TELKOM
NIKA
, Vol. 11, No. 5, May 2013, pp. 2545 ~
2552
ISSN: 2302-4
046
2545
Re
cei
v
ed
Jan
uary 9, 2013;
Re
vised Ma
rch 11, 2013; A
c
cepted Ma
rch 22, 2013
Soldiers’ Group Behaviors Simulation Based on
Improved MAPRM
Meng Li*
1
, Jia-hon
g Lian
g
1
, Shi-lei Li
2
1
Colle
ge of Me
chan
ical En
gi
n
eeri
ng an
d Aut
o
matio
n
, Natio
nal U
n
ivers
i
t
y
of Defence tec
hno
log
y
, 4
100
73
Cha
ngsh
a
, P.R. China
2
Departme
n
t of Information Se
curit
y
, Co
ll
ege
of Electroni
c E
ngi
neer
in
g, Na
val Univ
ersit
y
o
f
Engine
eri
ng,
430
00
0, W
uha
n, P.R. China
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: {mengsh
u
q
i
n
198
4, lin
gji
aho
ng_
nu
dt, stoneli}@1
63.com
A
b
st
r
a
ct
Sold
iers
’
gro
u
p
beh
aviors
are
a class
of flo
cking
beh
avi
o
r
s
in w
h
ich
a s
qua
d of so
ldi
e
r
s
marc
h
forw
ard accord
ing to
a sp
ecifi
ed for
m
ati
on
a
nd try to ke
ep i
t
w
hen enc
oun
tering
narrow
c
o
rridors. It has
a
lea
d
in
g a
p
p
lica
t
ion i
n
d
e
terri
n
g
the
be
havi
o
r
and
activiti
es o
f
a pote
n
tia
lly
h
o
stile cr
ow
d a
n
d bri
ngi
ng
a
mob
eng
ag
ed
in
a
riot u
nder
cont
rol i
n
s
o
cia
l
st
abl
e
ma
in
te
na
nce. In th
is p
a
per, w
e
focus
on i
n
vesti
gati
n
g
soldi
e
r
’
s gr
oup
behav
iors b
a
s
ed on a
n
i
m
pr
oved MAPRM
alg
o
rith
m. F
i
rstly, w
e
use MAPRM alg
o
rith
m to
sa
mp
le
co
n
f
i
g
u
r
a
t
i
o
n
s
fro
m
th
e
me
d
i
a
l
a
x
i
s
o
f
the
simu
la
ti
o
n
sp
a
c
e
.
Seco
n
d
l
y
, i
n
the
co
n
s
tru
c
ti
on
and
query
phas
e o
f
MAPRM, clearanc
e infor
m
a
t
ion of the s
p
a
c
e is intro
duc
e
d
to loc
a
l p
l
an
ner a
nd us
ed
a
s
heur
istic infor
m
ati
on to A* a
l
gorit
hm w
h
ic
h
impr
oves
the
MAPRM alg
o
ri
thm. T
h
irdly,
w
hen the sol
d
i
e
rs
pass through narrow corridors
, their formatio
n
transiti
ons ar
e ach
i
eve
d
by
sampli
ng th
e d
e
sire
d for
m
ati
o
n
shap
e. T
he simu
lati
on resu
lts show
that
our appro
a
ch is ef
fective and fe
a
s
ible.
Ke
y
w
ords
:
Cr
ow
d simulati
on
, flocking, sold
i
e
rs
’
gro
up b
e
h
a
vior, MAPRM
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Cro
w
d
s
,
ubiq
u
itous in
the
real
wo
rld f
r
o
m
group
s
of h
u
man
s
to
flocks of i
n
sect
s,
are
vital
feature
s
to m
odel in
a virtual environm
ent. Vari
ou
s
simulatio
n
m
odel
s an
d archite
c
ture
s h
a
ve
been d
e
velo
ped. Crowd
manag
eme
n
t mean
s all meas
ures ta
ken in the
n
o
rmal p
r
o
c
e
s
s of
facilitating th
e movem
ent
and
enjoym
ent of pe
opl
e,
as
well
a
s
all me
asure
s
p
r
ep
are
d
t
o
be
taken i
n
the
emergent p
r
o
c
e
ss
of peopl
e evac
uation
s
. Crowd ma
nagem
ent ca
n be succe
ssful
only whe
n
viewe
d
as
a combinatio
n of
manag
em
en
t of all the crowd
s
, enviro
n
ment an
d their
relationshi
p
. This is
especi
ally necessary for
the cro
w
d incid
ents in
urba
n ci
rcum
stan
ce
s.
Soldiers’ g
r
o
up beh
aviors
are a
cla
s
s of flocki
ng be
h
a
viors i
n
whi
c
h a sq
uad
of soldi
e
rs
march forwa
r
d according t
o
a spe
c
ified
formation. Its appli
c
atio
n area in
clu
d
e
s
the mean
s to
influen
ce the
behavio
r a
n
d
activities
of
a potentia
lly hostile cro
w
d
,
as well as, the
capability
to
bring
a mob
engag
ed in
a riot und
er control. So
ldiers’ group
i.e. a squa
d
of soldie
rs is
a
coh
e
re
nt gro
up
which n
e
ver
splits up i
n
to seve
ral
su
b-g
r
ou
ps. In
current a
pplica
t
ions, the
qua
lity
of coh
e
rent g
r
oup
be
havio
r is i
n
ge
ne
ra
l not very
go
od, espe
cially
few resea
r
ch
ers inve
stigat
es
the soldie
rs’
grou
p b
ehavi
o
rs.
The
prim
ary rea
s
on
s
are th
e n
eed
s to
com
pute
the path
s
in
real-
time and co
ntrol the gro
up’s formatio
n transitio
ns
whe
n
the sol
d
iers’ gro
up
passe
s thro
u
gh
nar
ro
w co
rri
d
o
rs.
In respe
c
t of motion pla
nni
ng problem f
o
r g
r
ou
ps of
entity, the most com
m
on
a
ppro
a
ch
to simulating
group move
ment is to use flocki
ng.
The co
ncept of flocking
was intro
d
u
c
e
d
by
Reynol
ds.
Hi
s boi
ds m
o
d
e
l de
scribe
d
the beh
avio
r
of the entitie
s in a
gro
up
usin
g only lo
cal
rule
s for th
e individual
entities [1]. Later, Reynold
s
extend
ed the tech
nique to in
clude
autonom
ou
s
reactive
beh
a
v
ior. But in
Reynold
s
’s work, little motio
n
pla
nning
was
co
nsi
dere
d
. In
the rob
o
tics community, on
e of the domi
nating te
chni
que
s is the
probabili
stic
roa
d
map a
pproa
ch
(PRM
). Efficient probabili
stic (centralize
d
) te
chniq
u
e
s
for multiple
e
n
tities have
b
een d
e
velop
e
d
.
They treat th
e differe
nt en
tities togethe
r as
one
l
a
rg
e
rob
o
tic
syste
m
. Unfortu
nat
ely, each
enti
t
y
has t
w
o
deg
rees
of free
do
m (a
ssumin
g
it is defin
ed
b
y
its po
sition
on a flo
o
r
su
rface
)
so the
to
tal
roboti
c
syste
m
ha
s 2
n
de
gree
s of fre
e
dom
whe
n
th
ere
are
n
ent
ities. When
n
gets la
rge
r
t
he
runni
ng time
become
s
too
large. To
ove
r
co
me th
is
problem d
e
cent
ralized te
chni
que
s, like p
a
th
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2545 – 255
2
2546
coo
r
din
a
tion,
have b
een
develop
ed, e
nablin
g the
p
l
annin
g
of m
o
tion for a la
rge
r
n
u
mbe
r
of
entities. Still though, the
s
e method
s fa
il when th
e
n
u
mbe
r
of enti
t
ies grows a
nd the re
sulti
n
g
motion is no
t coherent [1]. Recently,
Bayazit, Lien and Amato have com
b
ined the PRM
approa
ch wit
h
flocki
ng techniqu
es. The
entities u
s
e the roa
d
ma
p cre
a
ted by PRM to guid
e
thei
r
motion to
wa
rd the
goal
wh
ile they u
s
e
flocking
to a
c
t
as
a g
r
ou
p a
nd avoi
d lo
ca
l colli
sion
s. B
u
t
grou
ps still
split u
p
e
a
sil
y
. Li and
Chou
dev
elo
p
ed a
ne
w
a
ppro
a
ch that
allo
ws dyna
mic
stru
cturi
ng
of the e
n
tities such
that the
centra
lized
pl
annin
g
of th
e
motion
s i
s
g
r
eatly imp
r
ov
ed.
Ho
wever, thi
s
app
roa
c
h l
a
cks the abil
i
ty of
guaran
teeing co
he
rence [1]. So
it can’t be used
dire
ctly
in sol
d
iers’ gro
up behavio
rs
si
mulation.
A. Kamphui
s, J.F.
Hel
gers
a
nd
M.
H.
Ove
r
mars
introdu
ce
d a new a
pproa
ch to motion planni
ng for
grou
ps of en
tities in virtual environm
e
n
ts.
They model
e
d
the grou
p a
s
a deformab
l
e sha
pe of large e
nou
gh
area a
nd pla
nned the mot
i
on
for this
sha
p
e
throug
h the
environ
ment
usin
g an
exte
nsio
n of the
prob
abili
stic roadma
p
met
hod
[1, 2].
But A. Kamphuis ig
nore
d
the cle
a
ran
c
e info
rmation of the spa
c
e wh
en
generating the
grou
p’s t
r
avel
ing path. Alth
ough Steven
A. Wilmar
th d
e
velope
d MAPRM alg
o
rith
m and u
s
e
d
it to
gene
rate
pat
hs o
n
the
me
dial axis of th
e free
sp
ace, he di
dn’t u
s
e
the cl
earan
ce to imp
r
ove
the
local pl
anne
r
and qu
ery ph
ase of MAPRM [3].
In respe
c
t of formation
control p
r
obl
e
m
of group
s,
currently, the popul
ar m
ean
s of
forming
a target group fo
rmation i
s
to specify the
d
e
s
ire
d
po
sition
of ea
ch a
g
e
n
t at a pa
rticular
moment and
then gene
rat
e
reali
s
tic tra
n
sition
s bet
ween the cu
rre
n
t position an
d the target. So
most existin
g
work a
dopt
s the com
m
on
pro
c
e
ss
of
first sampli
ng t
he de
sired fo
rmation
sh
ap
e
,
followe
d by a path plannin
g
stage for e
a
ch
samp
l
e
points. Sub
s
eque
ntly, agents in the flock
follow
th
e co
rre
sp
ondi
ng sampl
e
point
s, while at
the same time
exhibiting flocki
ng beh
aviors.
Based
on thi
s
ide
o
logy, Anderso
n et al
. con
s
ide
r
ed an
iterative sampling meth
od
to
ge
nera
t
e
grou
p ani
ma
tions
with predefine
d
formation.
More re
cently, Xu et al. propo
sed
a shap
e
con
s
trai
ned fl
ock alg
o
rithm
to handl
e co
mplex 2D
an
d 3D
sh
ape
con
s
trai
ned fl
ocks th
at mo
ve
along p
r
ed
efined path
s
[4, 5, 6].
This p
ape
r focu
se
s on th
e soldi
e
rs’ group be
haviors, whi
c
h i
s
u
s
ed m
a
inly for soci
al
stable m
a
inte
nan
ce an
d m
ob co
ntrol
s
. Our a
pproa
ch
is in
spired
most he
avily by A. Kamphuis’
s
innovational
rese
arch i
n
th
e dom
ain
of moti
on pl
anni
ng for group
s of
entities
and Steven
A.
Wilma
r
th’s m
o
mentou
s wo
rk on
MAPRM
(a
p
r
o
babil
i
stic roa
d
map
plann
er with sampli
ng
o
n
t
he
medial axi
s
of the free space),
Choon S
i
ng Ho’
s
work
is also helpf
ul to fulfill this paper [4, 7,
8
,
9].
Our pa
per co
nsi
s
ts
of thre
e ph
ases:
prepr
o
c
e
s
sing
pha
se with MAPRM,
qu
e
r
y
ph
ase
con
s
id
erin
g the cl
earan
ce
informatio
n of
the sp
ac
e i
n
A* algorithm,
formation tra
n
sition
s ph
ase.
In the first p
h
a
se,
milesto
n
e
s
are
ge
ne
rated
with
MA
P
R
M sampli
n
g
st
rat
egy
wh
ich ca
n
in
cr
e
a
se
the sa
mpling
s
on th
e me
di
al axis of the
corrid
or
s in th
e sp
ace. The
node
s lie
on
the medial
axis
of the free
space which
potentia
lly provides th
e soldiers’ g
r
o
u
p
a maximu
m pa
ssa
ge.
The
medial
axis facilitates acquirin
g process of
the
clearance inform
ation
whi
c
h i
s
used in local
planner
of M
APRM to im
prove the collision-fr
ee links. In the
second phase, the MAPRM
algorith
m
is
improve
d
by con
s
ide
r
in
g
the clea
ran
c
e info
rmatio
n of the sp
a
c
e. The
gro
up’s
traveling p
a
th
is ge
ne
rated
by A* algo
rithm with
an i
m
prove
d
he
u
r
istic fun
c
tion
. As the soldi
e
rs’
grou
p
fo
rmati
on cha
nge
s according
to the
current
cl
eara
n
ce info
rmation, form
ation tra
n
sitio
n
s
are a
c
hieve
d
in the third ph
ase u
s
in
g B-spline curve.
2. Impro
v
e M
A
PRM
Algori
t
hm
2.1. Using
MAPRM
Algorithm to G
e
nera
te
Mileston
es
and
Impro
v
e Loc
al Planner
w
i
th
Cleara
n
ce In
formatio
n
MAPRM (A
Proba
bilisti
c
Roa
d
map
Pl
anne
r
with S
a
mpling
on
the Me
dial A
x
is of the
Free
Spa
c
e
)
use
s
a
sam
p
li
ng
strate
gy of
gen
eratin
g
random
mile
st
one
s lie
on
th
e me
dial
axis
o
f
the free
sp
ace. It efficiently retra
c
ts
any
sam
p
led
co
n
f
iguration, fre
e
or
not, onto
the medi
al a
x
is
of the free sp
ace
without h
a
ving to com
pute the m
edi
al axis explici
t
ly. Sampling and ret
r
a
c
ting
in
this way give excellent p
e
rform
a
n
c
e o
n
probl
ems
of generating
paths
re
quiring traversal
of
narro
w pa
ssa
ges [3].
In our
pap
er
we
reg
a
rd th
e sp
ace a
s
a
2D
co
nfigura
t
ion sp
ace in
whi
c
h the
fre
e
sp
ace
and
ob
stacl
e
s a
r
e
polygo
n
a
l. For si
mpli
city, we
co
nsi
der
only
sets
P
that a
r
e th
e
disjoi
nt union
of a finite
nu
mber of
clo
s
e
d
polyg
on
s
(i
nclu
ding
the i
n
terio
r
, po
ssi
b
ly with
hole
s
). Fo
r
x
P
, we
define
()
P
Bx
to b
e
the la
rge
s
t closed
disc cente
r
ed
at
x
that is
a su
bset of
P
,
i.
e.,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Soldiers
’
Group Beha
vio
r
s
Sim
u
lation Base
d on Im
prove
d
MAPRM (Meng Li
)
2547
()
,
(
)
PP
Bx
B
x
x
, where
,
Bx
r
den
otes the
clo
s
ed di
sc of
ra
dius
0
r
cente
r
e
d
at
x
and
2
()
t
a
n
,
\
P
x
di
s
c
e
x
P
R
is the
dista
n
c
e to
the
bo
u
ndary
for
poi
nts in
sid
e
P
a
nd 0
for poi
nts
outsid
e
P
[3]. T
he medial axi
s
()
M
AP
of
P
is defined as
(
)
|
w
ith
(
)
(
)
PP
M
AP
x
P
y
P
B
x
B
y
Ø
(1)
Formul
ation
(1) m
ean
s th
a
t
()
M
AP
are the
set of all
point
s
x
of
P
w
h
os
e as
so
c
i
a
t
e
d
()
P
Bx
is maximal.
A
point
x
P
is cal
l
ed a
sim
p
le point
if
x
has a
unique nea
rest point
0
x
in
P
.
Otherwise,
x
is called a
m
u
ltiple point
.
P
is defined to be
P
minus its no
n-convex vert
ex points.
Figure 1. Medial axis of a clo
s
ed p
o
lygo
n
The followi
ng
facts hold a
c
cording to referen
c
e [3]:
1. Let
x
P
, Then
x
is in
()
M
AP
if and only if
x
is a con
v
ex vertex of
P
.
2. Any multiple point of
P
is contai
ned in
()
M
AP
. If
()
x
MA
P
, then
x
is in the interior of
P
if and only if
x
is a
m
u
ltiple
point
of
P
[5].
3. For e
a
ch
x
P
,
()
P
Bx
is conta
i
ned in
a u
n
ique m
a
xim
a
l disc
()
P
By
, where
()
yM
A
P
. Furthe
rmo
r
e if
\\
(
)
x
PP
M
A
P
, then
y
is on the
ray
0
x
x
. If
x
is a
non
-vertex
point of the bound
ary, then
y
is on the line throug
h
x
no
rmal to the bo
unda
ry at
x
.
4. The ma
p
\(
)
PM
A
P
P
taking
ea
ch p
o
int to
its nea
re
st
boun
dary p
o
i
nt is
contin
uou
s a
nd
()
M
AP
is a
clo
s
ed
set. Esp
e
cially imp
o
rtant is the f
a
ct that
()
M
AP
is
an
deform
a
tion retract of
P
because
P
can be
contin
uou
sly
deform
ed o
n
t
o
()
M
AP
without m
o
ving
any of the points of
()
M
AP
.
Let
C
den
otes the co
nfigu
r
ation spa
c
e i
n
2
R
and
BC
the
Co
b
s
t
a
c
l
e
be a di
sjoi
nt
union of a fin
i
te numbe
r of
polygon
s, then
\
FC
B
is the fre
e
sp
ace whi
c
h co
nsi
s
ts of
a finite
numbe
r of co
nne
cted poly
gono
us comp
onent
s.
Acco
rding to afore
m
entione
d facts,
Proposition
1
is obtain
ed
[3].
Proposition
1.
The canoni
cal retraction ma
p
()
FM
A
F
can be extended
contin
uou
sly to map
\(
)
(
)
CM
A
B
M
A
F
.
On
the ba
sis of
Propositio
n 1
, we
can i
n
crea
se th
e
sampling
s
i
n
medial
axis o
f
all the
corrid
ors whi
c
h h
a
ve vario
u
s
clea
ran
c
e
s
in the
spa
c
e. In our p
r
ep
rocessin
g ph
ase, the MAP
R
M
algorith
m
for sampli
ng the
mileston
es o
n
the medial
axis is given i
n
Algorithm 1
[3].
Algorithm 1
MAPRM sa
m
p
ling strategy
Input:
N
, the number of mile
stones to ge
ne
rate
Output:
N
milestone
s on the
medial axis o
f
F
1:
repea
t
2:
Gene
rate a u
n
iformly ran
d
o
m point
p
in
C
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2545 – 255
2
2548
3:
Find the nea
rest point
q
on
F
to
p
.
4:
if
p
is
free
the
n
5:
Take th
e retraction di
re
ction
v
to be
qp
, and let the start point
s
be
p
.
6:
else
7:
Take th
e retraction di
re
ction
v
to be
pq
, and let the start point
s
be
q
.
8:
end if
9:
Usi
ng bi
se
ction, move
s
in the dire
ction
v
until
q
is not the uniqu
e
nearest poi
nt of
F
to
s
. This
moves
s
onto the medial axis
of
F
.
10:
until
N
milesto
nes h
a
ve bee
n gene
rated.
After the milestone
s which are u
s
ed to
construct a roa
d
map a
r
e ge
nerate
d
by
Algorithm 1
, we can u
s
e the cle
a
ra
nce informatio
n co
mputed by
medial axis to
improve the
local pl
anne
r
of MAPRM. The cle
a
ra
nce of one point
()
x
MA
F
is
0
()
t
a
n
,
clearanc
e
x
dis
c
e
x
x
whi
c
h is illust
rated in Figure 2. A corri
dor is defined as:
()
a
n
d
,
|
,
(
)
,
(
)
,
(
)
0
,
w
z
cor
x
M
A
F
x
z
w
z
w
M
A
F
c
l
e
arnc
e
z
cl
earn
c
e
w
(2)
in whi
c
h
is a pred
efine
d
value of
width. The
cleara
n
ce info
rmation of
w
z
co
r
is
()
m
i
n
(
)
|
ww
zz
c
l
e
a
r
anc
e
c
or
c
l
e
a
r
a
nc
e
x
x
c
or
which is the
minimum
s
value of the cle
a
ran
c
e
s
of all
the points tha
t
are in the co
rrid
o
r (Fi
g
u
r
e
2).
Medial axis
x
F
B
B
0
x
z
w
(
)
w
z
cl
ea
ran
c
e
c
or
Group
sold
ie
r
Figure 2. The
clearan
ce inf
o
rmatio
n of a corrid
or
After the mile
stone
s o
n
me
dial axis a
r
e
gene
rated
an
d the cl
earan
ce info
rmatio
n of the
corrid
ors is
acq
u
ire
d
, we
can improve the ex
isting local pla
n
n
e
r of MAPRM by adding
the
following rule
to it.
If
()
w
z
c
l
e
a
r
anc
e
c
or
Then
th
e
mileston
es i
n
w
z
co
r
are
delete
d
. At the sam
e
time, we
sup
p
leme
nt the numb
e
r of
milestone
s that are del
ete
d
usin
g
Algor
ithm 1
.
In whic
h,
is a
n
u
s
e
r
d
e
fined valu
e
and
aforem
entione
d rule
mean
s th
at if the
clea
ran
c
e of a
co
rri
do
r
is smaller
th
an
, then the
co
rri
d
o
r i
s
forbidde
n to pa
ss. At the same tim
e
,
we
sho
u
ld
su
ppleme
n
t the
numb
e
r of m
ileston
es to
N
. Up to p
r
e
s
e
n
t, the prep
ro
cessing
pha
se
is accom
p
lished.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Soldiers
’
Group Beha
vio
r
s
Sim
u
lation Base
d on Im
prove
d
MAPRM (Meng Li
)
2549
2.2. Impro
v
e
Quer
y
Phase of MAPRM
In this pha
se, A* algorit
hm is u
s
e
d
to find a pat
h betwe
en t
he sta
r
t and
the goal
config
uratio
n, usin
g the
ro
admap
co
nst
r
ucte
d
in th
e
prep
ro
ce
ssin
g pha
se. Sim
u
ltaneo
usly,
we
improve
the
MAPRM al
go
rithm o
n
ce a
g
a
in by
ren
o
va
ting the
heu
ri
stic fu
nctio
n
o
f
A* algo
rithm
in
existing MAPRM.
We use
()
hz
to d
enote th
e h
e
u
r
istic fun
c
tion
in existin
g
A*
algorith
m
a
n
d
()
hz
to
den
ote
the improve
d
heuri
s
tic fun
c
tion.
()
()
(
)
m
a
x
(
)
,
(
)
h
z
h
z
clea
ra
n
c
e
x
clea
ra
n
c
e
z
cl
ea
ra
n
c
e
w
(3)
whe
r
e
()
m
a
x
(
)
,
(
)
c
l
e
a
r
anc
e
x
c
l
e
a
r
anc
e
z
c
l
e
a
r
anc
e
w
mean
s that we
should ta
ke i
n
to
accou
n
t the
variation
of t
he
clea
ran
c
e
inform
ati
on
of the
co
rrid
o
r
in
the
estim
a
ted
co
st of t
h
e
path from mil
e
ston
e
z
to the goal
config
uration. In thi
s
wa
y, the bigger i
s
the v
a
riation
of the
clea
ran
c
e
of
a co
rri
do
r, the larger th
e
cost of p
a
ss
in
g this
co
rrid
o
r to the go
al. As the va
riati
on of
the cle
a
ran
c
e
refle
c
ts the
changi
ng d
egree of
g
r
ou
p’s
formation
wh
en pa
ssing t
h
e co
rri
do
r,
()
hz
corresponds
to the
probability to sel
e
ct t
he path through the
co
rridor. Obviously,
()
hz
never
overe
s
timate
s the
cost to
rea
c
h the
g
oal. So
far, usin
g improved A* algo
rit
h
m, the gro
u
p’s
traveling path
can be g
ene
rated.
3. Formation
Transition o
f
the Soldier
s
’ Group
In this se
ctio
n, we p
r
e
s
ent
a flexible form
ation ha
ndli
ng by the pro
c
e
ss
of first sampling
the de
sire
d formatio
n sh
a
pe, followe
d
by a pat
h pla
nning
stage f
o
r ea
ch
sam
p
le point
s on
the
basi
s
of jiayixu’s
work [6]
.
Subse
que
ntly, so
ldie
rs i
n
the g
r
o
up
follow th
e pa
th between t
h
e
corre
s
p
ondin
g
sampl
e
poi
nts, while at the sam
e
time
avoiding collision
s
with e
a
c
h othe
r.
In ord
e
r to
d
r
ive the m
o
tion of an
enti
r
e g
r
ou
p in
a more
cohe
rent ma
nne
r
whe
n
it
migrate
s
fro
m
initial formation to desi
r
ed one, we spe
c
ify a glo
bal path
()
Rt
(B-s
pline c
u
rv
e
)
betwe
en the
cente
r
s of the
two formatio
ns [6].
Let
1
O
and
2
O
re
spectively re
prese
n
t the ce
nters of
the i
n
itial formatio
n and th
e de
sire
d
formation. We set
1
O
as the start point,
2
O
as the end poi
nt of the B-sp
line cu
rve. Similarly, we
assume
1
S
and
2
S
as the two
corre
s
p
ondin
g
sam
p
le p
o
ints of one
so
ldier in th
e g
r
oup
on the
initial formati
on and the d
e
s
ire
d
formatio
n, as illustrated in Figu
re 3
.
1
O
2
O
1
S
2
S
()
Ot
()
St
()
St
()
R
t
Figure 3. The
formation tra
n
sition of the
grou
p
Then the cen
t
er of the gro
up
()
Ot
betwe
en
the initial and
the desired formatio
n ca
n
be
comp
uted by formulatio
n (4
).
12
()
(
1
)
Ot
t
O
t
O
(4)
Similarly,
12
()
(
1
)
St
t
S
t
S
. Then the path b
e
twee
n
1
S
and
2
S
can b
e
co
nstructed a
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2545 – 255
2
2550
12
1
2
12
12
(
)
()
()
()
SS
S
S
St
R
t
O
t
S
t
OO
O
O
(5)
One maj
o
r a
d
vantage of t
h
is p
a
th co
ntrol schem
e is
that we only
need to p
a
ra
meteri
ze
the path on
ce for all sa
m
p
le point
s [6]. At
the sam
e
time, Homi
ng beh
avior
whi
c
h is u
s
e
d
to
drive the
sol
d
ier and
colli
sion
avoidance behavior whi
c
h
i
s
used
to avoid collisions
bet
ween
soldi
e
rs are utilized to
steer the soldier.
In this
way, the form
ation
trans
ition phase of the
group
is accom
p
lished.
4. Experimental Re
sults
and Con
c
lus
i
ons
In our
sim
u
l
a
tion, we
e
m
ploy a h
u
m
an a
n
imati
on softwa
r
e
packa
ge
call
ed DI
-Guy,
whi
c
h is com
m
ercially available from Boston Dy
n
a
m
i
cs In
c. we control the be
haviors of the
she
phe
rd an
d the flock u
s
ing SDK by
C++ pro
g
ra
ms. We sup
pose the followin
g
scena
rio to
prove
ou
r m
odel: a
squa
d of
soldi
e
rs form
s a
co
here
n
t g
r
ou
p
,
this g
r
o
up
tactical
targe
t
of
marchin
g
fro
m
the initial p
o
sition to th
e
goal p
o
si
tion
to get on the
army
tru
c
ks. Figure
4(a) gi
ves
the initial
co
nfiguratio
n of
enviro
n
ment
. Fig. 4(
b
)
, (c),
(d) give t
he snap
sh
ots of
simulati
on
res
u
lts.
(a)
initial simulati
on environment
(b)
sna
p
shot at t=20
s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Soldiers
’
Group Beha
vio
r
s
Sim
u
lation Base
d on Im
prove
d
MAPRM (Meng Li
)
2551
(c) sn
ap
shot
at t=29s
(d) snap
sh
ot at t=35s
Figure 4
.
Simulation re
sult
s of the soldi
e
rs’ g
r
o
u
p
From Fig
u
re
4, we can
se
e that improved
MAPRM a
l
gorithm ge
ne
rate a more realisti
c
path for soldi
e
rs’
gro
up by
con
s
ide
r
ing
the clea
ra
nce
information
of the spa
c
e.
And in narro
w
corrid
ors, the grou
p ch
ang
e
s
its formatio
n automatical
ly.
5. Summar
y
In this pa
pe
r
we h
a
ve p
r
e
s
ented the
sol
d
iers’ g
r
ou
p b
ehaviors by i
n
trodu
ce
an i
m
prove
d
MAPRM alg
o
rithm. In o
u
r ap
proa
ch,
we in
co
rpo
r
ated MAP
R
M algo
rithm
with cl
eara
n
ce
informatio
n o
f
the spa
c
e
by improve t
he lo
ca
l
pla
n
ner and
A*
algorith
m
in
query
pha
se
of
MAPRM. Simultaneo
usly, the soldie
rs’
g
r
oup
ca
n a
d
a
p
t its formatio
n acco
rdin
g to the cl
earan
ce
informatio
n. But, this pap
er do
esn’t take the inte
ra
ction bet
wee
n
the mob a
nd the soldie
r into
consideration. Further research
es will focus on this problem.
Ackn
o
w
l
e
dg
ments
This pa
pe
r was supp
orted
by national
n
a
tural scie
nce foundatio
n (6117
0160
).
Referen
ces
[1]
A Kamph
u
is,
MH Overmars. F
i
ndin
g
path for c
oher
ent gro
u
p
s usin
g cl
e
a
ranc
e.
ACM
SIGGRAPH/Eurographics Sym
p
osium
on Computer Anim
ation
. 200
4: 193
-202.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 5, May 2013 : 2545 – 255
2
2552
[2]
A Kamph
u
is,
JF Helg
ers, MH Overm
a
rs
. Motion
pla
nni
ng for
grou
ps of
entities.
ACM
SIGGRAPH/Eurographics Sym
p
osium
on Computer Anim
ation
. 200
3: 193
-202.
[3]
C Hol
l
ema
n
a
nd L Kavr
aki. A frame
w
ork for us
in
g the
w
o
rkspac
e med
i
al a
x
is i
n
prm
plan
ners.
In
Internatio
na
l C
onfere
n
ce o
n
Rob
o
tics an
d Automatio
n
. 200
0; 2: 1408-
141
3.
[4]
SH Cho
on, H
N
Quang, O Ye
w
-
S
o
o
n
. Auton
o
m
ous m
u
lti-a
gents i
n
fle
x
i
b
l
e
flock formati
on.
MIG2010
,
LNCS 6
459
. 2
010: 37
5–
38
5.
[5]
F
Aurenh
amm
e
r. Voron
o
i d
i
agrams: A su
rve
y
of a fun
dame
n
tal g
e
o
m
etric data st
ructure.
ACM
Com
p
ut. Surv
.
199
1; 23: 345
–
405.
[6]
Xu
J, Jin
X, Y
u
Y, She
n
T
,
Zhou M.
S
hap
e-
constrai
ned
flo
ck anim
a
tion. I
n
: Comp
uter
A
n
i
m
ati
on a
n
d
Virtual W
o
rl
ds
. 200
8; 19: 313
–
330.
[7]
CW Re
y
n
o
l
ds.
Flocks, herds
, and scho
o
ls:
A distribute
d
beh
avior
a
l mo
del.
in Co
mp
ut
er
Graphics
.
198
7: 25–
34.
[8]
RT
Vaughan,
N Sumpter, J Hen
derso
n, A F
r
os
t, and S Camero
n. Exp
e
riments i
n
au
tomatic flock
control. Ro
bot and Auto
nom.
S
y
s. 20
00; 31:
09-1
17.
[9]
OB Bay
a
zit, JM Lien, NM Amato.
Better Group B
ehav
iors
using
Rul
e
-Ba
s
ed Ro
ad
maps
. In Proc. Int.
W
kshp. on Alg.
F
ound. of Rob
.
(W
AF
R), Nice, F
r
ance. 2002:
95-11
1.
Evaluation Warning : The document was created with Spire.PDF for Python.