TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5331 ~ 53
4
1
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.519
9
5331
Re
cei
v
ed
No
vem
ber 2
4
, 2013; Re
vi
sed
March 21, 20
14; Accepted
April 6, 2014
A Virtual Physics-based
Approach to Multiple Odor
Sources Localization
Xiaoping Ma
1
, Yuli Zhang*
1,2
, Yanz
i Mi
ao
1
1
School of Infor
m
ation a
nd El
e
c
trical Eng
i
ne
e
r
ing, Ch
ina U
n
i
v
ersit
y
of Min
i
n
g
and T
e
chno
l
o
g
y
,
X
u
zh
ou
, C
h
in
a
2
School of Scie
nce, Dal
i
an Ji
a
o
T
ong Univers
i
t
y
,
Dali
an, Ch
ina
*
Corresp
on
din
g
author, e-mai
l
: zhang
yu
li
10
21
@hotmai
l
.com
A
b
st
r
a
ct
T
he detecti
on
of an o
dor so
urce l
o
catio
n
has b
een
en
h
ance
d
by us
in
g multi
p
le
plu
m
e-trac
in
g
mo
bil
e
r
obots.
So far,
ma
ny
researc
hers fo
cus o
n
l
o
catin
g
a s
i
n
g
le
sou
r
ce in
vari
ed
e
n
viro
nments. T
h
e
prese
n
t study i
s
concer
ne
d w
i
th the
pro
b
le
m
of multi
p
le
che
m
ic
al so
urc
e
s loc
a
li
z
a
t
i
o
n
usin
g
mu
lti-ro
bo
t
system. In
this
study,
mu
ltipl
e
gr
oups
of r
o
bots w
e
re
use
d
a
nd
c
oor
din
a
ted by a mu
lti-robot
coo
per
ati
o
n
strategy with virtual physics force,
which includes structur
e
form
ation forc
e, goal forc
e, repulsion forc
e and
rotary force. In order to test the effectiven
ess
of
the propos
e
d
strategy, plu
m
e
mo
de
l w
i
th tw
o sources was
constructed
by
computati
on fl
uid dy
na
mics simulati
ons
. Si
mu
lati
on ex
per
iment disc
usse
d the infl
ue
nce
of
the vari
ed fre
que
ncies
of w
i
nd d
i
rectio
n/ s
pee
d a
nd
me
t
han
e rel
eas
e
w
i
th different i
n
itial
pos
itio
ns of
mu
ltipl
e
gro
ups
to the search perfor
m
a
n
ce. Simulati
on co
mp
ariso
n
exp
e
r
iments usi
ng three ki
nds of pl
u
m
e
tracing a
l
gor
ithms: ch
e
m
ota
x
is, ane
motax
i
s and flux
ota
x
is w
e
re carried out res
p
e
c
tively an
d th
e
compar
ative re
sult abo
ut thre
e plu
m
e
tracin
g alg
o
rith
ms w
a
s illustr
a
ted.
Ke
y
w
ords
: virtual p
h
ysics for
c
e, plu
m
e traci
ng, source l
o
ca
li
z
a
tio
n
, mobi
le
robot
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
The recent i
n
crea
sing th
reat
of ha
za
rdou
s chemi
c
al age
nts tha
t
are relea
s
e
d
into an
environ
ment has highlig
hted
the
ne
ed for
su
pe
rior
detectio
n
of hazard
o
u
s
e
m
issi
on
sou
r
ce
s.
Odor source
l
o
cali
zatio
n
u
s
ing plu
m
e-t
r
a
c
ing
ro
bots
h
a
s th
e pote
n
tial to dete
c
t
such
dan
ge
ro
us
odor relea
s
e
d
from
sou
r
ces
su
ch
a
s
e
x
plosive
s
,
toxic o
r
harmful
gas an
d fire.
Haye
s et
al. [
1
]
bro
k
e do
wn o
dor so
urce
l
o
cali
zation
i
n
to
three
subta
s
ks:
plume
fin
d
ing-co
ming i
n
to conta
c
t with
the odor, pl
ume traversal-follo
wing t
he odo
r pl
u
m
e to its so
urce and
so
urce de
clarat
ion
-
determi
ning f
r
om od
or a
c
quisitio
n
ch
aracteri
stics
th
at the sou
r
ce
is in the im
mediate vici
n
i
ty.
Many ob
sta
c
l
e
s h
a
ve hi
nd
ered
od
or
so
urce lo
cali
zat
i
on in th
e pa
st two
de
cad
e
s. Su
ch a
s
t
he
unsta
ble
win
d
in
the
natu
r
al
enviro
n
m
ent, the
dete
c
tion
of the
o
dor an
d the
wind
with
mo
bile
robot
s. In thi
s
p
ape
r, the
experim
ents
were
se
tup
o
n
the a
s
sum
p
tion of a
st
rong
and
co
n
s
tant
velocity but varying di
re
ctional ai
rflow i
n
t
he enviro
n
m
ent. Mean
while, the rob
o
t
s can
dete
c
t the
odor a
nd the
wind p
r
e
c
isel
y.
So far, many
plume
-
tra
c
ing
resea
r
ch ha
s focu
sed o
n
l
o
catin
g
a si
n
g
le source
an
d many
strategi
es a
nd method
s have been
desig
ned,
su
ch
a
s
gra
d
ient-follo
win
g
-ba
s
e
d
stra
tegy
(ch
e
motaxi
s)
[2, 3], combin
ation of chem
otaxis
and a
n
e
motaxis [4, 5], infotaxis [6], evolutiona
ry
approa
ch [7]
,
model
-ba
s
ed
strategy
[8, 9], mu
ltiple robot
s
coope
ration
b
a
se
d o
n
swarm
intelligen
ce [
10, 11], virtual physi
cs
based st
rate
gy [12-14],
and st
rategy
fusing visi
on
informatio
n [1
5]. While
ma
ny of the
s
e
al
gorithm
s
wo
uld li
kely
su
cceed
in
fin
d
ing
on
e
so
urce
even in a mul
t
i-sou
r
ce envi
r
onm
ent, they offer no guid
ance on ho
w to partition the robot
s du
rin
g
a se
arch to
ensure
that
all so
urce
s a
r
e lo
cate
d in
minimal tim
e
, how to av
oid ob
sta
c
le
s and
other robot
s durin
g
glo
bal sea
r
ch,
and how
to contin
ue se
archin
g
for other
source
s on
ce
a
sou
r
ce ha
s b
een foun
d, and ho
w to avoid re-fin
din
g
the sam
e
sou
r
ce. In this problem, the
intensity of e
a
ch
so
urce m
a
y vary with time
and th
e
positio
n of the so
urce may
be o
cclu
ded
by
obsta
cle
s
o
r
other robot
s.
The
ab
ove-mentione
d
m
u
lti-so
urce p
r
oblem involv
es a va
riety o
f
distin
ct chall
e
nge
s that have received little a
ttention in the singl
e so
urce literatu
r
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5331 – 53
41
5332
This
pap
er
discu
s
ses an
extensi
on
o
f
our
ea
rlier wo
rk on
m
u
ltiple o
dor
sou
r
ces
locali
zation
u
s
ing
virtual
physi
cs ba
se
d ro
bots,
pu
blish
ed
re
ce
ntly in Adva
nce
d
Mate
ria
l
s
Re
sea
r
ch [16
], which had
discu
s
sed a new ap
proa
ch base
d
virtual physics for coordinatin
g two
grou
ps of m
obile
rob
o
ts i
n
the
sea
r
chi
ng
of two m
e
thane
so
urces in
op
en e
n
vironm
ent. The
results tol
d
u
s
that th
e
swarm b
ehavi
o
r u
s
e
d
in
th
e propo
se
d
approa
ch
en
sured th
e ro
bots
detect
all sou
r
ce
s.
He
re, we focu
s
on th
e detaile
d a
n
a
lysis
of th
e
colla
borative
control a
nd gi
ve
more
si
mulat
i
on
comp
ari
s
on exp
e
rim
e
nts o
n
th
ree
algo
rithm
s
: chem
otaxis, anemotaxi
s
and
fluxotaxis at three diffe
rent
frequen
cie
s
of wind
direct
ion/ spee
d a
nd methan
e relea
s
e with six
different initia
l po
sition
s of
mult
iple g
r
o
u
p
s, a
nd th
e
result
s
sho
w
t
hat the
propo
sed
st
rategy
can
effectively navigate the mobile ro
botics swa
r
m
s
to the ch
emi
c
al
sou
r
ces
and
the com
parative
result about three pl
ume tracin
g algo
rith
ms is illu
strat
ed.
2. Rese
arch
Metho
d
2.1. The Simulation Ar
en
as
The plu
m
e m
odel ad
opted
in this stu
d
y to test the co
ntrol st
rategi
e
s
was d
e
velo
ped by
[17]. This pl
u
m
e wa
s
sim
u
lated u
s
in
g
a co
m
putatio
n fluid dyna
mic (CF
D
)
software
pa
ckage,
FLUE
NT (Fl
u
ent, Inc.). Th
e plume d
a
ta
prod
uce
d
by
Fluent CF
D
were impo
rte
d
into MATAL
AB
(Math
w
orks, Inc.) and
integrate
d
with
simul
a
ted
mobile
ro
bot
s mo
del fo
r plumin
g tra
c
ing
behavio
r
sim
u
lation. T
he
gas in thi
s
pa
per was meth
ane
(CH
4
).
T
he
simulatio
n
environme
n
t we
use
d
in
the
pape
r
(figure
5)
wa
s
a 2
D
e
n
viro
n
m
e
n
t of 20
m×2
0
m. The
po
si
tions
of meth
ane
sou
r
ces were
(1, 1
1
) and
(3, 9). T
w
o
so
urces ha
d th
e same
rele
a
s
e
rate
s
whi
c
h were
50
0
kg
/m
3
-s. Varyin
g airflow di
re
ction
s
we
re a
dopted in
stea
d of the fixed
airflow direct
ion; the varying
airflow
entere
d
into the left-han
d sid
e
b
ound
ary of
the domain
at a con
s
tant 5
m
/s velocity but
varied directi
ons b
e
twee
n
±22.5º du
rin
g
the
simulat
i
on and exist
ed from the right-han
d si
de
boun
dary of t
he dom
ain. F
r
om figu
re 5
we
can
see t
hat, the plum
es fro
m
the t
w
o
sou
r
ces
a
r
e
conve
r
ge
d together, which
make
the group ro
bots
can not tell which
sou
r
ce the plume the
y
detecte
d co
mes f
r
om. S
o
, one
odo
r
sou
r
ce
can
b
e
tra
c
ked by
more
than
o
ne g
r
ou
p. Th
is is
pointle
ss the
waste of time with two
group
s fo
r
only a singl
e
sou
r
ce.
He
nce, the del
ay in
sea
r
ching th
e next odor
can p
o
ssibly occur. Th
e n
e
w propo
se
d
strategy wit
h
forbidd
en
area
setting shoul
d cop
e
with this di
sadva
n
tage.
2.2. The Con
t
rol Algorith
m
s
The
strate
gy ba
sed
virtua
l physi
cs
we
have
propo
sed
con
s
i
s
ts
of the foll
owi
ng th
ree
parts: plu
m
e finding, plum
e
traversal and
odor sou
r
ce decl
a
ratio
n
.
2.2.1. Plume
Finding
At the begin
n
ing, wh
en the ro
bot find
s no
plu
m
e, the rob
o
t wo
uld perfo
rm
passive
monitori
ng [1
8] to find plu
m
e, that is, the rob
o
t rem
a
ins
stationa
ry and waits fo
r an odo
ur pl
u
m
e
to interse
c
t the robot’
s
cu
rre
nt location
. The mal
e
si
lkworm
moth
employ
s this strate
gy.
When
trying to dete
c
t the plume
of pheromon
e relea
s
e
d
by
the female si
lkworm m
o
th, the male mo
th
waits, he
ad in
to the wind in
an exposed
pos
itio
n, until it detects the
pheromo
ne.
2.2.2. Plume
Trav
ersal
To make se
arching time
faster, we are usin
g pa
ral
l
el sea
r
ch by
two gro
u
p
s
’
robot
s.
Each g
r
o
up h
a
s
six rob
o
ts
and a virtu
a
l
robot, an
d
th
en there is to
tal twelve ro
b
o
ts u
s
ed fo
r t
w
o
odor sou
r
ce
locali
zation. Each
g
r
oup runs
by
itself. Members of
each g
r
ou
p
send an
d take
informatio
n a
m
ong th
eir
group. If one
g
r
oup
foun
d a
nd lo
cate
a source, the
group
woul
d
stop
moving.
2.2.2.1. The Con
t
rol Algo
rithms Ba
se
d Virtual Ph
y
s
ics for ea
ch
Group of
Ro
bots
The control
a
l
gorithm
s ba
sed virtual phy
sics
for e
a
ch
grou
p of six robots d
e
velo
ped by
[14] for
ro
bot
s a
r
e
ad
opte
d
in
this stu
d
y
. The
cont
rol
force i
n
cl
ude
s t
w
o
kin
d
s of
effort,
which
are
virtual stru
ctu
r
e force
F
(
VS
) (a
cting o
n
the six robots) and virtual g
oal force
F
(
VG
) (actin
g on
the
virtual robot).
Virtual stru
ctu
r
e force
F
(
VS
) can b
e
state
d
as follo
ws:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Virtual Physics-ba
sed A
ppro
a
ch to Multip
le Odo
r
Source
s Lo
cali
zation (Xi
aopi
ng Ma)
5333
6
22
2
2
1,
6
22
2
2
1,
(
)
cos(
)
(
)((
)
(
)
)
(
)
s
i
n(
)
(
)((
)
(
)
)
ki
x
k
r
k
i
s
kc
kc
k
c
ii
k
ki
ki
y
k
r
k
i
s
kc
k
c
kc
ii
k
ki
qq
F
V
S
k
k
xx
xx
y
y
r
d
qq
F
V
S
k
k
yy
x
x
yy
r
d
(1)
Whe
r
e
k
r
,
k
c
are po
sitive
g
a
in,
q
k
a
nd
q
i
are
the
K
-th and
i
-th el
ect
r
ic
ch
arges (
q
k
=
q
i
=1 in t
h
is
pape
r)
,
a
nd
d
ki
is the
distan
ce b
e
twee
n the
m
, (
x
k
,
y
k
) i
s
the po
sition of rob
o
t
k
.
ki
i
k
ki
ki
i
k
ki
d
y
y
d
x
x
)
sin(
,
)
cos(
.
The force
def
ined in
(1)
m
a
ke
s the
rob
o
t
move towa
rd the ci
rcl
e
with cente
r
(
x
c
,
y
c
) a
n
d
radiu
s
r
w
h
en (
x
k
,
y
k
)
≠
(
x
c
,
y
c
) an
d form
a regul
ar h
e
xagon.
We obtai
ned
united vecto
r
as follo
ws:
((
)
,
(
)
)
((
)
,
(
)
)
((
)
,
(
)
)
xk
yk
xk
yk
xk
yk
F
VS
F
V
S
FV
S
F
V
S
F
VS
F
V
S
(2)
Virtual go
al
forces
F
(
VG
) a
r
e
co
nstructed
by d
i
fferent plum
e-tra
c
in
g alg
o
rithm
s
:
chem
otaxis, anemotaxi
s
a
nd
fl
uxotaxis respe
c
tively.
The spe
c
ific con
s
tru
c
tion
method of
F
(
VG
)
is defined as
follows:
Chem
otaxi
s
: The gradient
strategy si
mp
ly follows t
he chemi
c
al g
r
a
d
ient, so the dire
ction
of the la
rge
s
t
ch
emical
co
nce
n
tration
is the g
oal
dire
ction. T
he virt
ual robot
re
ceives th
e
sen
s
o
r
data availabl
e on robot
k
(
k
=1, 2, 3
,
… , 6), choo
se
s the robot
j
wh
o
has th
e high
est
con
c
e
n
tration
and moves t
o
wa
rd it a distance of ste
p
length
s
1
. Th
e virtual goal force i
s
defin
ed
as
follows
:
()
()
()
()
()
jv
jv
Xt
X
t
FV
G
Xt
X
t
(3)
Whe
r
e
X
v
an
d
X
j
are po
sitions of virtual
robot an
d ro
b
o
t
j
.
ǁ
•
ǁ
repre
s
ents the Eu
clidean n
o
rm
operator.
Anem
otaxi
s
: The intuition
behin
d
the a
nemotaxis i
s
to move the l
a
ttice up
stre
a
m
while
kee
p
ing the
robots in
sid
e
the plume. If the
gas
co
n
c
entration
s al
l robot
s sen
s
ed exce
ede
d
a
pred
efined
th
reshold
ρ
T
an
d the
wind
ve
locitie
s
were
not all
ze
ro,
The virtu
a
l ro
bot re
ceive
s
t
h
e
s
e
ns
or
da
ta
ava
ila
b
l
e
on
r
o
b
o
t
k
(
k
=1, 2,
3, … , 6
)
, record
s all
the
wi
nd velo
cities :
6
2
1
,
,
,
v
v
v
and
cal
c
ulate
s
th
e average
wi
nd velo
city
6
1
6
1
i
i
v
v
. Then th
e virt
ual robot
ch
o
o
se
s
up
wind
dire
ction
v
and move
s
toward it a d
i
stan
ce of
step lengt
h
s
1
.
The virtual
goal force i
s
defined a
s
follows
:
()
v
FV
G
v
(
4
)
fl
ux
o
t
ax
is
:
In
this pa
per,
si
x robot
s com
posed of
reg
u
lar h
e
xago
n
grid
s. The
fo
rmula
of
cal
c
ulatin
g mass flux by a
robot is:
cos
v
n
v
(5)
Whe
r
e
and
v
are che
m
ical con
c
e
n
tration
and wind
ve
l
o
city of the robot re
sp
ecti
vely.
n
and
are sh
own in
Figure 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5331 – 53
41
5334
Figure 1.
Th
e
Method of Calcul
ating Flu
x
by a
Rob
o
t
Figure 2. The
Method of Calcul
ating Re
pulsi
on
Force bet
wee
n
the Virtual Rob
o
ts of Two
Grou
ps
If the ga
s
co
n
c
entration
s
al
l ro
bots sen
s
ed ex
cee
ded
a p
r
ed
efined
threshold
ρ
T
and
the
wind
velocity
wa
s n
o
t zero,
the virtual
ro
bot follo
wing
the ro
bot
i
wit
h
minim
a
l ne
gative ma
ss flux
will ta
ke th
e
robots up
win
d
whi
c
h
towards th
e
so
urce
. Then
the
virtual robot
ch
o
o
se
s th
e
rob
o
t
i
with mini
mal
negative m
a
ss flux a
nd
m
o
ves to
wa
rd
i
t
a di
stan
ce
of step
len
g
th
s
1
. T
he vi
rtual
goal force is
defined a
s
fol
l
ows:
()
()
()
()
()
iv
iv
X
tX
t
FV
G
X
tX
t
(6)
Then, the discrete
-time mo
del of the vi
rtual rob
o
t movements
can b
e
stated a
s
:
1
(1
)
(
)
(
)
vv
Xt
X
t
sF
V
G
(7
)
Whe
r
e
X
v
(
t
+1) and
X
v
(t)
are po
sition
s
of the virtual robot at time step
t
+1 and
t.
As the virtual
robot move
d
to a new p
o
s
ition,
the ro
b
o
ts wo
uld al
so move a di
stance of
step len
g
th
s
2
under the a
c
t
i
on of the virtual stru
ctu
r
e force
F
(
VS
).
So, the discre
te-time model
of the
robot movement
s can be state
d
as:
2
(1
)
(
)
(
)
kk
k
X
tX
t
s
F
V
S
(8)
Whe
r
e
X
k
(t+1) and
X
k
(t)
are po
sition o
f
the robot
k
(
k
=1, 2, 3, … , 6) at time step
t
+1 and
t
.
It
should be noted
that step
len
g
th
s
1
of the virtual
robot
sho
u
ld
less than th
a
t
of the
robot
s
s
2
such that the robots
c
a
n follow it.
2.2.2.2. The Con
t
rol Algo
rithms Ba
se
d Virtual Ph
y
s
ics bet
w
e
e
n each Gro
u
p
of Rob
o
ts
Parallel
Search with repul
s
ion
forc
e: P
a
rallel
search logi
cally m
a
ke
s
se
archi
ng time
faster. Several grou
ps of robot run an
d find odor
source
s se
parately. Bu
t, robot grou
ps ru
n
sep
a
rately ca
n make one
grou
p ca
nnot
find the other and
will not
kno
w
if it goes to the sa
me
odor
sou
r
ce a
s
the others. It is very inefficient
that one
odor sou
r
ce can b
e
tracke
d by more tha
n
one
group. In
additio
n
, the
robot
s from
o
ne g
r
o
up
are
movable
ob
st
acle
s
be
cau
s
e robot
s h
a
ve
a
physi
cal
sha
pe an
d foot p
r
int of
fi
nite size, the other group must
avoid collisions
with them.
In
orde
r to gu
arantee of the
positio
ns of o
ne group’
s ro
bots to be
aw
ay from the o
t
her, we
assu
me
that there
is
a re
pul
sion f
o
rce
f
v
bet
we
en the virtu
a
l
rob
o
ts. Th
e
repul
sio
n
force
f
v
between
the
virtual rob
o
ts
of two grou
ps is defined a
s
follows:
()
,5
0,
jk
jk
jk
v
XX
XX
r
XX
f
ot
he
r
w
i
s
e
(9)
Whe
r
e
X
k
an
d
X
j
are p
o
siti
ons of virtual
robot
s of t
w
o
grou
ps.
ǁ
•
ǁ
re
pre
s
ent
s
the
Euclide
an no
rm
operator.Th
e
spe
c
ific
cal
c
u
l
ation pro
c
e
s
s is sho
w
n in
Figure 2.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Virtual Physics-ba
sed A
ppro
a
ch to Multip
le Odo
r
Source
s Lo
cali
zation (Xi
aopi
ng Ma)
5335
Then, the virtual goal force
is modified b
y
:
()
()
v
F
VG
F
V
G
f
(10)
Parallel Sea
r
ch with re
pul
sion force and
forbidd
en are
a
setting
: Although repul
sio
n
force
betwe
en the
virtual ro
bots
of two g
r
ou
ps can
gu
ara
n
te
e the p
o
sition
s of o
ne g
r
ou
p’s
rob
o
ts to
be
away fro
m
the other, it ca
n not gua
ra
n
t
ee one o
d
o
r
sou
r
ce to b
e
tracke
d by
only one g
r
o
u
p
.
Con
s
id
erin
g t
he case
sho
w
n in
figure
3, we
assum
e
that, the g
r
oup first lo
ca
ted a
sou
r
ce
is
denote
d
by g
r
oup
1 an
d th
e so
urce
whi
c
h
wa
s first
lo
cated i
s
d
eno
ted by so
urce
1. So, when
the
distan
ce
bet
ween t
w
o virtu
a
l ro
bots is le
ss than
5
r
,
by analy
z
ing th
e force
s
F’
(
VG
) of th
e virtu
a
l
robot of grou
p 2, we ca
n e
x
plain this problem.
Figure 3. The
Method of Calcul
ating Fo
rce
s
of
the Virtual Ro
bot of Grou
p 2
Figure 4. The
Method of Setting the Forbidde
n
Area an
d the goal force
)
(
2
VG
F
vr
of
Grou
p 2’
Virtual Ro
bot
The ci
rcula
r
forbid
den
are
a
is its
cente
r
at the virtual robot of g
r
ou
p 1 and
radi
u
s
r
T
(
r
T
>
r
) which is shown as follows:
2
2
1
2
1
)
(
)
(
T
v
v
r
y
y
x
x
(11)
Whe
r
e (
x
1
v
,
y
1
v
) is the posit
ion of the virtual rob
o
t of group
1.
In ord
e
r t
o
b
y
pass the
so
urce1,
we m
odified the
g
oal force
F’
(
VG
) of
gro
u
p
2’ virtua
l
robot a
s
the rotary force
f
2vr
, whic
h is
s
h
own as
follows
:
j
f
i
f
f
vr
y
vr
x
vr
2
2
2
(12)
The follo
win
g
equatio
ns
expre
ss th
e
dire
cti
on of t
he rota
ry force in two di
fferent
conditions:
2
1
2
2
2
2
2
1
2
2
2
2
,
,
vrcc
y
vr
y
vrcc
x
vr
x
vrc
y
vr
y
vrc
x
vr
x
f
f
f
f
f
f
f
f
(13)
Whe
r
e (
f
x2vrc
,
y
2vrc
) is cl
ockwise force an
d (
f
x2vrcc
,
y
2vrcc
) is counte
r
cl
ockwi
s
e force
.
θ
1
is the angle
betwe
en
th
e positive dire
ction
of
x–axis and
the lin
e
con
n
e
c
ted t
w
o virtual
rob
o
t
s of two g
r
ou
ps
in cou
n
t
e
r
c
lo
ck
wi
se dir
e
ct
i
on.
A
nd
θ
2
is the angle be
tween the p
o
s
itiv
e dire
ctio
n of x–axis and
the goal force
F’
(
VG
) of the group 2 in
co
unterclo
c
kwi
s
e dire
ction.
)
(
)
(
1
2
2
1
2
2
v
v
vrc
y
v
v
vrc
x
x
x
f
y
y
f
(14)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5331 – 53
41
5336
And,
)
(
)
(
1
2
2
1
2
2
v
v
vrcc
y
v
v
vrcc
x
x
x
f
y
y
f
(15)
Whe
r
e (
x
1v
, y
1v
) and
(
x
2v
, y
2v
) are the
position
s
of
the virtual robot of gro
u
p1 and g
r
ou
p 2
r
e
spec
tively.
Let us no
rmal
ize it as follo
ws:
j
f
f
i
f
f
VG
F
vr
vr
y
vr
vr
x
vr
2
2
2
2
2
)
(
(16)
We can se
e,
whe
n
|
θ
1
-
θ
2
|<
π
/2, group
2 by force
F’
2vr
(
VG
) c
an
su
cc
es
sf
ully bypass
sou
r
ce
1 a
n
d
locate
sou
r
ce 2. It
sho
u
ld
be
noted
tha
t
once th
e virtual robot
of
grou
p
2
cho
s
e a
rotary meth
o
d
(cl
o
ckwise
or counte
r
clo
c
kwi
s
e), it
sh
ould
kee
p
on
until it escap
e
d
from
sou
r
ce
1
.
Whe
n
|
θ
1
-
θ
2
|>
π
/2, gro
up 2
by force
F’
(
VG
) can succe
ssfully move
away from
so
urce 1.
2.2.3. Source Declar
atio
n
Plume sou
r
ce
identification
is
the p
r
o
c
e
s
s wh
e
r
e
b
y the robot
s ident
ify the odor source in
the enviro
n
m
ent. This p
a
p
e
r ad
opted
source id
entific
ation i
n
[12], whi
c
h tell u
s
that, if the robots
surro
und
a susp
ecte
d emi
tter, and the t
o
tal mass flu
x
measu
r
e
d
b
y
the sen
s
o
r
grid
con
s
i
s
te
ntly
excee
d
s som
e
sm
all, empi
rically
-dete
r
m
i
ned th
re
shol
d
Φ
T
i
n
a
given n
u
mbe
r
of
step
s
n
s
, then
the robot
s ha
ve locali
zed t
he emitter.
3. Results a
nd Analy
s
is
In this section, we will test the search
efficiency of the propose
d multi-robot
system
based virtual
physi
cs fo
rce
at three diffe
rent fr
e
que
ncies of
wind
directio
n/ sp
ee
d and m
e
tha
n
e
relea
s
e,
whi
c
h are:
stan
d
a
rd frequ
en
cy, twice fre
q
uen
cy and t
r
eble fre
que
n
c
y. In additi
o
n
,
different initial positions of
tw
o groups
will al
so have great in
fluence
on the sear
ch efficiency.
So, we give six kinds of init
ial positio
ns (positio
n of the virtual rob
o
t), whi
c
h are: Position1
(18,
4)
, (18,14
), de
noted by P1;
Position2
(1
8,10), (18,
14
), denoted by
P2; Position3
(18,4
)
, (18,1
0
),
denote
d
by P3; Position4
(18,14),
(18,1
6
), den
oted
b
y
P4; Position5 (1
8,8),
(18
,
10), den
oted
by
P5; Position
6
(1
8,4),
(18,6
)
, de
noted
by
P6; T
w
o
gro
ups with
P1
have m
a
ximu
m spa
c
ing, th
en
P2 and P3, then P4-P6.
By simulatin
g
extensive
nume
r
ical se
arch
trials
for eac
h
parameter, we chose the
para
m
eter giv
en by able 1.
Table 1. Para
meters of Algorithm
k
c
r
k
r
s
1
s
2
ρ
T
Φ
T
n
s
0.0001
0.3[m]
5
0.06[m]
0.12[m]
0.005[kg/m
3
]
0.05[kg/m
2
•s] 10
3.1 Chemo
t
a
x
is
To get a better un
derstan
ding of the ef
fect
that we p
r
opo
se
d pa
ra
llel sea
r
ch ha
s on the
plume tra
c
ing
task, we give
a serie
s
of snap shot
s of the traci
ng od
or plum
es p
r
o
c
e
ss of the two
grou
ps’
robot
usi
ng
ch
em
otaxis in
si
m
u
lati
on
environment. A
s
a sample,
we introdu
ce
the
tracin
g od
or
plume
s
p
r
o
c
e
ss
at the tre
b
l
e freq
uen
cy
with six initial
positions
respectively shown
in Figure 5. Twelve ro
bots
are indi
cate
d by “
○
” an
d two virtual rob
o
ts are in
dicate
d by “+”.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Virtual Physics-ba
sed A
ppro
a
ch to Multip
le Odo
r
Source
s Lo
cali
zation (Xi
aopi
ng Ma)
5337
(a) P1
(b) P2
(c) P3
(d) P4
(e) P5
(f) P6
Figure 5. Plume Tra
c
ing P
a
ths u
s
ing
Ch
emotaxis
at the Tre
b
le Fre
quen
cy with
Six Initial
Pos
i
tions
Then,
we giv
e
the
sea
r
ch
time used by
sou
r
ce
s lo
ca
ting at thre
e
different fre
q
uen
cie
s
with six different initial posi
t
ions re
sp
ecti
vely.
(a) Stand
ard frequ
en
cy
(b) T
w
ice freq
uen
cy
(c) Tre
b
le fre
quen
cy
Figure 6. Co
mpari
s
o
n
s o
n
Search Time
used by
Lo
cating Two So
urces u
s
in
g Chemotaxis at
Thre
e Differe
nt Frequ
en
cie
s
with
Six Different Initial Pos
i
tions
Figure 5-6 tel
l
us that, the
prop
osed
se
arch
st
rategy
usin
g ch
emo
t
axis is effe
ctive and
obtain
s
10
0
%
su
ccess
rate; in ad
dition, with
th
e
increa
sin
g
of the wi
nd
dire
ction/
sp
eed
freque
ncy
an
d metha
ne
re
lease fre
que
ncy, the
sea
r
ch time
used
by the
robot
s in
crea
se
s. The
main
rea
s
o
n
i
s
that
the i
n
crea
sing
of th
e wi
nd
directi
on/ spee
d fre
quen
cy a
nd
methane
rele
ase
freque
ncy ma
ke the plu
m
e
drift up and
down dra
s
ti
cally, the robot
s usi
ng the chemotaxis m
o
ve
towards the
d
i
rectio
n of th
e
larg
est
ch
em
ical
co
ncentration
whi
c
h
make
the
tra
c
ing p
a
ths sim
ilar
to the variation pattern of the wind di
rection.
Fo
r e
x
ample, the averag
e time
is 780
s un
d
e
r the
ca
se the wi
nd dire
ction/
spee
d freq
uen
cy
and methane
rel
ease frequ
e
n
cy is stan
dard
freque
ncy,
while the
average
sea
r
ch ti
me is 118
0s unde
r
the ca
se
th
e
wi
nd dire
ction/ spe
ed
freque
ncy a
n
d
methan
e relea
s
e fre
q
u
ency i
s
trebl
e frequ
en
cy. Con
s
id
erin
g
six different i
n
itial
positio
ns
we
had te
sted,
Grou
ps
sta
r
t out from P3
spe
n
t the mo
st time to locate two
sou
r
ce
s
and g
r
ou
ps
start from P
2
sp
ent the
least time ev
en at thre
e
different fre
q
uen
cie
s
of wind
dire
ction/ spe
ed
an
d
meth
ane relea
s
e. For
exa
m
pl
e,
the average
time is 1
040
s with P1, 8
7
0
s
with P2, 1160
s with P3, 92
0s with P4, 9
70s
with P5,and 1090
s wit
h
P6 under th
e ca
se the wi
nd
dire
ction/ sp
eed fre
que
ncy and metha
ne rel
e
a
s
e f
r
equ
en
cy is
treble frequ
e
n
cy. The m
a
in
rea
s
on
is that
P1, P3 a
n
d
P6 have
a
sa
me
coo
r
din
a
te (18,4) which is fart
h
e
st
f
r
om t
w
o
sou
r
c
e
s;
the farther th
e initial positi
ons fro
m
two
sou
r
ces the l
onge
r time sh
ould be
spent
to located th
em.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5331 – 53
41
5338
In additio
n
, t
a
ke
a
c
count
of se
arch
efficien
cy
with
P1, P3 a
nd
P6, two
gou
rps
with
P1
h
a
ve
maximum sp
acin
g. Clo
s
e
r
position
s
al
ways m
a
ke two g
r
ou
ps
co
me in conta
c
t with ea
ch ot
her
and th
e
rep
u
l
s
ion
force
bet
wee
n
the
virt
ual
robot
s
of
two g
r
o
u
p
s
repel
s e
a
ch ot
her which
le
a
d
s
to the grou
ps spent mo
re t
i
me to find all the sou
r
ces.
So, different initial positio
n
s
of two grou
ps
usin
g ch
emot
axis also exert great influen
ce on the
sea
r
ch effi
cien
cy.
3.2. Anemotaxis
Figure 7 give
s the
plume t
r
aci
ng
path
s
of two
g
r
ou
ps of rob
o
ts
usi
ng an
emotaix
s
at the
treble fre
que
ncy with six i
n
itial position
s
re
spe
c
tively.
(a) P1
(b) P2
(c) P3
(d) P4
(e) P5
(f) P6
Figure 7. Plume Tra
c
ing P
a
ths at the Trebl
e Frequ
en
cy with Six Initial Positions
Figure 8 give
s the search
time use
d
by sou
r
ces lo
ca
ting at three
different freq
uen
cie
s
with six different initial posi
t
ions re
sp
ecti
vely.
(a) Stand
ard frequ
en
cy
(b) T
w
ice freq
uen
cy
(c) Tre
b
le fre
quen
cy
Figure 8. Co
mpari
s
o
n
s o
n
Search Time
used by
Lo
cating Two So
urces u
s
in
g Anemotaxis at
Thre
e Differe
nt Frequ
en
cie
s
with
Six Different Initial Pos
i
tions
Figure 7-8 tell us that
anemotaxi
s
obtain
s
100
% failure ra
te at three
different
freque
nci
e
s with
initial po
sition
P4, whil
e
ane
motaxi
s is effe
ctive a
nd obtai
ns 10
0% su
cce
s
s rate
with othe
r five different ini
t
ial position
s
.
The main
re
aso
n
is that,
P4 is at up
pe
r rig
h
t co
rne
r
of
the are
a
which i
s
clo
s
e t
o
the above
sou
r
ce an
d two g
r
ou
ps
wi
th initial posit
ion P4 are q
u
ite
clo
s
e. So, the plume-traci
n
g path of two grou
ps
a
r
e al
most the sa
m
e
. Anemotaxis whi
c
h movi
ng
towards the
upwi
nd di
re
ct
ion ma
ke
s t
w
o g
r
o
u
p
s
al
l locate
d the
sam
e
sou
r
ce that is lo
cated
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Virtual Physics-ba
sed A
ppro
a
ch to Multip
le Odo
r
Source
s Lo
cali
zation (Xi
aopi
ng Ma)
5339
upwi
nd the ot
her
sou
r
ce. F
o
rbid
den
are
a
and
rotary
f
o
rce could
no
t guara
n
tee t
he group to fi
nd
the sou
r
ce
d
o
wn
win
d
to t
he lo
cate
d
so
urce
(an
e
mot
a
xis ma
ke
s the
robot
s to
move to
ward
s the
upwi
nd dire
ct
ion,
not do
wnwin
d
di
re
ction) and
t
he
sea
r
ch faile
d. Co
nsi
deri
n
g initial positions
except for P4
, with the increasi
ng of the
wi
nd di
re
ctio
n/ spee
d freq
uen
cy and m
e
thane
relea
s
e
freque
ncy, th
e traci
ng path
s
of the ro
bot
s u
s
ing a
nem
otaxis is ve
ry similar to
ea
ch othe
r, so the
sea
r
ch time
use
d
by th
e
robot
s i
s
al
so simil
a
r.
T
h
e main
rea
s
o
n
is that the
increa
sing
of
th
e
wind di
re
ctio
n/ spee
d fre
quen
cy and
methane
re
le
ase frequ
en
cy make the
plume to hig
h
con
c
e
n
tration
s
which al
wa
ys exce
ede
d
threshold
ρ
T
and the
ro
bot
s u
s
ing th
e a
nemotaxis m
o
ve
towards the
u
p
win
d
di
re
ction
whi
c
h
ma
ke th
e t
w
o
group
s
run
an
d
find o
d
o
r
so
urces
sep
a
rat
e
ly
and
ma
ke th
e sea
r
ch tim
e
in
crea
se
sl
ightly. The
a
v
erage
time
are
le
ss tha
n
40
0s at th
ree
different freq
uen
cie
s
of wi
nd directio
n/ spe
ed fre
que
ncy and m
e
thane
rele
ase
which is fa
r le
ss
than the g
r
ou
ps u
s
ing
ch
e
m
otaxis. Gro
ups
start o
u
t from P1 spen
t the least time to locate
two
sou
r
ces, the
n
grou
ps
start
from P2 and
P3, then
gro
u
ps
start from
P5 and P6. F
o
r exampl
e, the
averag
e time
is 350
s with P1, 390s
with P2 and
P3 and 460
s with P5 an
d P6. So, more
disp
ersed init
ial positio
ns
of two gro
u
p
s
ca
n
ma
ke
the anemota
x
is more effe
ctive in parallel
sea
r
ch than the ch
emotaxi
s
.
3.3. Fluxotax
is
Figure 9-10 g
i
ve the plume
tracin
g path
s
and
se
a
r
ch time of two g
r
oup
s of ro
bot
s u
s
ing
fluxotaxis at the sta
nda
rd f
r
equ
en
cy, twi
c
e fr
equ
en
cy and treble
freque
ncy
with
three
differe
nt
initial positions respectively.
(a) P1
(b) P2
(c) P3
(d) P4
(e) P5
(f) P6
Figure 9 Plume Tra
c
ing P
a
ths u
s
ing Fl
uxotaxis
at the Trebl
e Fre
q
uen
cie
s
with
Six Different
Initial Positions
(a) Stand
ard frequ
en
cy
(b) Twi
c
e freque
ncy
(c)Trebl
e frequ
e
n
cy
Figure 10. Co
mpari
s
o
n
s o
n
Search Time
used by
Lo
cating Two So
urces u
s
in
g F
l
uxotaxis at
Thre
e Differe
nt Frequ
en
cie
s
with
Six Different Initial Pos
i
tions
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5331 – 53
41
5340
From
Figu
re
9 -10
we
can
see
that, the
perfo
rm
an
ce
of plume
tra
c
i
ng u
s
in
g flux
otaxis i
s
very similar t
o
that of anemotaxis.
Figure 5 -1
0 tell us that, the pro
posed
multi
-ro
bot co
operation st
rat
egy with ch
emotaxis
is very effecti
v
e and obtain
s
100%
success rate at
all
three differe
nt frequen
cie
s
win
d
dire
cti
on/
spe
ed and m
e
thane relea
s
e with six di
fferent init
ial positio
ns. Bu
t, the search
time using the
chem
otaxis i
s
mu
ch m
o
re t
han th
at of th
e an
em
otiaxi
s a
nd fluxota
x
is an
d al
so
the p
e
rfo
r
man
c
e
is influe
nced
greatly by differ
ent frequ
en
cie
s
wi
nd di
re
ction/ spee
d
and meth
ane
relea
s
e
and t
he
initial posotio
ns. The main
reason is th
at the
robots using the ch
emotaxis mo
ve toward
s the
dire
ction of the larg
est
chemic
al co
ncentration
whi
c
h ma
ke the
tracin
g path
similar to t
he
variation p
a
ttern of the
wi
nd directio
n and al
ways fi
rst lo
cate the
sou
r
ce do
wnwin
d
the ot
her
sou
r
ce. Fo
rbi
dden
area
a
nd rotary fo
rce
ca
n g
uarantee th
e ot
her
group
to
find the
sou
r
ce
upwi
nd to
th
e lo
cated
so
urce. In
addi
tion, we
can
se
e th
at the
varie
d
wind
directio
n/ sp
eed
freque
ncy a
n
d
methan
e re
lease freq
ue
ncy exert little influen
ce o
n
the se
arch
perfo
rman
ce
for
the robot
s u
s
ing fluxotaxis and
an
emot
axis. A
rea
s
o
nable
expla
n
ation i
s
that,
both
fluxota
x
is
and a
nemota
x
is co
mbin
e i
n
formatio
n a
bout wi
nd ve
l
o
city whi
c
h
can ma
ke th
e
two g
r
ou
ps
run
and fin
d
o
d
o
r
so
urce
s
se
p
a
rately
and
p
a
rallelly. F
u
rt
herm
o
re,
mo
re di
spe
r
sed
i
n
itial po
sition
s of
two gro
u
p
s
can make the anemotaxi
s
a
nd fluxot
axis more effe
ctive in parall
e
l search.
4. Conclusio
n
Aims of thi
s
study a
r
e
to
tackl
e the
ch
alleng
es of tracin
g pl
ume
s
and
lo
cate
multiple
odor sou
r
ces.
In this stu
d
y
a coop
erative
se
arch
strate
gy ba
sed
on
mobile
ro
botics swa
r
m
s
wi
th
virtual phy
sics force
s
wa
s adopte
d
to
control a
nu
mber
of mob
ile ro
bots to
locate t
w
o
o
dor
sou
r
ces. A n
e
w meth
od f
o
r avoidi
ng t
he on
e odo
r
sou
r
ce is t
r
a
c
ed by mo
re t
han o
ne g
r
ou
p is
introdu
ce
d. T
h
is m
e
thod
b
a
se
d o
n
a
rot
a
ry force
avoi
ds th
e
robot
s
to re
-lo
c
ate t
he
same
sou
r
ce
whi
c
h h
a
s
b
een lo
cate
d
by other
rob
o
ts, and l
e
a
d
s the
m
to
move toward
other
so
urce.
Simulation
experim
ents co
mpared th
re
e
plum
e-trac
i
n
g
alg
o
rithm
s
: chem
otaxis, anemotaxi
s
a
n
d
fluxotaxis and discusse
d the in
fluence
of the varied wind direct
ion/ spee
d freque
nci
e
s a
nd
methane
rel
ease freq
ue
ncie
s an
d di
fferent initial
position
s
of
two gro
u
p
s
to the sea
r
ch
perfo
rman
ce.
After co
ndu
cting the
exp
e
rime
nts,
we
may de
rive
the followi
ng
con
c
lu
sio
n
from
this re
se
arch:
first, the varied win
d
directi
on/ spee
d freque
ncy an
d
methane rel
ease freq
uen
cy
exert g
r
eat i
n
fluen
ce o
n
the search
e
ffici
ency
fo
r chem
otaxis but
not
fo
r anemotaxi
s
and
fluxotaxis; se
con
d
, mo
re
d
i
spe
r
sed i
n
itial po
sition
s
o
f
two g
r
o
u
p
s
, the m
o
re
effective by
pa
rallel
sea
r
ch when
using pl
ume
-
tra
c
ing al
gorithms: anem
o
t
axis and fluxotaxis. In the future, we
will
cho
o
se m
o
re
group
s
of robots to m
u
l
t
iple od
or so
urces lo
cali
zation in
the
more
compl
e
x
environ
ment i
n
whi
c
h inten
s
ity of each
sou
r
ce ma
y
vary with time, the positio
n of the sou
r
ce
may be
occlu
ded
by ob
sta
c
le
s o
r
oth
e
r
robot
s, od
or
plume
may b
e
mo
re tu
rbul
ent an
d furth
e
r,
transplant the
control
strate
gy on the true
swam
rob
o
ts in the experi
m
ent are
na.
Ackn
o
w
l
e
dg
ements
This work was finan
cially
suppo
rted b
y
the National Natural Scien
c
e Fo
un
dation of
Chin
a un
der Gra
n
t 613
0
3183, the S
p
eciali
ze
d Re
sea
r
ch Fu
nd
for the
Do
ctoral Progra
m
o
f
High
er Edu
c
a
t
ion of China
unde
r Grant 2012
0095
120
023 an
d Post
docto
ral Research Fu
nd
s of
Jian
gsu Provi
n
ce u
nde
r Grant 1101
107
C.
Referen
ces
[1]
Ha
yes
AT
, Martinoli
A, Go
od
man
RM. Distr
ibute
d
odor
so
urce
loca
lizati
o
n.
IEEE Sens
ors Journal
.
200
2; 2(3): 260
-271.
[2]
Sand
ini G, L
u
c
arini G, Var
o
li M.
Gradie
n
t
Driven S
e
lf-
O
rgani
z
i
ng Sy
stems
. Proc
ee
din
g
s of the
IEEE/RSJ Internatio
nal C
onfer
ence o
n
Intell
ig
ent
Rob
o
ts and
S
y
stems. Yok
oham
a. 199
3: 429-
432.
[3]
Nurmai
n
i S, T
u
tuko B,
T
h
ohars
i
n A R.
In
tellig
ent M
obil
e
Olfactio
n
of S
w
arm
Rob
o
ts.
IAES
Internatio
na
l Journ
a
l of Ro
bo
tics and Auto
mation (IJRA)
. 2
013; 2(4): 1
89-
198.
[4]
Ku
w
a
na Y, S
h
imo
y
a
na I. Ph
eromo
ne-g
u
i
d
e
d
mo
bil
e
ro
bot
that b
ehav
es
like
a s
ilk
w
o
r
m
moth
w
i
t
h
livin
g ante
n
n
a
e
as pheromo
n
e
sensors.
Internatio
nal Jo
urn
a
l of Rob
o
tics Rese
arch
. 199
8; 17(9): 924
-
933.
[5]
Ha
yes AT
, Martino
li A, Go
odma
n
RM. S
w
a
rm ro
botic odor loca
liz
ati
on:
off-lin
e o
p
t
imizatio
n an
d
valid
atio
n w
i
th
real
ro
bots.
Ro
botica
. 20
03; 2
1
(4): 427-
44
1.
Evaluation Warning : The document was created with Spire.PDF for Python.