TELKOM
NIKA
, Vol.12, No
.4, Dece
mbe
r
2014, pp. 79
5~8
1
0
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i4.495
795
Re
cei
v
ed Se
ptem
ber 18, 2014; Revi
se
d Octob
e
r 24,
2014; Accept
ed No
vem
b
e
r
7, 2014
Cooperative Avoidance Control-based Interval Fuzzy
Kohone
n Networks Algorithm in Simple Swarm Robots
Siti Nurmaini
1
, S
i
ti Zaiton
2
, Ric
y
Firna
ndo
3
1,3
Robotic and
Contro
l Rese
ar
ch Lab, F
a
cult
y of Com
puter S
c
ienc
e, Univer
sit
y
of Sri
w
i
j
a
y
a,
Jl Ra
ya Pa
lem
ban
g-Prab
umu
lih, Km
32, Ind
e
ral
a
ya-Oga
nIli
r, Indonesi
a
2
Soft Computing Res
earch G
r
oup, F
a
cult
y o
f
Computin
g, Universiti T
e
knol
ogi Ma
la
ysi
a
813
10 Sku
dai,
Johor Ba
hru, Mala
ysi
a
e-mail:
siti_nu
rmaini
@
unsr
i
.a
c.id
1
, sitizaiton@utm.my
2
, blacklist161@y
ahoo.co.id
3
A
b
st
r
a
ct
A nove
l
tech
ni
queto c
ontro
l s
w
arm ro
bot
’
s
mov
e
me
nt
is p
r
esente
d
a
nd
a
naly
z
e
d
i
n
this
pap
er. It
allow
s
a
grou
p
of robots to
mov
e
as
a un
i
que
entit
y p
e
rforming t
he fol
l
o
w
i
ng functi
on
such as
obst
a
cle
avoi
danc
e at grou
p leve
l.T
he contro
l strat
egye
n
h
ances t
he mob
ile ro
b
o
t
’
s p
e
rfor
ma
n
c
e w
hereby th
eir
forthcomin
g d
e
cisio
n
s ar
e i
m
p
a
cted
by its previo
us
ex
peri
ences
duri
ng the
navi
gat
ion a
part fro
m
the
current ran
ge
inp
u
ts. Interval
F
u
zz
y
-
Ko
hon
en Netw
ork (I
FKN) algor
ith
m
is util
i
z
e
d
i
n
this strategy
. By
empl
oyin
g a sma
ll nu
mber o
f
rules, the IF
KN algor
ith
m
s can be a
dapte
d
tosw
arms rea
c
tive control. T
h
e
control strate
g
y
provi
des
mu
ch fa
ster resp
onse c
o
mpar
e
to F
u
zz
y
K
o
h
one
n Netw
ork
(F
KN) algor
ith
m
to
expecte
d eve
n
ts. T
he effectiveness of the pro
pose
d
techn
i
qu
e is also d
e
m
o
n
strated in a s
e
ries of practic
a
l
test on our ex
peri
m
e
n
tal by
usin
g five low
costrobots
w
i
th limited sens
or
abiliti
e
s an
d l
o
w
comp
utatio
nal
effort on eac
h
singl
e ro
bot in
the sw
arm.
T
h
e
results sh
ow
that sw
arm ro
b
o
ts base
d
o
n
p
r
opos
ed tec
hni
qu
e
have th
e a
b
il
ity to perfor
m
c
oop
erative
b
e
h
a
vior,
pr
oduc
e
s
mi
ni
mu
m c
o
llisi
on
and c
a
p
abl
e to n
a
vig
a
t
e
arou
nd sq
uare
s
hap
es obstac
l
es.
Ke
y
w
ords
: co
oper
ative av
oid
ance, sw
arm r
obots,
interva
l
fu
zz
y
-
k
o
h
one
n
netw
o
rks
1. Introduc
tion
A swarm i
s
a complex
a
daptive
syste
m
, wh
i
c
h
is
decentrali
ze
d
,
self-org
ani
zed a
n
d
who
s
e
indivi
dual
s a
r
e
si
mple, h
o
mog
eneo
us an
d
autonom
ou
s
[1]. In term
s
of ro
botics i
s
a
n
approa
ch for
coo
r
din
a
ting
multi-ro
bot sy
stem. This
is
the fields co
n
s
ide
r
ing a g
r
oup of relatively
simple
individ
uals able co
o
perate
s
to pe
rfor
m
com
p
le
x tasks, in a
decentrali
ze
d
mann
er
[2]
. T
he
con
c
e
p
t of swarm ro
bots i
s
ba
sed
on n
a
ture.Th
e
inspiration fo
und
s in the first line within
ani
mal
so
ciet
ie
s,
su
c
h
as
bir
d
s,
a
n
t
s and
bee
s.
S
o
cial
in
se
ct
s exhibit su
cce
ssful
beh
avio
r in p
e
rfo
r
min
g
compl
e
x tasks on the l
e
ve
l of the grou
p, and a
r
e a
b
le to elimina
t
e noise, e
r
ro
r, and failu
re
of
s
w
a
r
m m
e
mbe
r
s
.
The
rele
vant pro
p
e
r
ties
of ma
ny o
f
these
biol
og
ical
syste
m
s
are
revie
w
e
d
in
[3],[4]. The main objective of this approach i
s
t
he idea that collective in
telligence can arise from
the inte
ra
ctio
n of
a hi
gh
n
u
mbe
r
of
rel
a
tively simpl
e
unit
s
[5].
I
n
s
w
a
r
m
ro
b
o
t
sy
st
ems,
eac
h
robot mu
st
behave
s
by
itself acco
rding to
the
its states and env
iro
n
ments, a
n
d
if
necessa
ry,must co
ope
rate
s with othe
r robot
s in o
r
de
r to carry out a
given task.
Today, num
b
e
r of
swarm
robotic
syste
m
have b
een
propo
sed fo
r n
u
mero
us
appli
c
ation
s
whe
r
e h
u
ma
n interventio
n is not fe
asibl
e
such
a
s
ra
dioa
ctivity detection,
firefighting
and
landmin
e det
ection,the ro
bots ne
ed to
be dispe
n
sa
ble [6]. A large numb
e
r o
f
robots allo
w for
redu
nda
ncy
a
nd in
crea
se
the
robu
stne
ss of
the
sw
arm.The in
crea
sing
interest
i
n
swa
r
m
ro
b
o
tic
system in
dica
tes that empl
oying multiple
inex
pen
sive
and si
mple m
obile robot
s a
s
opp
osed to
a
singl
e expe
n
s
ive [7]. An
expen
sive ro
bot may b
e
able to
achie
v
e the task
b
u
t its failu
re
can
prove to be costly and da
ngerou
s in missi
on critic
al
applicatio
ns.
By building swarm of ro
b
o
ts
with eleme
n
tary feature
s
,
the same t
a
sk
ca
n be
achi
eved for a lowe
r co
st and increa
sed
reliability [6]. In sea
r
ch ap
plicatio
ns al
so hav
e an
a
d
vantage of l
a
rge
r
covera
ge of the se
arch
spa
c
e
and its simpli
city of impleme
n
tatio
n
. They
co
uld
perfo
rm expl
oration ta
sks
in a large-sca
l
e
area mo
re ef
ficiently [8]-[10]due to, the swa
r
m
sh
ares inform
atio
n about the environ
ment and
individual
me
mbers i
n
tera
ct with e
a
ch ot
her [1
1].With
out bei
ng a
b
l
e
to p
r
o
c
e
s
s
and
re
spo
nd
to
new info
rmati
on, a robot lo
se
s its ability to adapt.
The
swarm
ro
bots
nee
d to
be a
b
le to
pro
c
e
s
s an
d a
c
t
upon
any
ne
w info
rmatio
n
from it
s
environ
ment.
Otherwi
se, the rob
o
ts wil
l
have to
rely on a static set of rules which may
be
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 795
– 810
796
inade
quate if the rob
o
t’s ci
rcum
stan
ce
s
cha
nge
signi
f
i
cantly. In the new envi
r
on
ment, they must
cap
able
of a
v
oiding o
b
st
acle
s, contro
lling t
he g
r
o
up spee
d a
nd mo
difying
the inter-ro
bot
distan
ce [1
2],[13]. Several cont
rol a
c
t
i
on solution
s are
propo
sed robot
s a
g
ree
on th
e
i
r
spe
e
d
s
;theref
ore they nee
d to kno
w
the spe
e
d
s
of the neighb
ors’ rob
o
ts [14]
-[17]. However,
theimplem
ent
ation such al
gorithm
on
real robot
s
ca
n be
quite
difficult due to
thedifficulty
in
o
btaining the
neigh
bors’
spe
e
d
s
by o
b
se
rvati
on a
nd it produ
ces a
m
ount
o
f
comp
utatio
nal
resou
r
ces. In
addition, forcing e
a
ch ro
bot to carry such
re
sou
r
ce
s internally mean
s dupli
c
at
ing,
perh
a
p
s
unn
ece
s
sarily an
d expensive
assets [10
]. The ca
pability of the controll
er, sen
s
o
r
s an
d
comm
uni
cati
on syste
m
is
a signifi
cant
perfo
rman
ce
para
m
eter fo
r swa
r
m robot
s. Ho
weve
r, the
eco
nomi
c
co
st p
r
obl
em i
s
often a
s
soci
a
t
ed with
s
w
a
r
m a
p
p
lica
t
io
ns
du
e to
th
e la
r
g
e nu
mb
er
o
f
robot
s requi
red an
d the
building
of in
telligent ro
bo
ts with a
nu
mber
of se
n
s
ors fo
r vari
ous
para
m
eters i
s
expe
nsive.
Thus,
a lo
w
co
st swa
r
m robot platfo
rm
with inexp
e
n
s
ive sen
s
o
r
for
spe
c
ific
singl
e and multi task
sc
ena
rio is highly favourable.
The devel
op
ment of co
ntrol algo
rithm
s
for
swa
r
m robots i
s
a
ch
allengin
g
task a
s
the
global b
ehavi
o
r that em
erges from the
many inte
ra
ction
s
bet
we
en the robot
s is often ha
rd to
predi
ct
and experim
ents with
sw
arm robots a
r
e often expen
sive
and time co
nsumi
ng.In the
pre
s
ent
pa
pe
r,
a ne
w algo
rithm
Inte
rval
Fu
zzy
-Koho
nen Net
w
o
r
k (IFKN)
i
s
introdu
ced
in
s
w
a
r
m
robot
s re
sea
r
ch and dem
o
n
stratio
n
of
t
e
ch
niqu
e
for smalle
r
m
u
lti-robot syste
m
s.
Thi
s
pa
pe
r
is
orga
nized
as follows: Se
ction 2
prese
n
ts
coo
p
e
r
ati
v
e beh
avior
of swa
r
m
ro
bots. Se
ction
3
descri
b
e
s
the
desig
n process of
the propo
se
control strategy b
a
se
d on Interval
Fuzzy-Koh
o
n
en
Network (IFK
N) alg
o
rithm.
Sect
ion 4 di
scusse
s the
experim
ental
desi
gn and
result
s. Sectio
n
5
gives the con
c
lu
sion a
nd future work.
2. Proposed
Metho
d
Behavior-ba
sed de
sign i
s
t
he mo
st com
m
on wa
y to d
e
velop a
swa
r
m ro
botic
sy
stem. In
an iterative
way, the indi
vidual beh
avior of ea
ch
ro
bot is impl
e
m
ented, stu
d
ied and i
m
proved
until the
de
sired
coll
ectiv
e
be
havior i
s
o
b
taine
d
. In be
havior-b
a
se
d
swarm
rob
o
ts
de
si
gn,
inspi
r
ation i
s
often taken from the ob
se
rvation of
the behavio
rs
of so
cial anim
a
l
s
. The individ
ual
behavio
rs of
the robot
s whi
c
h re
sult
s in the
colle
ctive behavio
r of t
he swa
r
m. They m
u
st
integrate
sev
e
ral
goal
s
ori
ented beh
aviors con
c
u
r
ren
t
ly
in order to
rea
c
h
a
goal.
In this situ
atio
n,
multi-ro
botwo
uld com
m
uni
cate with e
a
c
h othe
r a
n
d
coop
erate t
o
execute a
spe
c
ific glo
bal
behavio
r.
With the
in
creasi
ng
req
u
e
s
ts
on th
e ap
plicatio
n of m
u
lti-ro
bot
syst
em for compl
e
x tasks,
the obsta
cle
avoidan
ce b
ehavior
o
f mu
lti-
r
o
bo
t s
y
s
t
em b
e
c
omes
mo
r
e
an
d
mo
r
e
cr
uc
ia
l [18
]
.
Many algorith
m
s have be
e
n
prop
osed for ob
stacl
e
a
v
oidan
ce of a single robot.
In addition, it is
found th
at re
alizin
g ob
sta
c
le
avoidan
ce of ea
ch
ro
bot in m
u
lti-robot
system
is n
o
t en
ou
gh.
Furthe
r, it re
q
uest
s
multipl
e
ro
bots to
h
a
ve the
a
b
ility to avoid ob
stacl
e
while
keepin
g
form
a
t
ion.
If a multi-rob
o
t system in the un
kno
w
n
environ
ment,
it become
s
m
o
re difficult to
realize ob
sta
c
le
avoidan
ce
while
keepi
ng f
o
rmatio
n [19]
. This
sy
ste
m
ca
n imp
r
ove
the effe
ctiveness
of a
rob
o
t
action in te
rms of the
perfo
rman
ce
and the
re
liability.However, the di
fficulty arise
s
in
coo
r
din
a
tion
and coop
eration of multi-ro
bot to per
fo
rm a singl
e, gl
obal task. Th
ere a
r
e imp
o
rtant
asp
e
ct
s
su
ch
as ho
w to d
e
s
ign
the l
o
cal
beh
avior
a
c
tion of
ea
ch
si
ngle?
an
d h
o
w
to
control
the
coo
perative behavior of a swarm?.
The ch
an
ce
of robots
coll
iding agai
nst
each othe
r i
s
yet another challen
ge in
swa
r
m
robot
s forma
t
ions. Some
re
sea
r
chers use th
e term colli
sion
avoidan
ce
synonymou
s
with
obstacle
avoi
dance [20],[21]. Collis
i
o
ns are
avoided by maintain
i
ng st
rict buffer di
stances
and
con
s
i
s
tent co
mmuni
cation
betwe
en ro
bots [
22].Co
operative avoidan
ce beh
avior
tend
s
t
o be
based o
n
sp
eed a
daptati
on, ro
ute de
viation by
on
e ro
bot only
and
route
d
e
viation by b
o
th
robot
s, al
so
a combin
ed
spe
ed a
nd
route a
d
justm
ent. Each
ro
bot could
co
mmuni
cate
with
surro
undi
ngrobots
in dyn
a
mic environ
mental
p
e
rce
p
tion a
nd it
should
be
abl
e to d
e
termi
n
e at
least one of the
informati
on con
c
e
r
nin
g
the rela
tive
po
sition, o
r
i
entation, an
d
sp
eed
of ot
her
robot
s. However, the
ab
ility of the robot
to
com
m
unicate d
e
pend
s
on th
e computatio
nal
resou
r
ces a
n
d
also the typ
e
and amo
unt
of s
ensors th
at are empl
oyed on the ro
b
o
ts [23].
Multiple robo
ts re
quire fre
quent u
pdati
ng
of
sen
s
o
r
-ba
s
ed
information b
e
tween e
a
ch
individual u
n
i
t
and they
must ma
nag
e a large a
r
ray
o
f s
e
ns
ory in
fo
r
m
a
t
io
n to
d
e
t
e
r
mine
its
environ
ment. Each sen
s
o
r
provide
s
som
e
input
about
the environ
ment aro
und
the robot
s. That
input bein
g
in
corpo
r
ated int
o
a kn
owl
edg
e base.
Fro
m
this kn
owl
e
d
ge ba
se, ap
p
r
op
riate
s
trate
g
y
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cooperative Avoidance
Control based Interval Fuzzy K
ohonen
Net
w
orks .... (Siti
Nurm
aini)
797
about
co
ntrol
actio
n
s take
n in
re
sp
on
se to the
inp
u
t is
gen
erated. Th
ese a
c
tion
s all
o
w
the
robot
s, to in
teract
with i
t
s su
rroun
di
ngs fo
r a
c
hi
eving the g
oal. Ho
weve
r, creating
a
n
d
maintainin
g these cont
rol action
s, as well as gathe
ri
ng new d
a
ta for the kno
w
l
edge ba
se p
o
se
s
signifi
cant
ch
alleng
es.
Thi
s
i
s
due
to m
e
mory
and
p
r
oce
s
sing
po
wer
are
i
ssu
es in thi
s
re
sea
r
ch
area.
Fuzzy logic
has been a widely used
met
hod for
swarm robot
s [6],[12],[25]. Other
method
s expl
ored
in literature i
n
cl
ude
n
a
vigation
b
a
sed on
intellig
ent data
ca
rri
er
system
s [1
4],
neural network [25],[26],
particl
e
swarm optimi
z
at
ion [27],[28], support
vector machine [17],
Fora
ging al
g
o
rithm [29] have bee
n explore
d
for this purp
o
se.Ho
w
ever, i
n
the hard
w
are
impleme
n
tation, som
e
in
strume
ntation
element
s
u
s
u
a
lly introdu
ci
ng some
sort
of unpredi
ctable
values in th
e
coll
ecte
d inf
o
rmatio
n of
e
n
vironm
ental
measurement
, calle
d u
n
ce
rtainty, su
ch
as
amplifier,
sen
s
ors, digital t
o
anal
og (DAC), an
alog
to digital con
v
erters (A
DC), and a
c
tuat
or.
These un
ce
rtainties
cau
s
e
difficulty in determini
ng th
e cont
rol a
c
tion in re
al time situation.
While
control a
c
tion
s a
r
e
dete
r
mi
ned
and
tune
d in
ce
rtai
n e
n
vironm
ental con
d
ition
s
,
it might
have
to
be ch
ang
ed i
n
other envi
r
o
n
ments [30],[
31].
Fuzzy logic systems (F
LS
s) empl
oy a mode of app
roximate rea
s
oning that ma
ke
s them
a suitabl
e tool to implem
ent a robu
st
robot
beh
avior tolerating
noisy and u
n
relia
ble sen
s
or
informatio
n.
Ho
wever, th
e
FLSs have
the comm
o
n
pro
b
lem
tha
t
they ca
nnot
fully han
dle
or
accomm
odat
e for the
lin
guisti
c
a
nd
nume
r
ical
u
n
c
ertai
n
ties a
s
soci
ated
wit
h
chan
ging
and
dynamic envi
r
onm
ent be
cause they
u
s
e p
r
e
c
i
s
e f
u
zzy set
s
. Hence, it nee
ds a
metho
d
to
overcome th
e
uncertai
n
ty probl
em to
re
alize
a
safe a
nd efficie
n
t of
swarm
ro
bot
s movem
ent.
It
mean
s
th
e swarm rob
o
ts do
n
o
t collide
with
both
of obsta
cle
s
an
d robot
s, and
they can rea
c
h a
destin
a
tion in
a small amo
unt of time.
In the real
wo
rld, num
ero
u
s
natu
r
al a
g
e
n
ts
like anim
a
ls to
recogni
ze thei
r envi
r
onment
s
just with lo
w
sen
s
itive se
n
s
ors
without
a geom
etric
map [32]. Th
ey
must lea
r
n
to recogni
ze
the
new envi
r
on
ment by itself. Hence, dev
elopin
g
adapt
ive techniqu
e
in real enviro
n
ment for swarm
behavio
r i
s
d
e
sireabl
e. Th
is p
ape
r d
e
scrib
e
s a
pra
c
tical impl
eme
n
tation that h
i
ghly intellige
n
t
and cap
able
swarm
ro
bots
for coo
perative avoidan
ce
throu
gh
env
i
r
onm
ental re
cog
n
ition
b
a
sed
on a
ne
w
alg
o
rithm
o
finterv
a
l
fuzzy-koh
o
nen network (IFKN).
The
algorith
m
i
s
i
m
pleme
n
ted t
hat
integrate
s
g
r
oup
s of
smaller l
o
w-cost r
obot
s, whi
c
h a
r
e
equip
ped
wi
th limited-ra
nge
comm
uni
cati
on ability and inexpen
sive distance
sensors.
3
.
Research
Method
In this paper,
the Fuzzy-K
ohonen Network
(F
KN) algorithm [30],[31] is extended to an
Interval Fuzzy-Kohonen
Network (IFKN) algorith
m. To overcome the
uncertai
nty in some
element
s inst
rume
ntation i
n
the hardwa
r
e implem
ent
at
ion, interval
fuzzy set
s
a
r
e u
s
ed. Interval
mean
s that the input dom
ains a
r
e
characteri
ze
d by interval fuzzy sets. In the IFKN algorith
m
desi
gn
only f
e
w
rule
s are
create
d
c
om
p
a
re
FKN alg
o
rithm fo
r
ge
neratin
g
spe
ed a
nd
stee
ring
angle of swa
r
m mobile rob
o
ts.Kohon
en
netwo
rk
ha
s the advanta
g
e
of the patterns
re
cog
n
itio
n
mech
ani
sm
w
hile the inte
rv
al fuzzy logi
c
plays a
rol
e
i
n
man
aging
the inp
u
t and
output p
r
o
c
e
s
s of
pattern
re
co
gnition.The
IFKN
control
algorith
m
i
s
i
m
pleme
n
ted
on e
a
ch rob
o
t to de
cid
e
i
ts
action
and
co
mputational
metapho
r in
spired by soci
al inse
cts.
3.1.
Pa
tt
e
r
n
Cl
as
s
i
fi
ca
ti
on
In ord
e
r to
en
able the
mobi
le ro
bot to av
oid the o
b
sta
c
le
s with
re
active action, th
e better
mappin
g
rela
tion bet
ween
the
sen
s
o
r
data a
s
in
put
and th
e
spe
ed
control a
s
output m
u
st
be
establi
s
h
ed. Environme
n
tal cla
ssification is mod
e
ll
ed to summa
rize all e
n
vironmental p
a
ttern
s
as de
scri
bed
in Figure 1.
Each
sampl
e
of such patt
e
rn
con
s
i
s
ts
of rang
e re
ad
ings o
b
taine
d
b
y
five
sen
s
o
r
s. For red
u
ci
ng compl
e
xity
of
stor
age, re
prese
n
tation an
d
lea
r
nin
g
a
l
ong se
quen
ce
of each sam
p
le is
mapp
e
d
to one
cla
ss
by IFKN
algorith
m
. All possibilitie
s
of mobile
ro
bot
environ
ment
s are
con
s
ide
r
ed t
h
ro
ugh
fuzzifying p
r
o
c
e
s
s an
d
co
mbining
the
s
e 15
cl
asse
s of
interval fu
zzy
rule
ba
se. T
he rule tabl
e
is con
s
tru
c
te
d exploiting t
he sequ
en
ce
of enviro
n
me
ntal
pattern
and
speed l
e
vels. I
n
this
strate
g
y
15 rul
e
s
are employe
d
t
o
ke
ep
fe
w
c
o
mpa
r
e to F
K
N
algorith
m
an
d IFKN algorit
hm.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 795
– 810
798
Figure 1. Environm
ental cl
assificatio
n
s
3
.
2
.
Interv
al Fuzzy
-KohonenNet
w
o
r
k
(IF
K
N) Design
The IFKN alg
o
rithm is p
r
o
posed to inco
rpo
r
ate lea
r
ni
ng rule
s to d
e
termin
e the distan
ce
and
simila
rity betwe
en the
input an
d in
itial pattern
s. Infrare
d
sen
s
ors a
r
e i
n
g
eneral u
s
ed t
o
obtain inform
ation reg
a
rdi
ng the local environ
m
ent.
The input p
a
tterns a
r
e
constructe
d fro
m
curre
n
t sen
s
o
r
rea
d
ing
s
an
d the algorith
m
desig
n is d
e
scrib
ed in th
e followin
g
st
eps:
Step 1. Fuzzi
fication d
esi
gn
It is obviou
s
that the interval fuzzy
set
s
a
r
e in
a re
gion
con
s
tru
c
ted byapri
n
ci
pal fuzzy
s
e
tmembers
h
ip func
tions
(FSMFs). An interval fu
zzy
system
is obt
ained
by u
s
in
g the fu
zzy set
s
to partition
th
e inp
u
t dom
ai
ns
of the
ba
seline FSM
F
s
with a
n
inte
rval
o
f unce
r
tai
n
ty as
sh
own
in
Figure 2. Th
e
inputs
(
)
w
h
ic
h
ar
e
d
i
s
t
anc
es
o
f
th
e
obs
ta
c
l
e
s
to
the s
e
ns
ors
.
F
u
zz
ific
a
t
ion
s
t
age
of interval fu
zzy syste
m
pro
duces fo
ur
d
egre
e
s
of inp
u
t membe
r
shi
p
, su
ch
as
up
per
nea
r, lower
near,
uppe
r f
a
r, an
d lo
wer far a
s
sho
w
n in Fig
u
re
2. The me
mbe
r
ship fu
nctio
n
s
that dete
r
m
i
ne
the degree of
farness o
r
ne
arne
ss to
the obsta
cle a
r
e
defined a
s
fol
l
ows,
0
1
15c
m
25c
m
40c
m
50c
m
µ
(
di
s
t
a
n
ce)
nea
r
fa
r
di
s
t
a
n
ce
Figure 2. Interval fuzzificati
on stag
e
In this wo
rk triangul
ar me
m
bership fu
ncti
ons
(MF
s
) a
r
e used. The
uppe
r an
d lo
wer
MFs
for interval fu
zzy sets can
be written in
equatio
ns (1),
(2), (3
) and
(4) re
sp
ectivel
y
,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cooperative Avoidance
Control based Interval Fuzzy K
ohonen
Net
w
orks .... (Siti
Nurm
aini)
799
μ
x
1,
x
1
5
,1
5
x
5
0
0,
x
5
0
(1)
μ
x
1,
x
1
5
,1
5
x
4
0
0,
x
4
0
(2)
μ
x
0,
x
1
5
,1
5
x
5
0
1,
x
5
0
(3)
μ
x
0,
x
2
5
,2
5
x
5
0
1,
x
5
0
(4)
Step 2. Euqlidian distan
c
e
For cal
c
ul
ating the Euqlidian dista
n
ce in upper
and lower m
e
mbe
r
ship
that is
respon
sibl
e for co
mpa
r
ing
the input pattern
(
and
(
with every initial pattern
and fin
d
the
winn
er ta
ke
s
all neu
ron.
T
he tw
o
eq
uati
onsca
n be
written in eq
uati
on (5)
and
(6) res
p
e
c
tiv
e
ly
,
̅
(5)
(6)
Whe
r
e,
is upper eu
clide
a
n
distance,
is
lowe
r eucli
de
an distan
ce,
n
is the number
of sen
s
or,
w
is
prototype pattern
an
d
j
is pattern in
dex. There a
r
e many initi
a
l pattern
s, whi
c
h
r
e
pr
es
e
n
t
a
c
h
ar
ac
te
r
i
s
t
ic p
a
tte
r
n
in
eve
r
y la
yer. T
he ori
g
inality
of
the learn
i
ng process
i
n
koh
one
n network is u
n
su
pervised, ho
wever that is
taking a lon
g
time in the training p
r
o
c
es, to
find the weig
hts that mee
t
the good p
e
rform
a
n
c
e.
To red
u
ce the spa
c
e
co
mplexity and to
facilitate fast learni
ng of sample
sequences by the IFKN algorith
m, all these patt
e
rns are set
as
a weig
ht in the distan
ce lay
e
r.
To obtai
n info
rmation
re
garding the
envi
r
onment
it i
s
e
s
sential to
ke
ep tra
c
k of
ch
ange
s
in the
pattern
of a
se
que
nce. Often th
e
chang
es
in
the
se
que
nce
are mo
re
impo
rtant than
sim
p
ly
the seque
nce
itself. For si
mple
cal
c
ulati
on, the
ru
le
b
a
se t
able i
s
u
t
ilized. Th
e pl
an of the
rul
e
‘s
table
i
s
cre
a
ted based on
environ
menta
l
clas
sif
i
cat
i
o
n
s,
w
her
e,
is
s
e
ns
or
1
,
is sensor 2,
is
s
e
ns
or
3
,
is sen
s
or
4,
and
issen
s
or5
.
The num
be
r of rul
e
s
eq
uals
with e
n
v
ironme
n
ta
l
patterns
.In this
work
, initial pattern are der
ived fro
m
previou
s
exper
im
ental
data base. The
pattern
sa
re a
s
soci
ated wit
h
a pair of mo
tor spe
ed referen
c
e M
1
an
d M2 as
sho
w
n in Tabl
e 1
.
Step 3. Rule
tables
Once the si
m
ilarity value is obtained
by us
in
g the Eu
clide
an di
sta
n
ce, then th
e
degree
s
of membership
can
cal
c
ulate
d
. The
r
e are two
similarity value
s
of upper an
d lower
membe
r
ship
are referred t
o
the equatio
n (7) a
nd (8
) resp
ectively,
1,
,
0,
(7)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 795
– 810
800
1,
,
0,
(8)
The mem
bership d
e
g
r
ee
rep
r
e
s
ent
s thesimila
rity be
tween
cu
rre
n
t
patterns
an
d
prototype p
a
tterns
, and
∈
(0,1)
.
Where
,
is up
per de
gree
of mem
bership,
is l
o
wer
degree of membe
r
ship,
is minimum Euclid
ean dista
n
ce and
is the
maximum
E
u
c
lidean di
stan
ce. The sum of
the membership deg
ree
out
puts
equal to 1. After th
e
rule ba
se a
n
d
simila
rity value
s
are
kn
own,the
n
the
spee
d of the motors
can
be determine
d by
finding th
e ru
le ba
se t
hat
has the
high
est level
of
similarity.
The results are calcul
ated with
the
referen
c
e spe
ed.
Table 1. Rul
e
base table
of IFKN
Pattern
Sensor Input
Motor
Output
S1 S2 S3
S4 S5
M1
M2
1
Far
Far
Far
Far
Far
90
90
2
Near
Far
Far
Far
Far
73
90
3
Far
Far
Far
Far
Near
90
73
4 Near
Near
Far
Far
Far
90
83
5
Far
Far
Far
Near
Near
83
90
6
Near
Near
Near
Far
Far
85
55
7 Far
Far
Near
Near
Near
55
85
8
Near
Near
Near
Near
Far
85
53
9
Far
Near
Near
Near
Near
53
85
10 Near
Near
Far
Far
Near
87
77
11
Near
Far
Far
Near
Near
77
87
12 Near
Far
Far
Far
Near
87
87
13
Near
Near
Far
Near
Near
86
86
14
Far
Far
Near
Far
Far
87
60
15 Far
Near
Near
Near
Far
85
53
Step 4.
De
ter
m
ine the outputs
Cal
c
ulate
the
crisp o
u
tput
value from av
erag
e m
e
mb
ership
d
eg
ree
.
The
avera
g
e value
s
of the upp
er
and the l
o
we
r deg
re
e of
membe
r
ship
as a
sin
g
le d
egre
e
of me
mbershi
p
for
each
pattern is
cal
c
ulate
d
by usi
ng equ
ation (9) belo
w
,
̅
/2
(9)
After the rule
base and m
a
ximum simil
a
rity pattern
s (
) a
r
e
kno
w
n, the mobil
e
robot
output
s are
determin
ed by findin
g
a rul
e
that
ha
s the
highe
st
level of simil
a
rity from up
p
e
r
and lo
we
r in
dex (
). The fi
nal value i
s
cal
c
ulate
d
by
mu
ltiplying t
he si
milarity l
e
vel of the
maximum de
gree of me
mbershi
p
with percent
a
g
e
of maximum left and right motor sp
eed
values, given
in equatio
n (1
0) and
(11
)
a
s
follow,
μ
∗
(10
)
μ
∗
(11
)
Whe
r
e,
is pe
rce
n
tage of speed for left motor outp
u
t,
is percenta
ge of spee
d
for right moto
r output,
is p
a
tterns ind
e
x whi
c
h ha
s a maximum
d
eg
ree of memb
ership,
left motor output speed p
e
rcentag
e an
d
is output right motor
speed pe
rcent
age for ea
ch
pattern.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cooperative Avoidance
Control based Interval Fuzzy K
ohonen
Net
w
orks .... (Siti
Nurm
aini)
801
3.3.
H
a
r
d
w
a
r
e
D
e
s
i
g
n
The experi
m
ental wo
rks a
r
e co
ndu
cted
to ev
aluate the swa
r
m’
s co
ntrolle
r perfo
rmance.
To validate t
he characte
ri
stic
of
swarm
behavio
ural,
the expe
rim
ents a
r
e im
pl
emented
on f
i
ve
smalle
r robot
s. The d
e
vel
opment of
th
ehardware archite
c
ture i
s
t
he initial work that ma
rks
the
transl
a
tion of
the desig
ne
d softwa
r
e int
o
the hardwa
r
e platform. In this pha
se
all algorith
m
s of
embed
ded
control
s
ystem
s
as the de
stination pl
at
form, are im
plemente
d
u
s
ing a sta
n
d
a
rd
microcontroll
er platfo
rm. The swa
r
m ro
bots pl
atfo
rm
employs AV
R ATMEGA
16 micro
c
ont
roller
seri
es a
s
the
main pro
c
e
s
sor fo
r mana
ging all mod
u
les. They u
s
e cl
ock at 8
.
0 MHz u
s
ing
the
internal
RC oscillator to
provideth
e
n
e
ce
ss
ary co
mputational power
for re
al-time sen
s
ory
system. T
h
is
seri
es ha
s 16
kilo
bytes
of
prog
ram
m
abl
e flash m
e
mo
ry an
d o
ne
kil
obyte of inte
rnal
SRAM that
provide
s
m
o
re than
enou
gh spa
c
e to
impleme
n
t basi
c
rea
c
tive beh
avior
an
d
different c
o
mplexity of s
w
arm algorithms.
The
swarm
robots have
three
infrare
d
s
ensors fo
r ob
stacl
e
d
e
t
ection
and
a digital
comp
ass for
headi
ng mea
s
ureme
n
t. Communi
catio
n
is the most important a
s
pe
ct of swa
r
m
robot
s the
r
ef
ore m
u
ltiple i
ndividual
s ca
n sh
are i
n
formation to fu
nction
as a
whol
e. An IEEE
802.15.4/Zig
Bee com
p
lian
t
X-Bee wirel
e
ss modul
e with a ran
ge
of approxim
ately 20 m indo
ors
are
used fo
r
comm
uni
cati
on bet
wee
n
robots and
be
tween th
e ro
bots
and
a
compute
r
con
s
ole.
Whe
r
e th
e mi
cro
c
o
n
troll
e
r
will serve th
e
input
wi
th its interrupt rout
ine, 20 M
H
z
AVR AT Me
g
a
16 micro
c
ont
rolle
r,whi
c
h
provide
s
a
control
step
duration of
110 ms is employed. The
microcontroll
er ca
n
b
e
p
r
ogra
mmed
th
roug
h a wire
less comm
un
ication
lin
k. The
l
o
w-po
wer
desi
gn of it
s
system
s let
s
swarm
ro
bots ope
rate
fo
r
10 h
ours
with
a 20
00
mAh
Li-Poly b
a
ttery.
Figure 3 sh
o
w
s the d
e
tail impleme
n
tatio
n
of hard
w
a
r
e
desig
n by usi
ng Proteu
s Simulator.
Figure3. Ha
rd
ware de
sign
4. Results a
nd Analy
s
is
Real
-time n
a
v
igation invol
v
es de
ci
sion
makin
g
a
c
cording to the
p
e
rception
of the lo
cal
environ
ment.
The fu
zzy i
n
feren
c
in
g m
e
thod h
a
s b
een
sho
w
n t
o
be
su
cce
s
sful in
real
-ti
m
e
navigation wi
th cluttered
environ
ment
s [30]. Ho
we
ver, when th
e environ
me
nt is filled with
obsta
cle
s
i
n
t
he form of l
o
ops,
mazes,
and
other
co
mplicate
d
structures the
robots tend
s t
o
lo
se
track of di
re
ction an
d gets t
r
app
ed. In th
is sectio
n vari
ous fu
ndam
e
n
tal ben
chm
a
rk p
r
o
b
lem
s
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 795
– 810
802
navigation a
r
e experime
n
tally demon
strated to allo
w
us to evaluat
e overall pe
rforma
nce and
to
examine po
ssible fault
s
in recognitio
n
a
nd com
m
uni
cation betwee
n
obje
c
ts.
In the implementation of
the propo
se
d IFKN
algorithm on real
swa
r
m ro
bo
ts, som
e
pra
c
tical
issu
es
are
expe
ct
ed to
re
solve.
In this
sectio
n, som
e
m
a
jo
r issu
es in th
e light
of som
e
prelimi
nary studies a
r
e p
o
inted out,
with an expe
rimental
setu
p com
p
o
s
ed
of a numbe
r of
mobile ro
bot
s. Before im
plementin
g the pro
p
o
s
ed
IFKN algorit
hm, it is necessary to perform
some
te
sts i
n
order to
dete
r
mine
the
stru
cture
of th
eIF
K
N in
cludi
ng
numbe
r
of inp
u
t patterns an
d
the environ
m
ental co
nditio
n
.
4.1. Obstacle
Av
oidance
Be
h
a
v
i
or in Single Robot
In ord
e
r to ve
rify the effe
ctiveness
of the
pr
o
p
o
s
ed
al
gorithm, m
obi
le ro
bot i
s
se
t up in
several environment
s. In this work, IFKN algo
rithm i
s
compa
r
ed t
o
FKN
algorith
m
in single
ro
bot
action
as de
monst
r
ated
in
Figu
re 4,
5 a
nd 6
re
sp
e
c
ti
vely for ob
sta
c
le avoi
dan
ce
behavio
r. F
r
om
the result
s,
both of th
e
algorith
m
s h
a
t uses in
si
ngle m
obile
robot
su
cce
s
sfully to p
e
rf
orm
obsta
cle avoi
dan
ce task.
Ho
wever, mo
bile rob
o
t ba
sed o
n
IFKN algorithm p
r
odu
ce
s stabl
e
movement, b
e
ca
use of in
IFKN algo
rith
m environ
me
ntal pattern
s
cha
ngin
g
in i
n
terval value
s
.
Therefore the
motor output
produ
ce
s co
nstant spee
d.
In the lon
g
corri
dor enviro
n
ment, si
ngle
mobile
ro
bot
based
on IF
KN alg
o
rithm
more
stable
co
mpa
r
e to FK
N al
gorithm, d
u
e
to the m
obil
e
ro
bot mov
e
ment al
way
s
keep
on i
n
the
centre of the corrido
r
e
nvironm
ent as shown in Figu
re 4(a
)
. The
recorded g
r
a
ph of the mobile
robot P
W
M
with FKN al
g
o
rithm an
d IFKN algo
rith
m whe
n
avoi
d
s t
he
wall a
s
sho
w
n in F
i
gure
4(b
)
.
(
a
) C
o
rr
idor
(b) PWM
output
Figure 4. Mobile rob
o
t mo
vement in the corrido
r
In the right corne
r
envi
r
on
ment, mobile
robot with IF
KN algo
rithm
prod
uce ke
e
p
a safe
distan
ce
f
rom
the rig
h
t wall com
p
a
r
e
to FKN alg
o
r
ithm. At the co
rne
r
a
r
ea
by usin
g IF
KN
algorith
m
, mobile rob
o
t is a
b
le to make t
he moveme
nt more sta
b
le than FKN al
go
rithm.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cooperative Avoidance
Control based Interval Fuzzy K
ohonen
Net
w
orks .... (Siti
Nurm
aini)
803
(a) Right
corn
er
(b) PWM
output
Figure 5. Single rob
o
t movement in the right wall
It’s
se
en
in Fi
gure
5(a)-(b) mobile rob
o
t move
ment
wi
th FKN al
gori
t
hm moving
close
r
to
the wall
when it makes
t
u
r
n left action,
whe
r
ea
s by
using IFK
N
algorithm
produ
ce
s sm
o
o
th
movement. T
h
is is b
e
cau
s
e in IFKN alg
o
rithm de
sig
n
,
the sen
s
ors
provide m
o
re
data within t
h
e
rang
e of interval for solvin
g impre
c
isi
o
n in the sen
s
or
s
yste
m wh
en it detect the enviro
n
m
ent.
While
in th
e
FKN
algo
rithm de
sig
n
, sensor
data
a
r
e exp
r
e
s
sed
only in
cert
ain valu
es. The
mobile
rob
o
t perfo
rman
ce
test is
co
ndu
cted by
pla
c
i
ng a m
obile
robot in
seve
ral environme
n
ts
su
ch as
sim
p
le environ
m
ent, complex
environme
n
t and the U-Shafe
s
ituation respectively
t
o
analysi
s
the compa
r
ison of the prop
osed
algorithm
. Figure 6
(
a
)
sho
w
s a traje
c
to
ry performe
d
by
one ro
bot usi
ng pro
p
o
s
e te
chni
que in
co
mplex enviro
n
ment co
ntai
ning four o
b
st
acle
s.
It can b
e
see
n
that si
ngle
mobile
rob
o
tmovem
ent uti
lize
s
IFKN
al
gorithm
more
ca
refully
comp
are to FKN algorith
m
, it keeps a di
stan
ce
from the unst
r
u
c
tured wall. Experimental resul
t
s
in clutte
red
e
n
vironm
ent with more ob
st
acle
s
as
depi
cted i
n
Fig
u
re 6(b).In thi
s
environ
ment,
we
dealt with
mo
re n
o
ise like
unstructu
re
d
wall. Th
e ex
p
e
rime
ntal results hig
h
light
the fact that
by
usin
g the IF
KN alg
o
rithm
ca
n en
han
ce environme
n
tal sen
s
ing
ca
pa
city. The
same
fact i
s
observed fro
m
the outpu
ts of
variou
s expe
rimen
t
s perfo
rme
d
in different
environm
en
tal
conditions.
In compl
e
x e
n
vironm
ents,
IFKN algo
rith
m can
solve
a pro
b
lem
co
mmonly en
co
untere
d
durin
g im
ple
m
entation
s
,
su
ch
as erro
rs in
ran
ge
readin
g
s due
to multiple
reflection
s
at
the
corne
r
s. In
complex an
d u
n
stru
ctu
r
ed
e
n
vironm
ents.
IFKN algo
rith
m are
able
to
pro
d
u
c
e
sta
b
le
movement
b
y
gene
rating
only 6
patte
rns,
compa
r
e
d
to FK
N
al
gorithm
p
r
od
uce
14
patte
rns.
Having goal on
the side o
f
the
wall
lo
n
g
-corrido
r
e
n
v
ironme
n
t
may
cau
s
e a mobile robot to
be
trappe
d
in a wro
ng boun
d
a
ry-follo
wing
dire
ction.
Th
e U-shap
e sit
uation i
s
diffi
cult to b
e
sol
v
ed
because the
sensing c
apability of infra-red sensor
s
and noi
ses in the sensor
s m
a
ke it
difficult
to
determi
ne th
e si
ze o
r
lo
cation of ob
st
acle
s
whe
n
this info
rmatio
n is requi
red
for the e
s
ca
pe
criterion.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 795
– 810
804
(a) Simple
enviro
n
ment
(b) Compl
e
x
environm
ent
(c) U-Sh
ape
situ
ation
Figure 6. Mobile rob
o
t traj
ectory
In some
con
d
ition, avoid-obsta
cle b
e
h
a
vior
may no
t function pro
perly and th
e
mobile
robotm
o
ve too clo
s
e to an obsta
cle, whi
c
h is a U-sha
pe
s
ituation. In such ca
se,
mobile rob
o
t is
requi
re
d to stop the move
ment or even,
in some
in
st
ances, the m
obile ro
bot ne
ed to make aturn
back move.
This
con
d
itio
n dep
end
s o
n
the safest
allowable di
st
ance bet
wee
n
the mo
bile
robot
and the
ob
st
acle. T
h
is
be
havior
ha
s th
e high
est
prio
rity. By using
an IFKN
algorithm
mobile
ro
bot
is able to
ma
ke turn left a
c
tion, but a
mobile robot
w
ith FKN alg
o
rithm g
e
tting
trappe
d in a
local
min
i
mu
m as
s
h
ow
n in
F
i
gu
r
e
6
(
c)
. T
h
is s
t
ra
te
g
y
p
r
o
duces go
od
a
c
tion to
guid
e
the mo
bile
ro
bot
out of the traps.
Evaluation Warning : The document was created with Spire.PDF for Python.