TELKOM
NIKA
, Vol. 11, No. 4, April 2013, pp. 2264
~
2
270
ISSN: 2302-4
046
2264
Re
cei
v
ed
De
cem
ber 2
4
, 2012; Re
vi
sed
March 3, 201
3; Acce
pted
March 10, 20
13
Velocity Perception: Collision Handling Technique for
Agent Avoidance Behavior
Na
zree
n Abd
u
llasim
1
, Ahmad Hairul Basori*
2
, Md Sah Hj Salam Abdullah Bade
3
1
Kulli
yya
h
of Information a
nd
Commun
i
cati
o
n
T
e
chnolo
g
y
,
Internatio
na
l Islamic Univ
ersit
y
,
531
00 Gomb
a
k
, Selang
or, Mala
ysi
a
2
UT
M VicubeL
ab, F
a
cult
y
of Comp
uting, Un
iversiti
T
e
knolo
g
i Mal
a
ysia Sk
uda
i, 813
10 Jo
hor, Mala
ys
ia
3
Sekola
h Sai
n
s
dan T
e
knolo
g
i
,
Universiti Mal
a
y
s
ia
Sa
ba
h, 8889
9, Kota Kin
aba
lu, Sab
ah, Mala
ysi
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: ajen
dgre
a
t@
gmail,c
o
m
1
, uchih
a
.hoir
u
l@
g
m
ail.com*
2,
aba
de
08@
ya
h
oo.com
3
A
b
st
r
a
ct
Collis
io
n avo
i
d
ance
beh
avi
o
r is alw
a
ys ab
ou
t mai
n
ta
i
n
in
g free col
lisi
on
bet
w
een virtua
l ob
jects. It
is als
o
ab
out
g
ener
ating
ev
asi
on r
out
in
g for
the
ag
ents i
n
v
i
rtual
envir
on
m
ent suc
h
as i
n
crow
d si
mu
lati
on.
It consists of three
process
e
s w
h
ich are c
onstructio
n
of
Field of Vis
i
o
n
,
Collis
ion
ha
n
d
lin
g a
nd co
lli
sion
respo
n
se. C
o
n
s
tructing fie
l
d
of visio
n
is
al
w
a
ys a da
unti
ng task
and
al
w
a
ys in e
n
ig
ma for the
desi
g
ne
r
beca
u
se it is subj
ected tow
a
r
d
s age
nt
’
s
perc
eptio
n w
h
ic
h is
varies to eac
h of them. T
here
are few
attemp
ts
on d
e
si
gni
ng fi
eld
of visi
on
b
a
sed
on th
e a
gent
’
s
dyn
a
m
i
c
focus tow
a
rd
its surrou
n
d
i
n
g
. T
herefor
e, w
e
prese
n
t a top
dow
n ap
pro
a
c
h
study fro
m
c
r
ow
d simula
ti
o
n
mod
e
li
ng u
n
t
il the col
lisi
o
n
han
dli
ng l
e
vel
in
order t
o
i
d
e
n
tify the s
u
itab
le
crow
d mod
e
li
n
g
for
our
ap
pro
a
ch. H
ence,
at
the
end
of th
is
pa
per w
e
w
i
l
l
be
abl
e to discuss
the possib
l
e techn
i
qu
es for constructi
n
g
a
gent
’
s
fie
l
d of vision
and a
n
a
l
y
z
e its pot
entia
l i
n
crow
d simu
lati
on env
iron
men
t.
Ke
y
w
ords
: av
oid
ance
beh
avi
o
r, collisi
on d
e
tection, crow
d simulati
on, aut
o
n
o
m
o
u
s ag
ent
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion and Ba
ck
ground
Cro
w
d
Simul
a
tion i
s
a
complicated
syste
m. Th
ro
ugho
ut years, the
r
e
are
many
developm
ent approa
che
s
i
n
trodu
ce
d that is called
crowd mo
delin
g. They are segre
gated by
the
different app
roa
c
h taken
on definin
g the virtual environme
n
t with the cro
w
d ag
e
n
ts.
Furthe
rmo
r
e, cro
w
d
simulat
i
on com
p
ri
se
s
of
many
ele
m
ents an
d o
n
e
of th
em i
s
crowd
behavio
r.
Cro
w
d
be
hav
ior i
s
a
syste
m
that p
r
ovid
es
actio
n
s to
the auto
nom
ous ag
ents
so that it
ca
n
react
to its virtual
environ
ment.
It adds
a se
nse
of re
a
lism to the cro
w
d livelih
ood
as if e
a
ch
o
f
the
agent h
a
s
a
sense of intelli
gent that
a
b
le
them to ma
ke de
cisi
on m
a
kin
g
up
on th
eir rea
c
tion th
eir
surro
undi
ngs.
One of th
e m
o
st ba
si
c cro
w
d b
ehavio
rs is Collisi
on a
v
oidan
ce. Ge
nerally
Colli
si
on
avoidan
ce i
s
a me
ch
ani
sm of which t
he a
gent
w
ill
avoid a
n
y a
gent fro
m
col
liding a
nd
avoid
in
te
r
s
e
c
tion
w
i
th
s
t
a
t
ic ob
s
t
ac
les
.
T
h
e
r
e is a
stu
d
y that di
sti
n
ct
colli
sion
avoidan
ce
a
s
a
n
avoidan
ce m
e
ch
ani
sm be
tween
dyna
mic obj
ect
s
and o
b
sta
c
le
avoidan
ce
as a
avoida
nce
mechani
sm between dynamic with
stati
c
obstacles[1].
However in this paper
we will mai
n
tai
n
colli
sion avoi
dan
ce a
s
a g
eneral avoida
nce b
ehavio
r for both stati
c
and dynami
c
obsta
cle
s
.
1.1 Cellular Automata
In orde
r to u
nderstan
d th
e archite
c
t of cro
w
d
beh
avior, it is e
s
se
ntial to unde
rstand it
s
cro
w
d
mo
del.
The
r
e
are
m
any ap
pro
a
ch
es
of cro
w
d
modelin
g a
n
d
the m
o
st
co
mmon
are
So
cial
Force, Cell
ul
ar Automata
(CA) an
d Rule
Based.
CA i
s
a commo
n model for
cro
w
d si
mulation
in
the early co
mputer g
a
me
developme
n
t espe
cially
fo
r a strategy and turn ba
sed gam
es. CA is
based o
n
spa
t
ial spa
c
e
def
ragm
entation
of the virt
ual
environ
ment.
It is se
gment
s into
whe
r
e i
t
is call
ed as
cells [2]. In order to avoid ag
ents in
terse
c
t
i
ng with ea
ch
other, CA can
define the ce
ll
to allow on
e
agent to
occupy only o
n
e
cell
at a tim
e
. Thu
s
by a
dapting thi
s
approa
ch,
CA will
guarantee fre
e
colli
sion/int
e
rsectio
n
bet
wee
n
age
nt
a
nd any stati
c
obje
c
t that de
fine in occu
pi
e
d
cell
s. Althoug
h CA i
s
sim
p
l
e
co
mpa
r
e fo
r many ot
h
e
r
cro
w
d
mod
e
li
ng, ho
wever i
t
is not
suitab
le
to simulate d
ense situatio
n sin
c
e it can
not
visuali
z
e
push or bo
dy conta
c
t between ag
ents.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Velocity Perception: Colli
si
on Handling
Te
chnique for Agent... (N
azreen Abdullasim
)
2265
1.2 Social Force
Social fo
rce
wa
s introdu
ced by
Helbi
ng
that in
co
rporate
Newt
on’s law into
crowd
simulatio
n
[3]. It functions
as colli
sio
n
avoidan
ce and
ensu
r
e
s
free
interse
c
tion
betwe
en age
nts.
More
over,
so
cial fo
rce a
b
l
e
to
simulate
pu
sh
effe
ct and body co
ntact
b
e
twe
e
n
pa
rticle
s where
CA co
uld not
perfo
rm. It is a strai
ght forward alg
o
rith
m to impleme
n
t in cro
w
d
si
mulation, thu
s
it
can
sim
u
late
a large
amou
nt of cro
w
d
s
.
Ho
wever, it
i
s
difficult to i
n
tegrate
soci
al
force
with
oth
e
r
behavio
r such as
path foll
owin
g, flee a
nd many
oth
e
r. Fu
rtherm
o
re,
soci
al fo
rce f
ond to
ward
sha
k
in
g movement wh
en the age
nts are
in hi
gh den
si
ty situation or at narro
w sp
ace
s
.
1.3 Rule Bas
e
d
In 1987, Crai
g Reynold
s
h
ad introd
uce
d
Boid
s that simulate flo
c
king be
havior of birds
[4]. The simul
a
tion is
reali
s
tic and a
b
le t
o
imitat
e flocking b
ehavio
r to an extent whe
r
e e
a
ch b
i
rd
has thei
r o
w
n
indepe
nde
nt behavio
r. Each individual
can pe
rform th
ree di
stin
ctive action
s which
are
co
hesi
on,
sep
a
ration a
nd alig
nment.
This
rule
ba
sed mo
del i
s
e
x
tensible
and
able to
creat
e
compl
e
x be
h
a
vior. Th
ro
ug
hout the
yea
r
s m
any
re
se
a
r
ch
ers i
n
tro
d
u
c
ed
thei
r exp
ansi
on
of cro
w
d
simulatio
n
su
ch a
s
Vi
Crowd [5] and
Cle
a
rPath [6
] wh
ich
a
r
e ba
sed
on Crai
g Re
ynolds’
s
wo
rk.
In
1999, Reyno
l
ds ha
d extended hi
s wo
rk by intr
od
ucin
g Steeri
ng Behavio
r into his crowd
s
i
mulation [7]. This
work is
im
po
rtant
as
far as colli
sion avoi
dan
ce
i
s
co
nce
r
n. Colli
si
on
avoidan
ce
was int
r
od
uce
d
in
steeri
n
g
behavio
r th
at dedi
cate
s only to pe
rform avoid
a
n
c
e
maneuve
r
action for autonomou
s ag
e
n
t. Howeve
r,
developing
cro
w
d
simula
tion using ru
le
based app
ro
ach i
s
com
p
li
cated
comp
a
r
e to CA and
so
ci
al force. This is du
e to the fact that
desi
gning
be
havior for thi
s
cro
w
d mo
d
e
l need to
b
e
co
nci
s
e a
n
d
yet comp
re
hen
sive as
o
n
e
behavio
r may
relate to anot
her.
2. Cro
w
d
Simulation De
sign Crite
r
ion
Accordi
ng to crowd
sim
u
lation
design crit
eri
on (whi
ch are
fl
exibility,
extensibility,
execution efficiency
and scalability [8]), rule based i
s
mo
re fulfilli
ng in
comparison
with other
cro
w
d
mod
e
l
. Neverth
e
le
ss,
executio
n efficien
cy
and
scala
b
ility are de
p
endin
g
on t
h
e
compl
e
xity of the behavio
r algorithm. A
l
though
rule
based is m
o
re com
p
licate
d
in term of i
t
s
developm
ent, it offer more
flexibility and extensibilit
y compa
r
e to
CA and soci
al force. Thu
s
, rule
based mod
e
l is more suita
b
le wh
en de
si
gning
sop
h
isti
cated b
ehavi
o
r.
Table 1. The
Differen
c
e
s
B
e
twee
n Crowd Model
s Pertaining To Th
e Colli
sion Handlin
g
Cro
w
d M
odel
Flex
ibility
Ex
tensibility
Scalability
Ex
ecution
CA
Rigid to Space
partitioning
Based on space
expansion
Large
Fast
Social Force
Rigid to
Ne
w
t
on’s La
w
No Large
Fast
Rule Based
Continuu
m
and
Based on Agent
decision
Yes
M
ediu
m
M
ediu
m
De
signi
ng re
alistic
cro
w
d
behavio
r ca
n be a com
p
licated task. As the cro
w
d si
mulatio
n
getting mo
re
reali
s
tic, cro
w
d b
ehavio
r
obviou
s
ly will
be mo
re
co
mplex and
h
ence affectin
g the
comp
utationa
l cost for co
mputer p
r
o
c
essor
a
nd m
e
mory. Ho
wever, the sol
u
tion doe
s n
o
t
necessa
ry so
lve with mo
re
memo
ry and
faster
pro
c
e
s
sor. It al
so
ca
n be d
one
by simplifying th
e
algorith
m
. Effectivene
ss a
nd efficien
cy are the
fun
d
a
m
ental de
sig
n
obje
c
tives to be co
nsi
dered
in a real-time simulation. To maintai
n
the inte
ractive
rate pe
rform
ance, the tra
de-off bet
we
en
pre
c
isi
on an
d
spee
d execu
t
ion must be
balan
ce a
c
co
rding to the a
pplication [9].
3. Adap
ting Steering
Be
hav
i
or
Steering Beh
a
vior whi
c
h
wa
s introd
uced by
Reynol
ds is o
ne of the best exa
m
ples o
f
distrib
u
tive crowd be
havior [1]. Distri
bution of
cro
w
d be
havior basi
c
ally co
mpri
se
s of b
a
si
c
steeri
ng b
e
h
a
vior an
d co
mbination th
ose. A
s
fo
r
example, flocking
beh
avior is comp
rise
s of
three b
a
si
c b
ehaviors
whi
c
h a
r
e
coh
e
si
on, se
parat
io
n and
alignm
ent. Dist
ributi
on facto
r
in
rule
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 4, April 2013 : 2264 – 2
270
2266
based
will
al
low it to
be
co
mbine
d
a
nd thu
s
able
to cre
a
te
sophi
sticate
d
and
eme
r
ge
nce
behavio
r. Tab
l
e 2 is the example
s
of
Basic a
nd Comb
ine stee
ring b
ehavior.
Table 2. Example of Basi
c And Combi
n
ed Steerin
g Behavior [1], [7]. [10]
Basic Steering Behavior
Combined Steeri
ng
Behavior
-Seek and Flee
-Pursue and
evade
-Wander
-Arri
v
a
l
-Obstacle avoidance
-
C
ollision avoida
nce
-Containment
-
W
all follow
i
ng
-Path follo
w
i
ng
-Flo
w
field follow
i
ng
-Cohesion
-Separation
-Alignment
-Cro
wd path follo
w
i
ng
-Leade
r follow
i
ng
-Unaligned collisi
on
avoidance
-Que
uing
-
F
locking
-Seek and Follo
w
i
ng
In this paper
we
will focus on collision
avoi
dance behavior as it i
s
rel
a
ting toward our
resea
r
ch focu
s. Colli
sio
n
a
v
oidan
ce g
e
n
e
rally
pe
rform avoidan
ce
maneuve
r
for the age
n
ts from
intersecting
or colliding with
ob
stacles i
n
the vi
rtual environm
en
t. It is
a fundamental behavior
in
cro
w
d
sim
u
la
tion be
cau
s
e
it’s bee
n
with many oth
e
r com
b
ination
of ba
sic
beh
a
vior in
ord
e
r to
cre
a
t
e
mo
re
compl
e
x
beh
av
ior.
Ho
wev
e
r,
colli
sion
a
v
oidan
c
e itsel
f
is a combi
n
ation of two b
a
si
c
behavio
rs which
are avo
i
dan
ce
wi
th
static
ob
stacl
e
and
avoid
a
nce with
dy
namic ob
sta
c
le.
Acco
rdi
ng to some
re
sea
r
cher, avoida
nce with st
atic
obsta
c
le i
s
known as Ob
stacle Avoidan
ce
and avoid
a
n
c
e with dyn
a
m
ic ob
stacl
e
is kno
w
n a
s
colli
sio
n
a
v
oidan
ce [1]. Figure 1 i
s
a
n
example of
combi
ned
st
eerin
g be
hav
ior in
ope
nS
teer
C++ lib
rary fo
r see
k
an
d flee t
hat
inco
rpo
r
ate
collision
avoid
ance and o
b
stacle avoid
a
n
c
e [11].
Figure 1: Example of the Structu
r
e Of Combine
d
Steering Be
havio
r
3.1 Collision Av
oidance behav
i
or description
In order to m
a
intain the belie
vability of crowd
simul
a
tion, it
is normal that each agent
sho
u
ld
not in
terse
c
t
with e
a
ch
othe
r a
n
d
not i
n
terse
c
t with
othe
r
obje
c
t a
s
wel
l
. This i
s
fo
r t
he
purp
o
se to
visuali
z
e
the
solidity of the
corre
s
p
o
n
d
ob
je
c
t
in
vir
t
ual e
n
v
ir
on
men
t. U
s
u
a
lly th
ere
will be two
types of coll
ision avoi
da
nce; that
re
spo
n
se to st
atic ob
stacl
e
s and
dyna
mic
obsta
cle
s
. Howeve
r, the
mech
ani
cs
o
f
these
two
colli
sion avoi
dan
ce
s are the same
whi
c
h
con
s
i
s
t of Constructio
n
o
f
agent’s pe
rce
p
tion,
coll
ision d
e
tecti
on/pre
d
ictio
n
, and colli
sio
n
respon
se.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Veloc
i
ty
Perception: Collis
i
on Handling
Tec
h
nique for A
gent... (Nazreen Abdullasim)
2267
Figure 2. Coll
ision Avoida
n
c
e Behavio
r
Mech
ani
cs
Con
s
tru
c
tion
of agent’
s
pe
rceptio
n is th
e
firs
t
p
r
ocess of
colli
sion avoidan
ce beh
a
v
ior.
It
is abo
ut cre
a
ting a se
nsor-like area to d
e
tect in
terse
c
ting obje
cts so that the agent woul
d avoid
from colliding. There are
many ap
proaches of
designing
the sensor such
as
using ray-casti
n
g,
boun
ding vol
u
me te
chni
qu
e, sp
atial pa
rt
ition, and V
e
locity ob
sta
c
le
s te
chniq
ue
with ea
ch
of t
he
approa
che
s
come with the
i
r colli
sion d
e
t
ection alg
o
rit
h
m. Collisi
o
n
detection i
s
about collisi
o
n
testing bet
we
en the age
nts and the o
b
s
tacl
es
whi
c
h
later on will
pro
c
e
ss th
e collision
re
spo
n
se
whe
r
e in colli
sion avoid
a
n
c
e case is a resp
on
se
that moving the a
g
ent aw
ay from the obsta
cl
e.
Ho
wever, in this pa
per
we
will discu
s
s a
bout
the co
nstruction of the
agent’s p
e
rception.
3.2 Collision Handling
w
i
th
Agent’s P
e
rception
The pe
rcepti
on in colli
sion
avoidan
ce is the fi
eld of vision of the ag
ent. It is an area that
rep
r
e
s
ent
s a sen
s
o
r
to detect ob
stacle
s. Obstacl
e
ca
n be the age
n
t or any object in the virtual
environment.
Moreover there are
set
of rules to be followed
when testing
collision in virt
ua
l
environ
ment
with multi ag
e
n
t. This i
s
for
the purpo
se
t
o
maintain f
r
e
e
colli
sio
n
an
d create p
r
io
ri
ty
avoidan
ce be
tween age
nt with
ag
ent
a
nd
ag
ent with
ob
stacl
e
s.
Table
3 is th
e colli
sio
n
te
sting
rule
s betwee
n
obsta
cle
s
(agent an
d oth
e
r obje
c
t) a
n
d
perception.
Table 3. Colli
sion Te
sting
Rule
s
Main O
b
ject
Collision T
e
sting
Obj
e
c
t
T
e
st Validity
Perception Obstacle
True
Obstacle Perception
False
Perception Perception
False
Obstacle Obstacle
True
Colli
sion te
sting is u
s
u
a
lly a seri
es
of te
st betwe
en
each
obje
c
t’
s edg
es
or v
e
rtice
s
.
Compl
e
x obj
ects
app
are
n
t
ly have more edg
es a
nd
ve
rtice
s
thu
s
having mo
re
testing. Thi
s
will
cause m
o
re processing re
sources and will affect t
he efficiency of the whol
e process. Therefore,
it is imp
o
rta
n
t
to sim
p
lify the obj
ect
re
pre
s
entat
io
n
so th
at the t
e
sting
is le
sser an
d fa
st
er.
Gene
rally, bo
th perceptio
n
and ob
sta
c
l
e
will be d
e
fine as
bou
n
di
ng volume o
r
basi
c
p
r
imitive
sha
pe such a
s
sp
here and
box. Figure 3
is an
exampl
e of agent wit
h
field of vision.
4. Velocit
y
P
e
rception for Collision Handling
There a
r
e m
any method
s on con
s
tru
c
ting
age
n
t’s
perceptio
n. In openSte
er that
simulate p
e
d
e
stria
n
in st
eerin
g beh
avior, t
he perception is ba
sed on the
con
s
tru
c
tion
of
boun
ding vol
u
me an
d ray
-
ca
sting [7][1][11]. There ar
e also
wo
rks
on con
s
tru
c
ting the pe
rcep
tion
based o
n
sp
atial pa
rtition
i
ng an
d u
s
in
g ro
botic
ad
aptation
o
n
obsta
cle avoi
dan
ce su
ch as
reci
procal ve
locity app
ro
a
c
h[12][6]. Ho
wever, th
ere
is la
ck of
resea
r
ch on
desi
gnin
g
t
h
e
perceptio
n b
a
s
ed
on
ag
ent’
s
dyn
a
mi
c fo
cal
point. So
as to
h
u
man
perceptio
n, th
e si
mulation
o
f
agent’
s
dyna
mic p
e
rcepti
on will
able
to pro
d
u
c
e
more va
ria
n
t rea
c
tion t
o
wa
rd
colli
si
on
avoidan
ce be
havior. Figu
re
4 and 5 dem
onstrate t
he differen
c
e
s
b
e
twee
n fixed perceptio
ns
with
dynamic p
e
rception.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 4, April 2013 : 2264 – 2
270
2268
Figure 3. Agent’s Field of Vision
Figure 4. Coll
ision Avoida
n
c
e u
s
ing Fixe
d Perception
Figure 5. Coll
ision Avoida
n
c
e u
s
ing Velo
city Perce
p
tio
n
4. Discussio
n
and Futu
r
e
Work
In this pap
er we present
additional f
eat
ure to a
u
tonomo
u
s
ag
ent that the dynamic
toward ag
ent’
s
pe
rception
focal p
o
int m
a
y re
sult
a di
fferent re
acti
on toward
col
lision
avoida
n
c
e
behavio
r. Thu
s
in som
e
scenari
o
su
ch a
s
in figure
5, the faster a
g
e
n
t may react first sin
c
e it has
longe
r focal p
o
int relative t
o
its velo
city. In addition, th
e slo
w
e
r
do
e
s
not h
a
ve to
respon
se
sin
c
e
its perceptio
n
does n
o
t detect any ob
sta
c
le by hav
ing
shorte
r focal point hen
ce it
does n
o
t nee
d
to go for unnecessary collision
testing. The
dynamic percepti
on may
also be extended it
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Velocity Perception: Colli
si
on Handling
Te
chnique for Agent... (N
azreen Abdullasim
)
2269
possibility by
able to generali
ze avoi
dance
maneuver behavior since its
collision
detection
gene
rali
zing
both static a
n
d
dynamic o
b
s
ta
cl
es a
s
onl
y one type of
obsta
cle.
In our future
work, we wou
l
d like to p
r
op
ose
our
co
nst
r
uctio
n
of pe
rceptio
n field i
n
crowd
simulation application. We also
woul
d like to investigate the ot
her possibilities
t
hat can affect
human fo
cal
point and the
r
efore be
co
me
agent’s
a
r
gu
ments to
ward
its dynamic p
e
rception.
Referen
ces
[1]
I Millin
gton, J
Funge.
Artificial Intelligence fo
r Gam
e
s, Sec
o
nd Edition
. Mor
gan K
aufma
nn
Publis
her
s
Inc., 2009.
[2]
J Dijkstra, J
Jesurun, H T
i
mmermans. A
mult
i-a
g
e
n
t
cellu
lar
autom
ata mo
del
of
pe
destria
n
movement.
Pedestria
n an
d Evacuati
on Dyn
a
mics
. 200
2; 173–
18
0.
[3]
D He
lbi
ng,
L
Buzn
a, A J
oha
nsso
n, T
W
e
r
ner. Self-
O
rganiz
ed P
e
destria
n Cr
o
w
d D
y
n
a
mics:
Ex
pe
ri
me
n
t
s, Si
mu
l
a
tio
n
s
, a
nd D
e
si
gn
So
l
u
tion
s.
T
r
ansportat
i
on Sci
ence
. 2
005; 39(
1): 1–2
4.
[4]
CW Rey
nolds. Flocks, herds and schools: A
distributed behavior
al
model. 1987; 25–34.
[5]
SR Muss
e, D
T
halmann. A
Mode
l of
Hum
an
Cro
w
d
B
e
h
a
vior: Gro
u
p
In
ter-Relati
ons
hi
p a
n
d
Co
llis
ion
Detectio
n Ana
l
ysis.
Proc. W
o
rkshop of Co
mputer Ani
m
at
io
n and Si
mul
a
ti
on of Euro
gra
phics.
19
97;
39–
51.
[6]
SJ Gu
y
,
J Ch
hug
ani, C Kim
,
N Satish, M Lin, D Man
o
c
ha, P Dub
e
y
.
ClearP
a
th: hi
ghl
y p
a
ra
lle
l
collis
io
n av
oi
d
ance
for m
u
l
t
i-age
nt simu
l
a
tion.
Proceedings of
the 2009 ACM SIGGRAPH
Eurogr
aph
ics Symp
osi
u
m on
Computer An
i
m
ati
o
n
. Ne
w
O
r
lea
n
s. Louis
i
a
na. 200
9; 177
–
187.
[7]
C Re
yn
olds. S
t
eerin
g Beh
a
vi
ors for Autono
mous Ch
aract
e
rs.
Game D
e
velo
pers Co
nf
erenc
e.
San
Jose. Cal
i
forni
a
. 1999; 7
63–
7
82.
[8]
S Z
hou, D Ch
en, W
Cai, L
Luo, MYH
Lo
w
,
F
T
i
an, VSH T
a
y
,
DW
S
Ong, BD Ham
ilton. Cro
w
d
mode
lin
g an
d simulati
on tech
n
o
lo
gies.
ACM T
r
ans.
Model. Co
mp
ut.
Simu
l
.
, 2010; 20(4):
1–3
5.
[9] C
Ericson.
R
eal-T
i
m
e
Co
lli
sion
Detecti
o
n
(T
he Mor
g
a
n
Ka
ufman
n
Series
in
Inte
ractive
3-
D
T
e
chno
logy)
.
Morga
n
Kaufm
ann. 20
05.
[10]
N Pe
lech
an
o,
JM Allb
eck, NI
Bad
l
er. Vi
rt
ual
Cro
w
d
s
: Meth
ods, Sim
u
lati
o
n
, an
d
Contro
l,”
Synthes
i
s
Lectures o
n
Co
mp
uter Graph
i
cs and Ani
m
ati
o
n
. 200
8; 3(1): 1–1
76.
[11]
OpenSteer. Av
aila
bl
e: h
ttp://
w
w
w
.
r
e
d
3d.com/
c
w
r/steer/.
[12]
J van de
n Be
rg, MC Lin, D
Manoch
a
. Re
ciproc
al Ve
loci
t
y
Obstacles f
o
r Rea
l
-T
ime Multi-Ag
en
t
N
a
vi
ga
ti
on
.
IEEE International Conferenc
e
on Robotics And Autom
a
tion
. 200
8; 192
8–
19
35.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 4, April 2013 : 2264 – 2
270
2270
Evaluation Warning : The document was created with Spire.PDF for Python.