant value.
Proof . Since the radii of the given three circles are constants, the traverses of the circumferences of these circles
are also constants. Eq. (1) can be rewritten as
L03 minfdistP 1 P 2 distP 2 P 3 cirf C 2 g
cirf C 1 cirf C 3 :

With adding and subtracting the two lengths of radii r1 and

r3 into Eq. (2), we have
L03 minfdistP 1 P 2 distP 2 P 3 cirf C 2
distO1 P 1 distO3 P 3 g cirf C 1
cirf C 3  distO1 P 1  distO3 P 3

where dist(O1P1) and dist(O3P3) are the length of r1 and r3,

respectively. Since a line with minimum path from a point
to a circle must pass through the center of the circle, the
three points P2, P1, and O1 are collinear and the three
points P2, P3, and O3 are also collinear, too. The Eq. (3)
can be rewritten as
L03 minfdistO1 P 2 distO3 P 2 cirf C 2 g cirf C 1
cirf C 3  distO1 P 1  distO3 P 3 :
Observing that the value of cirf(C1) + cirf(C3) 
dist(O1P1)  dist(O3P3) in equation is constant, we set a
constant variable K to substitute it and then have

2. Problem transformation

L03 minfdistO1 P 2 distO3 P 2 cirf C 2 g K

In this paper, we use Exy to represent the problem of nding the minimum traversal path for x distinct points and y
disjoint circles in a two dimensional plane. The goal of the
proposed paper is to nd the minimum traversal function
of three disjoint circles C1, C2, and C3. This problem is thus
to be represented as E03 , i.e., nding the minimum traversal
path for 0 distinct points and 3 disjoint circles in a two
dimensional plane. The path of E03 is an open route, i.e.,
the starting point of the traverse needs not to be coincided
with the end point of the traverse.
Under such E03 condition, we denote the length of the
minimum traversal path by L03 . The function dist(UV) rep-

Since the term min{dist(O1P2) + dist(O3P2) + cirf (C2)} is

the E21 problem for one circle C2 with two points O1 and
O3, the problem E03 can be reduced to the problem E21 by
adding a constant value. This constant value is the sum
of the two circumferential lengths of circles C1 and C3,
and is subtracted by the lengths of the two radii r1 and
r 3.
h
Since the term cirf(C2) in min{dist(O1P2) + dist(O3P2) + cirf(C2)} is constant, the minimum traversal
length L21 of E21 can be simplied as
L21 minfdistO1 P 2 distO3 P 2 g cirf C 2 :

238

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

y
B

x
O

C
Fig. 5. An example of the variations of h and (h).
Fig. 4. An example of

E21

problem.

With replacing the C2 by C, P2 by P, O1 by A, O3 by B and

removing the constant term cirf(C2), the problem can be
simplied as nding the minimum traversal length L of
the sum of the two interconnecting line segments AP and
PB, so that
L minfdistAP distPBjP 2 Cg:

In order to simplify the derivation of equation, we assume

that the center of circle C coincides with the origin. Fig. 4
shows the relationship between points A, B, P, and circle C
for an E21 problem. Let (xA, yA), (xB, yB), and (xP, yP) be the
coordinates of points A, B, and P, respectively. Let
(xP, yP) be the distance function of the real variables xP
and yP. It is dened by the equation
q
xP ; y P xA  xP 2 y A  y P 2
q
5
xB  xP 2 y B  y P 2 :
If we introduce the polar coordinates
xP r cos h

y P r sin h

then (xP, yP) becomes (r, h). Since the radius r in this paper is constant, the function (r, h) can then be denoted as
(h). Direct substitution in Eq. (5) yields
q
2
2
h xA  r cos h y A  r sin h
q
2
2
6
xB  r cos h y B  r sin h
Figs. 5 and 6 illustrate examples of the relationships of h
and (h) for E21 problem. The diagrams show samples for
(h) in the interval [0, 2p] and [0, 4p], respectively. The E21
example in Fig. 5 contains one circles C with its radius
length equal to 2 and two points A and B with their coordinates (3, 4) and (4, 5), respectively. Observing that the
shape of the discrete curve is approximately sinusoidal
curve. Nevertheless, not all cases appear in the sinusoidal

Fig. 6. Another example of the variations of h and (h).

curve as shown in Fig. 6. The E21 example in Fig. 6 contains

one circle C with its radius length equal to 2 and two points
A and B with their coordinates (3, 1) and (4, 1),
respectively. Since the points A and B are xed and the variable point P lies on the circumference of circle C, the periods of Figs. 5 and 6 are 2p and 4p. Although there is one
minimum value in Fig. 5 when h is equal to p/2, there
are two minimum values appeared in Fig. 6 with h equal
to 1.4p and 1.9p. Therefore, the derived solution for E21
problem is not unique.
3. Derivation of equations
There are many approaches to pursue the extreme values. The best-known approach is using the rst-order
derivative for function (h) with respect to h to be equal
to zero. With applying the non-linear programming or
numerical analysis methods, we also can nd the extreme
value or approximate extreme values. However, the general

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

239

root function of the E21 problem cannot be solved by using

all of these methods. In this paper, we propose a simpler
method which combines the law of light reection in physics and geometrical mathematics to nd the optimal value
of P which results in the minimum traversal path of the
E21 problem. Let the refractive index of the medium with
the incident ray be n1 and that of the medium with the
refractive ray be n2. Fig. 7 illustrates the refraction of light
between these two dierent mediums. The angles that the
incident and refracted rays made with the line N normal
to the interface between the media are b1 and b2, respectively. Then

equation n1 sin b1 = n2 sin b2 for the reection. Since the

light emission and reection are at the same side, the mediums of both light directions are the same, i.e., n1 = n2. We
get b1 = b2.
Based on the theory of Fermats Principle (Giannoni,
Masiello, & Piccione, 2002), light traverses along the least
time path. The light quickest traversal path from point A,
via plane M, and to point B is the set of line segments
AP and PB with b1 = b2. Since the mediums of both sides
are the same, the speed of light is constant and the light
quickest traversal path is equal to the minimum traversal
path. We then have

n1 sin b1 n2 sin b2 :

Lemma 2. With the incident angle b1 equal to the reflection

angle b2, the length of traversal path along the line segments
AP and PB is the shortest path.

This result, found by Willebrord Snell, is known as Snells

law (Azadeh & Casperson, 1997; Dijksterhuis, 2004). If the
mediums of the both sides are identical, i.e., n1 = n2, the
incident angle b1 is then equal to the refractive angle b2.
Reection is a special case of refraction. Fig. 8 shows an
example of light reection. The direction of dotted line N in
Fig. 8 is also perpendicular to the surface of the plane M.
~ to the plane M,
The light emits from the direction of AP
~
and then makes the reection with the direction of PB.
Let b1 denote the incident angle and b2 denote the reection angle. According to the Snells law, we also have the
N

n1

n2

Proof . In Fig. 9, the point A* is the mirrored point of A

against M. Point D is the intersection of lines AA and
M. Line AA is then perpendicular to M. Then the length
of AD is equal to the length of A D. Since the length of line
segment AP is equal to that of the line segment A P , the
angle b4 is equal to the angle b5. When the incident angle
b1 is equal to the reective angle b2, the angle b3 is equal
to the angle b5. Therefore, the angle b4 is equal to the angle
b3, i.e., the three points A*,P, andB are collinear.
Assume there exists another point p excluding P on M, it
results in the shorter path than P. Since the length of the
line segment Ap is equal to that of the line segment A p, The
length of the path from A via p to B is equal to the length of
the path from A* via p to B. According to the theory of
trigonometric inequality, the length of the path from A* via
p to B is greater than the length of the path from A* via P
to B. It contradicts the assumption. Therefore, the length
of the path from A via P to B is the shortest when the
incident angle b1 equal to the reection angle b2.
h

2
B

Fig. 7. The refraction of light between dierent mediums.

N
B

p
A

D
1

5
4

1 2

2
M

P
Fig. 8. Example of the light reection.

Fig. 9. Geometrical analysis of the light reection.

240

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

A( x A , y A )

yP 

y A y 2P 2xP xA y P y A x2P
x2P y 2P

xP 

xA x2P 2xP y A y P xA y 2P

x2P y 2P

yP  yB
:
xP  xB

If we introduce the polar coordinates

B ( xB , y B )

xP r cos h

P ( xP , y P )
*

A ( x A* , y A * )

10

y P r sin h
and rearrange terms, then equation becomes

r3 sin h  y A r2 sin2 h  2xA r2 cos h sin h y A r2 cos2 h

r3 cos h  xA r2 cos2 h  2y A r2 cos h sin h xA r2 sin2 h
r sin h  y B

:
r cos h  xB

Substituting the expression

sin2 h 1  cos2 h
Fig. 10. The relationship of the light reection and the

E21

problem.

By applying the principle of light reection to the E21

problem, M is the tangent line to circle C at point P.
Fig. 10. shows the relationship of the light reection and
the E21 problem. In order to simplify derivation of solutions, let the center of circle C be coincided with the origin.
The traversal path can be found by any given point P on
circle C. Hence, the minimum traversal path can be derived
from these traversal paths. Let (xA , y A ) be the coordinates
of point A*.
Since P is a tangent point for line M to circle C, the line
segment OP is perpendicular to M. Since the point A is the
mirrored point of A against M, the line AA is perpendicular to M, too. The slope of line AA is equal to the slope of
line segment OP , that is,
y A  y A y P
:
7
xA  xA xP
Both lengths of the line segments AP and A P are equal, we
have
q q
xP  xA 2 y P  y A 2 xP  xA0 2 y P  y A0 2 :
8
By solving the simultaneous Eqs. (7) and (8), the coordinate of A* is gained, that is,
xA x2P 2xP y A y P  xA y 2P
x2P y 2P
y y 2 2xP xA y P  y A x2P
A P
:
x2P y 2P

r3 sin h  2xA r2 cos h sin h 2y A r2 cos2 h  y A r2

r3 cos h  2y A r2 cos h sin h  2xA r2 cos2 h xA r2
r sin h  y B

:
r cos h  xB

11

The variable sin h can be replaced by variable cos h with

p
sin h  1  cos2 h:
Since p
we
can derive the same
result with using
p
sin h 1  cos2 h or sin h  1  cos2 h to substitute
into Eq. (11), we use the expression
p
sin h 1  cos2 h
here for the derivation. We obtain
p
p
r3 1  cos2 h  2xA r2 cos h 1  cos2 h 2y A r2 cos2 h  y A r2
p
r3 cos h  2y A r2 cos h 1  cos2 h  2xA r2 cos2 h xA r2
p
r 1  cos2 h  y B

:
r cos h  xB
By separating variables of cos h and
ing, we obtain

p
1  cos2 h and squar-

4x2A x2B x2B y 2A x2A y 2B y 2A y 2B cos4 h

 4rx2A xB xA x2B xB y 2A xA y 2B cos3 h

xA
y A

into equation, we have

Since points A*, P, and B are collinear, the slopes of line

segments A P and PB are collinear, that is,
y P  y A y P  y B

:
xP  xA xP  xB
By replacing the variables xA and y A with their values from
Eq. (9), we get

r2 x2A 2r2 xA xB r2 x2B  4x2A x2B r2 y 2A  4x2B y 2A

2r2 y A y B r2 y 2B  4r2 x2A y 2B  4y 2A y 2B cos2 h
2r2x2A xB 2xA x2B xB y 2A  xA y A y B  xB y A y B xA y 2B
 cos h x2A y 2B 2xA xB y A y B x2B y 2A  r2 x2B  2r2 xA xB
 r2 x2A 0
To evaluate the polynomial of degree 4 for variable cos h,
we generalize the four solutions as follows:

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

81
p
k  n  12 q  p
>
4
>
>1
p
<
k  n 12 q  p
cos h 41
p
>
k n  12 q p
>
4
>
:1
p
k n 12 q p;
4

12

where k, n, p, q are as follows:

k

raxB bxA
ab
s
1
2

where a, b, c, d, e, f, g, h, j, l, m are
a x2A y 2A
b x2B y 2B
c 2r2 xA xB r2 y 2A x2B r2  4y 2A 2r2 y A y B r2 y 2B
 4y 2A y 2B x2A r2  4b
d 2x2A xB xB y A y A  y B xA 2x2B  y B y A  y B
e x2B r2 y 2A  2xA xB r2  y A y B x2A r2 y 2B
f 36r2 6eaxB bxA 2 cdaxB bxA 6abd 2
g c2 48abe 24r2 daxB bxA
r
q
1 3 3

2c
hp

288abce

2f

4g3 4c3  144abce f 2

3
2
j 12ab
c
r2 axB bxA 2

3ab
2a2 b2
2
m rr axB bxA 3  abcaxB bxA  4a2 b2 d:

l

From Eq. (12), the solutions of angle h can be derived as

follows
8
p
cos1 14 k  n  12 q  p
>
>
>
< cos1 1 k  n 1 p
q  p
4
2
h
p
1
1
1
> cos 4 k n  2 q p
>
>
:
p
cos1 14 k n 12 q p:

13

Since cos h = cos (h), we have

81
p
k  n  12 q  p
>
4
>
>
< 1 k  n 1 p
qp
2
cosh 41
p
1
>
qp
k

n

>
2
>
: 41
1 p
k n 2 q p:
4

15

Let h1, h2, h3, h4, h5, h6, h7, and h8 be the eight solutions of
angle h. With combination of Eqs. (13) and (15), we have
8
p
h1  cos1 14 k  n  12 q  p
>
>
>
p
>
>
>
h2 cos1 14 k  n  12 q  p
>
>
>
p
>
>
h3  cos1 14 k  n 12 q  p
>
>
>
p

< h cos1 1 k  n 1 q  p
4
4
2
16
h
1 1
1 p
>
h
q p
5  cos 4 k n  2
>
>
>
p
>
1 1
1
>
> h6 cos 4 k n  2 q p
>
>
p
>
> h7  cos1 1 k n 1 q p
>
4
2
>
>
p
:
h8 cos1 14 k n 12 q p:

l 1 g
h
2 j h
m
p 3 3
8a b n
1 g
q l  h;
j h

8
p
 cos1 14 k  n  12 q  p
>
>
>
p
<
 cos1 14 k  n 12 q  p
h
p
>
 cos1 14 k n  12 q p
>
>
:
p
 cos1 14 k n 12 q p:

241

14

From Eq. (14), we can derive another four solutions of angle h, that is,

From Eq. (16), the resolved eight angles can derive the corresponding locations of eight points P1, P2, P3, P4, P5, P6,
P7, and P8 on the circle C with their coordinates xP 1 ; y P 1 ,
xP 2 ; y P 2 , xP 3 ; y P 3 , xP 4 ; y P 4 , xP 5 ; y P 5 , xP 6 ; y P 6 , xP 7 ; y P 7 ,
and xP 8 ; y P 8 , respectively. Substituting the resolved eight
angles into the polar coordinates of P in Eq. (10), we obtain their eight coordinates, that is,
8
p
< xP 1 r14 k  n  12 q  p
q
P1 :
17
2
: y P r 1  1 k  n  1 p
q  p
4
2
1
8
p
< xP 2 r14 k  n  12 q  p
q
P2 :
18
2
: y P r 1  1 k  n  1 p
q

p

4
2
2
8
1
1 p
< xP 3 r4 k  n 2 q  p
q
19
P3 :
2
: y P r 1  1 k  n 1 p
q

p

4
2
3
8
p
< xP 4 r14 k  n 12 q  p
q
20
P4 :
: y r 1  1 k  n 1 p
q  p 2
P4
4
2
8
p
< xP 5 r14 k n  12 q p

q
P5 :
21
2
: y P r 1  1 k n  1 p
q p
4
2
5
8
p
< xP 6 r14 k n  12 q p
q

22
P6 :
2
: y P r 1  1 k n  1 p
q

4
2
6
8
p
< xP 7 r14 k n 12 q p
q

23
P7 :
2
: y P r 1  1 k n 1 p
q

4
2
7
8
p
< xP 8 r14 k n 12 q p
q
24
P8 :
2
: y P r 1  1 k n 1 p
q p :
4
2
8
Eqs. (17) and (18) show that there exists reection between
points P1 and P2 about the x axis, that is, x values are kept
and y values are ipped. The pairs of points P3 and P4, P5
and P6, P7 and P8 also have the reection relationships.

242

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

Table 1
The eight dierent combinations of rotation
E21 rotation angle

Quadrant

Fig. 11. Illustration of E21 problem with eight solution points.

Substituting the eight solutions of angle h into Eq. (6)

yields the eight lengths. With nding the shortest length
of these eight lengths, we can get the solution of this problem. From Eq. (4), the minimum traversal length L can be
rewritten as
L minfh1 ; h2 ; h3 ; h4 ; h5 ; h6 ; h7 ; h8 g:
25
Fig. 11 shows the eight solutions for an E21 problem with
xA = 2.5, yA = 2, xB = 1.5, yB = 2.5, r = 1. The optimal
point which results in the shortest length is the solution
point P4 with the coordinate (0.16, 0.99). The minimum
traversal length L, showed by the thicker line segment, is
4.8.
4. Algorithm of the problem rotation
In order to simplify the E21 problem, the problem can be
restricted in the rst quadrant or the rst and the second
quadrants with a two-dimensional rotation of the E21 problem. There are two points A and B in the E21 problem, and
each point can lie in any quadrant of the two-dimensional
plane. The number of dierent combinations of quadrants
of points A and B lay, with repetitions of quadrant, is
42 = 16. Since the points A and B are exchangeable, without repetitions of quadrant, the number of dierent combinations can be reduced to 10. Moreover, there are two
combinations, namely, I and I quadrants, I and II quadrants, which need not make rotation. Then, there are 8 different combinations left needed to be rotated. Consider an
E21 problem rotated to the rst quadrant or the rst and the
second quadrants by an angle. Since a convention about
the direction of rotation must be adopted, assume that
counterclockwise (CCW) rotations are positive and clock-

Point A

Point B

II
III
IV
I
III
IV
III

II
II
IV
IV
III
III
I

IV

II

p/2
p/2
p/2
p/2
p
p
hB, if a < p
2p  hA, if a P p
hB, if a < p
2p  hA, if a P p

wise (CW) are negative. Let hA and hB be the angles,

~ and OB,
~
CCW, from positive x axis to the vectors OA
~
respectively. Let a be the angle from the vector OB to the
~ Table 1 lists the eight dierent combinations
vector OA.
that need to be rotated.
Less formally, we can summarize the descriptions of
Table 1 as the E21 rotation algorithm.
Algorithm. Rotation (E21 )
Input. An E21 consists of two distinct points A and B and
one circle C.
Output. A new E21 with two points A and B lied in I and I
or I and II quadrants.
1. If B lies in II quadrant and A lies in II or III quadrant
2. then Rotate E21 by an angle p/2.
3. If B lies in IV quadrant and A lies in I or IV quadrant
4. then Rotate E21 by an angle p/2.
5. If both of A and B lie in III quadrant
6. then Rotate E21 by an angle p.
7. If B lies in III quadrant and A lies in IV quadrant
8. then Rotate E21 by an angle p.
9. If B lies in I quadrant and A lies in III quadrant
10. then if a < p
11.
then Rotate E21 by an angle hB.
12.
else Rotate E21 by an angle 2p  hA and exchange
A and B.
13. If B lies in II quadrant and A lies in IV quadrant
14. then if a < p
15.
then Rotate E21 by an angle hB.
16.
else Rotate E21 by an angle 2p  hA and exchange
A and B.
17. Stop.
Fig. 12 shows the two possible E21 problems with r = 1
after using the proposed E21 rotation algorithm, and their
eight solutions. Their minimum traversal paths are represented by the thicker paths. The rst possible case, as
shown in Fig. 12a, is that two points A (0.5, 2) and B
(2.5, 2.5) lie in the rst quadrant. Fig. 12b shows the other
possible case whose two points A (1.5, 2) and B (2, 2.5) lie
in the second and rst quadrants, respectively.

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

243

Fig. 12. Two possible E21 problems after using the proposed E21 rotation algorithm and their eight solutions (a) two points lie in the rst quadrant and (b)
two points lie in the rst and the second quadrants.

5. Conclusions
In this paper, the minimum traversal path of three circles is dened as an E03 problem, and is transformed into
the E21 problem. With applying the law of light reection
and geometrical mathematics, the eight roots of solution
function of the minimum traversal path of three circles is
derived. Using the proposed algorithm of the problem
rotation, the problem can be transformed to the problem
in rst quadrant or in the rst and the second quadrants.
This solution can be quick eectively implemented in solving a variety of engineering applications, such as layered
manufacturing, robotic motion planning, and path planning. There are several possible extensions of this work.
For instance, the problem of nding the minimum traversal
path of more than three circles is still essentially open. By
taking advantage of this minimum traversal function of
three circles, we are now exploring the traversal path problem of n circles. Moreover, nding the minimum traversal
path of geometrical object with polygon al shapes is also
interesting to us.
References
Azadeh, M., & Casperson, L. W. (1997). Reection and refraction of light
at saturating active boundaries. Journal of Modern Optics, 44, 2940.
Brutzman, D. P. (1994). A Virtual World for an Autonomous Undersea
Vehicle. Ph.D. Dissertation, Naval Postgraduate School, Monterey,
CA.
Chiu, Y. Y., & Liao, Y. S. (2003). Laser path planning of burn-out rule for
LOM process. Rapid Prototyping Journal, 9(4), 201211.
de Berg, M., van Kreveld, M., Overmars, M., & Schwarzkopf, O. (2004).
Computational geometry algorithms and applications (chap. 4). New
York: Springer.
Dijksterhuis, F. J. (2004). Once Snell breaks down, from geometrical to
physical optics in the seventeenth century. Annals of Science, 61,
165185.

Giannoni, F., Masiello, A., & Piccione, P. (2002). The Fermat principle in
general relativity and applications. Journal of Mathematical Physics,
43, 563596.
Graham, R. L., Lagarias, J. C., Mallows, C. L., Wilks, A. R., & Yan, C.
H. (2005). Apollonian circle packings: Geometry and group theory I.
The Apollonian group. Discrete and Computational Geometry, 34,
547585.
Guibas, L., Hershberger, J., Suri, S., & Zhang, L. (2001). Kinetic
connectivity for unit disks. Discrete and Computational Geometry, 25,
591610.
Jacobs, P. (1992). Rapid prototyping and manufacturing: Fundamentals of
stereo lithography. McGraw-Hill.
Johnson, D. B. (1994). Routing in Ad-Hoc networks of mobile hosts. In
Proceedings of IEEE Workshop on Mobile Computing Systems and
Applications (pp. 158163).
Kaiser, T. (2002). Line transversals to unit disks. Discrete and Computational Geometry, 28, 379387.
Khatib, O., Craig, J. J., & Lozano-Prez, T. (1989). The robotics review.
MIT Press.
Lan, P. T., Chou, S. Y., Chen, L. L., & Gemmill, D. (1997). Determining
fabrication orientations for rapid prototyping with stereo lithography.
Computer-Aided Design, 29, 5362.
Leonhardt, L. B. (1999). Mission Planning and Mission Control Software
for PHONIX Autonomous Underwater Vehicle (AUV) Implementation and Experimental Study. Thesis, Naval Postgraduate School,
Monterey, CA.
Majhi, J., Janardan, R., Smid, M., & Gupta, P. (1999). On some geometric
optimization problems in layered manufacturing. Computational
Geometry: Theory and Applications, 12, 219239.
Murty, K. G. (1995). Operations research deterministic optimization
models. Prentice Hall.
Nehmzow, U., & Owen, C. (2000). Robot navigation in the real world:
Experiments with manchesters forty-two in unmodied, large environments. Robotics and Autonomous Systems, 15, 223242.
ORourke, J. (1997). Computational geometry (chap. 8) (2nd ed.).
Cambridge University Press.
Preparata, F. P., & Shamos, M. I. (1985). Computational geometry: An
introduction. New York: Springer.
Tang, K., & Pang, A. (2003). Optimal connection of loops in laminated
object manufacturing. Computer-Aided Design, 35, 10111022.
Toh, C.-K. (1996). A novel distributed routing protocol to support AdHoc mobile computing. In Proceedings of the 15th IEEE Annual

244

S.-Y. Chou et al. / Expert Systems with Applications 35 (2008) 235244

International Phoenix Conference on Computers and Communications

(pp. 480486).
Wah, P. K., Murty, K. G., Joneja, A., & Leung, C. C. (2002). Tool path
optimization in layered manufacturing. IIE Transactions, 34, 335347.

Wood, L. (2001). Internetworking with satellite constellations. Thesis,

Centre for Communication Systems Research, School of Electronics,
Computing and Mathematics, University of Surrey, Guildford, United
Kingdom.