TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 9, September
2014, pp. 66
8
2
~ 669
0
DOI: 10.115
9
1
/telkomni
ka.
v
12i9.626
5
6682
Re
cei
v
ed Ma
y 14, 201
4; Revi
sed
Jun
e
26, 2014; Accepted July 1
6
,
2014
Kinect and Optimization Algorithm Based Mobile Robot
Path Planning in Dynamic Environment
Zhen
z
hong
Yu, Weicou
Zheng, Qiga
o Fan*, Xin Liu, Jing Hui
Ke
y
Lab
orator
y of Advance
d
Process Co
ntrol
fo
r Light Ind
u
str
y
(Min
i
str
y
of
Educati
on),
Jian
gna
n Un
iv
ersit
y
, W
u
xi, C
h
in
a
*Corres
p
o
ndi
n
g
author, em
ail
:
qgfan@j
i
an
gn
an.ed
u.cn
A
b
st
r
a
ct
Based
on e
n
vi
ron
m
e
n
tal aw
a
r
eness
and
effective p
a
th pl
ann
ing
alg
o
rith
m, effective ro
bot pa
t
h
pla
nni
ng ca
n b
e
achi
eved. In this pap
er, the Kinect sens
or, the latest visio
n
sensi
ng tech
nol
ogy, is used
to
perce
ive th
e o
b
stacles
an
d t
e
rrain
inf
o
rmati
on i
n
dyna
mi
c
envir
on
me
nt in
real-ti
m
e, w
h
i
c
h e
nab
les r
o
b
o
ts
to reali
z
e effec
t
ive path
planning tasks
in com
p
lex
dy
namic
envir
onm
e
nt. Using the real-
i
me RGB im
ag
e
and 3
D
i
m
ag
e
prod
uced
by the Kinect sens
o
r
, the mob
ile ro
bot peri
p
h
e
ral
envir
on
me
nt in
formati
on ca
n b
e
prob
ed. T
h
e i
m
prove
d
artificia
l
pote
n
tia
l
fie
l
d
path
pl
ann
in
g alg
o
rith
m is op
timi
z
e
d
by ge
n
e
tic
trust met
h
od.
As a r
e
su
lt, it can
so
lve t
h
e
loc
a
l
mini
mu
m
po
ints
and
target u
n
re
ach
abl
e pr
ob
le
ms
in
the
traditi
o
n
a
l
artificial p
o
tent
ial fiel
d al
gorit
hm. More
over,
it can
effectively i
m
prov
e th
e real-ti
m
e p
e
r
forma
n
ce of the
alg
o
rith
m, a
nd
eventu
a
lly
rea
l
i
z
e
t
he
opti
m
i
z
ation
of
re
al-ti
m
e
pat
h p
l
an
ni
ng tasks f
o
r a
robot i
n
dyna
mi
c
envir
on
me
nt. F
i
nally, the
ex
peri
m
e
n
tal sys
tem is set
u
p
to verify the effectiveness
of the prop
o
s
ed
meth
ods.
Ke
y
w
ords
: kin
e
ct sensor, arti
ficial p
o
tenti
a
l field, g
e
n
e
tic trust region, pat
h pla
nni
ng, mob
i
l
e
robotics
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
Path plannin
g
techn
o
logy
is one of th
e co
re issu
e
s
of intelligen
t mobile rob
o
t
and its
path pl
annin
g
in
dynami
c
enviro
n
me
nt is
a h
o
t an
d
difficult topi
c in the
field
of mobile
rob
o
t
resea
r
ch. Path plannin
g
algorith
m
an
d environ
me
ntal awa
r
en
e
ss m
e
thod a
r
e two imp
o
rtant
resea
r
ch issues. Commo
nly used p
a
t
h plannin
g
techn
o
logy i
n
clu
d
e
s
Tem
p
late Matching
algorith
m
s, m
ap
con
s
tru
c
ti
on pl
annin
g
t
e
ch
nolo
g
y an
d artifici
al int
e
lligen
ce
met
hod [1,
2]. The
template mat
c
hin
g
metho
d
is com
p
a
r
ed
the curre
n
t state of the robot with p
a
st experien
c
e
to
make d
e
ci
sio
n
s. So, the method is mu
ch more
de
pe
ndent on the
past expe
rien
ce of the rob
o
t.
Map buil
d
ing
path pla
nni
ng metho
d
is divide
d i
n
to roa
d
m
a
rk and
gri
d
method.
The
disa
dvantag
e
of this meth
od is a
poo
r real
-tim
e. Artificial intellig
ence path pl
annin
g
meth
od
applie
s the a
d
vanced artifi
cial intellige
n
c
e tech
nol
o
g
y
to mobile robot path pla
nning, incl
udi
n
g
artificial
neu
ral net
work,
e
v
olutionary
al
gorithm
s, fu
zzy logi
c
and
t
he info
rmatio
n fusi
on
and
so
on. But artificial intellige
n
ce path
planni
ng exist
s
le
arning
sam
p
le
difficult to obt
ain an
d lea
r
ni
ng
lag pro
b
lem.
Traditional
artificial pote
n
tial
field path plannin
g
method re
gard
s
the movem
ent
robot i
n
the
environ
ment
as a
kin
d
of
motion in vi
rt
ual a
r
tificial
stress field, ex
isting d
e
fect
s of
local
minimu
m point a
nd
target u
n
re
achable
proble
m
[3-5]. In t
h
is p
ape
r, m
e
thod b
a
sed
on
improve
d
a
r
tificial
potential
field i
s
use
d
and
em
ploy geneti
c
trust regio
n
al
go
rithm
to solve
the
probl
em
of th
e sub-goal
p
o
int of im
prov
ed a
r
tifici
al p
o
tential
field, made up of multiple
mi
ni
mum
global o
p
tima
l path, to realize the real-ti
m
e optimal p
a
th planni
ng.
Ravari, with h
i
s partn
er, stu
d
ied the mobi
le robot path
planni
ng and
navigation ba
sed on
video came
ra [6], but thi
s
meth
od
ne
ed to
conve
r
t 2D vide
o i
m
age to
3
D
model
with l
a
rge
amount
of
cal
c
ulatio
n a
nd t
he e
r
ror is bi
gger too.
Nav
i
gation of
mo
bile robot
lo
calizatio
n
b
a
sed
on wi
rele
ss sensor n
e
two
r
k technol
ogy
combi
ne the
wirel
e
ss
sen
s
or net
wo
rk
wi
th mobile
rob
o
t
bringi
ng ab
o
u
t expanding
its range of
percepti
on,
providin
g localizatio
n and
path planni
ng,
expandi
ng th
e ability of
the
rob
o
t nav
igation [7
-10]
. But Kine
ct
is a
kind
of
3D
body fe
el
ing
came
ra
with
the co
re tech
nology of im
age ide
n
tifica
tion, captu
r
in
g RGB ima
g
e
s 30 time
s
per
se
con
d
, and
also d
e
tectin
g image d
ept
h, comb
i
ned
with the 2D
p
l
ane imag
e p
hotography a
n
d
3D de
pth image photo
g
raphy tech
nol
ogy, posse
ss
ing the fun
c
tion
s of re
al-time dyna
mic
captu
r
e, im
ag
e recognition,
3D me
asure
m
ent, colo
r reco
gnition
an
d so
on. Mi
crosoft p
r
ovid
e
s
a
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Kinect and O
p
tim
i
zation Algorithm
Base
d Mobile Rob
o
t Path Planning in… (Z
he
nzh
ong Yu
)
6683
Wind
ows pl
atform SDK
ai
ming at th
e
device
body
sen
s
o
r
,
the extensio
n
of the
u
s
ing
V
C
++
prog
ram
m
ing
tools whi
c
h can ea
sily
extent
the
big
p
o
we
r of Kine
ct, providi
ng
a goo
d techn
i
cal
for that m
obil
e
ro
bot
syste
m
ba
sed
on
Kinect
cr
eate
real
-time
3D
terrai
n
mo
del.
Benavide
z
,
with
his pa
rtne
r, studied the na
vigation and t
a
rget tra
c
kin
g
system of
mobile
robot
based on Kin
e
ct
and verifie
d
that the reli
ab
ility of the system wa
s val
i
dated by exp
e
rime
nt [11]. Smise
k
with
his
partne
r
, did
resea
r
che
s
a
bout the met
hod
s of real
-time 3D mo
deling a
nd the targ
et obj
ect
orientatio
n m
e
thod an
d do
t
he quantitative analysi
s
for the mod
e
li
ng and p
o
siti
oning a
c
cura
cy
[12]. Csaba
with hi
s p
a
rtn
e
r, stu
d
ied th
e ob
stacl
e
av
oidan
ce
of m
obile
rob
o
t b
a
se
d on
Kine
ct
sen
s
o
r
and fu
zzy logi
c [13].
In this pape
r, obtaining 3
D
depth ima
g
ing and
RG
B image information of d
y
namic
environ
ment by employing
Kinect body sen
s
o
r
device, doing real
-time detectio
n
and coll
ecti
ng
local
enviro
n
m
ent informa
t
ion for m
obi
le ro
bo
t, to improve th
e
environ
ment
awa
r
en
ess o
f
sen
s
in
g a
b
ility of the
syst
em. Robot
re
-plan
s
and
a
d
just
s the
pa
th acco
rdin
g
to the
cha
n
g
e
of
real
-time envi
r
onm
ental inf
o
rmatio
n, to make the
effective use of local pla
nnin
g
, to concl
u
d
e
a
more
optimized path, to timely pro
c
e
ss the info
rm
ation of en
cou
n
tered
ra
ndom
obsta
cle
s
, so
as
to improve the overall pat
h plannin
g
of mob
ile rob
o
t perform
an
ce
. Among them, the impro
v
ed
artificial pote
n
tial field based on the alg
o
rithm of
gen
etic tru
s
t regi
on is ad
opted
, can effectiv
ely
improve the e
fficiency of pa
th plannin
g
for mobile robot
.
2. Kinematic
Modeling of the Rob
o
t
In the first pl
ace, m
o
tion
model
of mo
bile r
obot
i
s
establi
s
h
ed, as sh
own
in Figure
1.
Rob
o
t po
sition vecto
r
i
s
rep
r
e
s
ente
d
by a 3
D
vector X=
T
[]
xy
.
（
x
，
y
）
is the lo
ca
tion
coo
r
din
a
te of
the ro
bot in t
he two
-
dim
e
n
s
ion
a
l spa
c
e.
is the
ori
ent
ation of the
robot relative to
the X axis, th
at is, the angle betwee
n
sp
eed direct
io
n and X axis. It mainly has two pa
ramete
rs
controlling th
e mobile robo
t: speed v an
d orientatio
n angle
ω
.
(,
)
x
y
Figure 1. Rob
o
t Kinematic
Model
The kin
e
mati
cs e
quatio
n can be e
s
tabli
s
he
d by the robot kin
e
mati
cs m
odel:
co
s
0
si
n
0
01
x
yV
(1)
Liner velo
city V and orienta
t
ion angle
ω
are the control variable
s
o
f
the mobile robot,
is the velocity
of the orienta
t
ion angle.
3. Path Planning
w
i
th Ki
nect s
e
nsor
In ord
e
r to
a
c
hieve th
e se
curity of m
o
b
ile rob
o
t in u
n
kn
own dyna
mic e
n
vironm
ent, the
followin
g
tasks n
eed
to be
compl
e
ted: th
e dete
c
ti
on
of
terrain, o
b
st
acle
re
co
gniti
on, definition
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 66
82 – 669
0
6684
movable a
r
e
a
, path pla
n
n
i
ng an
d navig
ation, etc.
Ki
nect b
ody se
nso
r
devi
c
e
can ge
nerate t
h
e
depth field i
m
age flo
w
a
t
the spee
d
of 30 frame
s
pe
r secon
d
, recre
a
te the re
al-time
3D
environ
ment
nea
rby, co
mplete the
detectio
n
of
terrai
n
, ma
p buildi
ng, reco
gnition
of the
obsta
cle
s
a
n
d
target a
nd
pass thi
s
info
rmation to th
e com
pute
r
, so a
s
to lay the foun
dation
for
path plan
ning
of mobile rob
o
t.
Figure 2. Path Planning
wi
th Kinect
Acco
rdi
ng to
the built map
s
and
ob
stacl
e
loca
tio
n
inf
o
rmatio
n, the comp
uter
ca
lculate
s
the coo
r
din
a
tes of the mobile rob
o
t, target and ob
sta
c
le sp
ace poi
nt. According
to the results of
the coo
r
di
nat
es, the pote
n
t
ial field mod
e
l of obje
c
tives an
d ob
sta
c
le
s is
set u
p
, using
gen
etic
trust re
gion
algorith
m
for solving the sub
-
go
al
poi
nt, with multiple su
b-g
oal
point eventually
make u
p
glob
al optimal pat
h.
4. Path Planning Algorithm of Mobile Robo
t
4.1. The Improv
ement Measur
es of I
m
prov
ed Artificial Potential Field Method
In orde
r to overcome the d
e
fects of the
tr
aditional a
r
ti
ficial potential
field path plannin
g
method, ma
ke the followin
g
improvem
e
n
ts:
(1) The
targ
e
t
’s attra
c
tion f
o
r the
ro
bot a
nd ob
sta
c
le
s’
s repelli
ng fo
rce fo
r the
ro
b
o
t ca
n
be tran
sferre
d into a kin
d
of potential field st
rength
and the meth
od of
co
mput
ing strength
of
potential field
can be a
dopt
ed to repla
c
e
the traditional
vector contro
l.
(2) Coeffici
en
t
item
g
2
X
X
is adde
d in the re
pu
lsion ve
ctor,
so that the repul
sion
vector i
s
de
creased u
n
til reaching
the t
a
rget at the
same time
wh
en the robot
get clo
s
e to t
h
e
target p
o
int a
nd the
attract
i
on get
de
cre
a
se
d,
with th
e re
pul
sion v
e
ctor getting
decrea
s
e
d
to
0
mean
while. T
hus the p
r
o
b
lem of target
unre
a
c
habl
e
due to th
e re
aso
n
that the
ob
stacle
an
d
target point a
r
e too clo
s
e i
s
solve
d
.
(3)
For "dea
d
lock" p
r
obl
e
m
ca
used by
local
minimu
m point, the i
n
trodu
ction
of
"bridgi
ng
potential field
"
to guide the robot out of the lo
cal mini
mum point
s, namely an ad
ditional poten
tial
field
ad
d
U
is increased in the local minimum point.
4.2. Model of Impro
v
ed Artificial Pote
ntial Field
Acco
rdi
ng to the above me
asu
r
e
s
, mode
l of
artificial potential field is esta
blished:
(1) M
odel of
the attracti
ve potential of the target
for the om
ni-di
r
e
c
tional
mobile
bodywork is shown in Form
ula (2
).
2
1
att
g
2
()
(
,
)
Uk
X
XX
(2)
Whe
r
e:
g
(,
)
XX
——the di
sta
n
ce b
e
twe
en
the curre
n
t position a
nd th
e target point
of the
cente
r
point o
f
the mobile robot’s b
ody.
k
——the coefficient of the proportio
nal po
sition gai
n.
X
——the lo
cati
on of the rob
o
t’s ce
nter po
int in the movement sp
ace
T
[]
x
y
.
g
X
——po
s
ition
of the target points
T
gg
[]
x
y
.
(2) T
he mo
de
l of the repul
sion potential
of the
i
th static ob
sta
c
le fo
r the omni
-di
r
ection
al
mobile car b
o
d
y is sho
w
n i
n
the formula
(3).
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Kinect and O
p
tim
i
zation Algorithm
Base
d Mobile Rob
o
t Path Planning in… (Z
he
nzh
ong Yu
)
6685
2
g
2
0
re
p
s
0
0
11
0.
5
,
()
if
,
0
if
,
i
i
i
i
U
XX
XX
X
XX
XX
(3)
Whe
r
e:
1,
2
,
,
in
,
n
is the total numbe
r of sta
t
ic obsta
cle
s
;
,
i
X
X
——Th
e
sho
r
test distan
ce
betwe
en the
curre
n
t positi
on and the
i
th obs
tac
l
e
of the cente
r
of the mobile car b
ody;
0
——the effect
ive influence distan
ce of o
b
sta
c
le
s
；
——the coefficient of the proportio
nal po
sition gai
n.
(3) Th
e posit
ion of the dynamic ob
sta
c
le
and the
orientatio
n of the motion can be
obtaine
d with
the de
pth fi
eld ima
ge flo
w
at th
e
spe
ed of 3
0
fra
m
es
per second by
usi
n
g
the
sensory abilit
y of the m
obi
le environm
ent of the
Kinect. The environm
ent information
can’t
be
fully reflected
only con
s
id
ering the
po
sition, as
dy
namic o
b
sta
c
les a
r
e in m
o
tion. Whe
n
the
relative
spe
e
d
bet
wee
n
th
e dynami
c
o
b
sta
c
le
and
the robot i
s
i
ndu
cted i
n
to
the fun
c
tion
of
potential field
,
the model of
the repul
si
on potential
of the
r
th dynamic o
b
sta
c
le for the om
ni-
dire
ctional m
obile car bo
d
y
is set up an
d as sho
w
n in
Formula
(4).
2
g
2
0
re
p
m
0
0
11
0.5
,
si
n
()
if
,
0
if
,
r
r
r
r
r
VV
U
XX
XX
X
XX
XX
(4)
Whe
r
e:
1,
2
,
,
rm
,
m
is the total numbe
r of the mobile ob
st
acle
s
——propo
rtio
nal co
efficient
;
V
——the curre
n
t motion spe
ed of the mob
ile car b
ody,
max
m
ax
23
,
VV
V
r
V
——the curre
n
t motion spe
ed of the
r
th dynamic o
b
st
acle;
——the motio
n
orientatio
n of the mobile car b
ody
——the curre
n
t motion orie
ntation of the
r
th dynamic
obsta
cle;
(4)
Whe
n
the robot is in a l
o
cal mini
mu
m poi
nt, the paved over pot
ential is intro
duced to
solve
th
e pro
b
lem of
lo
cal
minimum
poi
nt, the m
odel of the paved over potential is sho
w
n a
s
in
Formul
a (5
).
2
gg
a
ad
d
ga
(,
)
(
,
)
()
0(
,
)
s
U
XX
XX
X
XX
(5)
Whe
r
e:
a
——the jud
g
i
ng dista
n
ce whether the m
obile car bo
d
y
has rea
c
h
e
d
the target p
o
int.
s
——the p
r
op
ortional
coeffi
cient.
So the overall
strengt
h of potential field of t
he omni-di
r
ectio
nal mo
b
ile car b
ody is sho
w
n
as Fo
rmula
(6).The
stre
ng
th of the overall potent
ial field that the Formul
a (6
) ad
d the paved o
v
er
potential whe
n
the robot is
at the local m
i
nimum poi
nt.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 66
82 – 669
0
6686
at
t
r
ep
s
r
ep
m
11
()
(
)
(
)
nm
ir
ir
UU
U
U
X
XX
(6)
att
r
eps
r
e
p
m
a
dd
11
()
(
)
(
)
(
)
nm
ir
ir
UU
U
U
U
X
XX
X
(7)
Based o
n
the
above model
, in the proce
ss of
the ro
b
o
t’s moveme
nt the minimum of the
sum
of the
st
rength
of the
potentia
l
field that
can
be
rea
c
h
ed i
s
seen as
the sub-g
oal point
by
each sam
p
lin
g perio
d and
multiple su
b-goal poi
nt fo
rm the global
optimizatio
n path. The su
m of
the strength
o
f
the potential
field
is
sh
own as Fo
rmula
(6). In
or
der to avoid the
shocks when t
h
e
sub
-
go
al poin
t
is nearby the local minim
u
m poin
t, the method of vector
synthe
sis is ado
pted
to
make the ju
d
g
ment wh
eth
e
r the ro
bot situates in the
local minimu
m point. if so, the strength
of
the potential
field co
ntain
s
the pave
d
o
v
er pote
n
ti
al, s
h
own as
in formula (7).
If s
e
t the
robot’s
maximum sp
eed be
and sampling p
e
rio
d
be
, the range that the ro
bot can rea
c
h in each
sampli
ng
peri
od i
s
a
ci
rcl
e
with the
curre
n
t location to
be its center
and
to b
e
its radi
us.
In orde
r to ensure the
motion stabili
ty and exec
ution efficien
cy, the spee
d of the rob
o
t’s
movement
can’t be too
small or to
o l
a
rge,
so th
at the su
b-goal
point can b
e
sel
e
cte
d
In
the
annul
ar regi
on
ma
x
0
ma
x
0
23
,
R
Vt
Vt
,
(0
,
2
)
. As shown in Fig
u
re 3, the a
nnula
r
sha
ded
part i
n
the figure i
s
the a
r
ea
of the su
b-
g
oal
point can b
e
sel
e
cte
d
. T
he poi
nt of the
sha
ded p
a
rt i
n
the figure can be d
e
scrib
ed a
s
co
s
xx
R
,
sin
yy
R
. Thus, formula
(6) a
nd form
ula (7
) is th
e function b
e
twee
n varia
b
les
and
,
ma
x
0
ma
x
0
23
,
R
Vt
Vt
,
(0
,
2
)
. Set
1
cos
zx
R
,
2
si
n
zy
R
,
12
(,
)
zz
z
. From F
o
rm
u
l
a (5
) and
Formul
a (6), i
t
can be
co
n
c
lud
ed that when
Usin
g ge
netic tru
s
t re
gion alg
o
rith
m for solving
the
obje
c
tive function of the sub-g
oal poi
nt sho
w
n a
s
F
o
rmul
a (8
) , that is, solvin
g a cla
ss
of linear
con
s
trai
ned
o
p
timization
problem
s a
nd
usin
g vect
o
r
synthe
sis me
thod jud
ge
o
u
t that the
ro
bot
locate in a l
o
cal mi
nimu
m point, Formula (8
) ad
d
ed with pav
ed over p
o
te
ntial is u
s
ed
as
obje
c
tive function.
att
r
eps
r
epm
11
mi
n
(
)
(
)
(
)
(
)
nm
ir
ir
UU
U
U
zz
z
z
(
8
)
Usi
ng the qu
adrati
c
app
ro
ximation, con
s
tru
c
t su
b-p
r
o
b
lem of con
s
t
r
aints trust re
gion.
TT
mi
n
(
)
0
.
5
kk
k
q
d
g
dd
G
d
(9)
..
s
t
2
k
d
，
kk
zd
Whe
r
e,
()
kk
U
g
z
,
k
is th
e ra
diu
s
of th
e tru
s
t re
gion
,
2
()
kk
U
Gz
, it is very
complex to
solv
e
k
G
so that
Qua
s
i-n
e
wto
n
method
is
use
d
to con
s
truct
He
ssi
on
matrix B for
rep
r
e
s
entin
g
k
G
approximatel
y.
k
d
is
the following tes
t
s
t
ep.
is the value
range of
R
and
.
The
symb
ol in
stru
ction the
a
l
gorithm
in
volved:
2
R
z
,
12
(,
)
kk
k
zz
z
,
1
(
1)
1
(
1)
2
(,
)
kk
k
zz
z
,
1
(
1)
1
(
1)
2
(,
)
ee
e
kk
k
zz
z
,
1
kk
k
yg
g
,:= rep
r
e
s
ent
assign
ment,
BFG
S
1
k
B
TT
T
T
//
kk
k
k
k
k
k
k
k
k
k
k
B
yy
y
dB
d
d
B
d
B
d
,
the ratio of actual de
cre
a
se am
ount
and the fore
cast
decrea
s
e a
m
ount is:
()
(
)
/
(0
)
(
)
kk
k
kk
k
kk
k
UU
rU
q
qq
zz
d
d
(10)
max
V
0
t
ma
x
0
Vt
R
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Kinect and O
p
tim
i
zation Algorithm
Base
d Mobile Rob
o
t Path Planning in… (Z
he
nzh
ong Yu
)
6687
Among them,
whe
n
re
solv
ing
1
mi
n
(
)
e
k
U
z
, Using
g
enetic
algo
rithm for q
u
ick solving
1
mi
n
(
)
e
k
U
z
obtain
s
the
i
t
eration
point
Supe
rior to t
he
current
po
int and th
us result i
n
the fi
nal
s
u
b-
g
o
a
l
po
int r
e
c
e
ivin
g the
va
r
i
ab
les
R
a
nd
. The fin
a
l
linear velo
city V an
d o
r
ient
ation a
ngle
ω
of the mo
b
ile ro
bot i
s
ca
lculate
d
out
u
s
ing
varia
b
le
s
R
and
, thus
the path
plan
ning of th
e
mobile robot i
s
obtain
ed.
V
m
a
x
T
0
2
V
m
a
x
T
0
/3
R
¦È
X
Y
C
u
rr
e
n
t
p
o
s
it
ion
O (
x
,
y
)
Figure 3. Selected
Ran
ge
of Sub-go
al Point
5. Experiment and Analy
s
is
5.1. The Sy
st
em Structur
e of the Mo
b
ile Robot Path Planning Based on
Kinect
The structu
r
e
of the control
sy
stem of the mobile ro
bo
t is mainly made up of Kinect body
sen
s
o
r
devi
c
e, com
puter,
mobile
rob
o
t. The
system
structu
r
e of th
e path pl
anni
ng is
as
sh
own in
Figure 4:
Figure 4. Con
t
rol Fram
ewo
r
k of t
he Mobi
le Rob
o
t Based on Kine
ct
The pe
rceptio
n of obst
a
cle
s
an
d terrain i
n
t
he dynami
c
environme
n
t
is reali
z
e
d
b
y
using
the rgb im
ag
e an
d 3
D
im
a
ge g
ene
rated
by the
body
sen
s
o
r
d
e
vice of Kin
e
ct,
make
the
re
al
-time
detectio
n
a
n
d
receive th
e surrou
ndin
g
dyna
mic e
n
vironm
ent.
The
com
put
er
analy
z
e
and
pro
c
e
ss th
e origin
al data
strea
m
of the 3D de
pth image, di
stan
ce sen
s
o
r
s a
nd rgb im
ag
e to
reali
z
e the id
entification a
nd po
sitionin
g
for t
he goa
l and obsta
cl
e. The cha
r
a
c
teri
stics of the
surro
undi
ng dynamic e
n
vironm
ent of the mobile
ro
bot are su
ppl
ied by the rgb image and
3D
depth ima
ge,
inclu
d
ing th
e ch
ara
c
te
rist
ics
of the
si
ze, distan
ce
a
nd color. T
h
e
self ori
entati
o
n
and targ
et po
sitionin
g
of the robot in u
n
known
enviro
n
m
ent is re
alized by usin
g SLAM algorith
m
or VF
H/VFH+ al
gorith
m
. The
co
ordi
nate of th
e
goal, o
b
sta
c
l
e
an
d robot
is
obtaine
d, the
improve
d
artif
i
cial p
o
tential
field is
con
s
tructed,
sub
-
g
o
a
l point i
s
cal
c
ulate
d
with t
he alg
o
rithm
of
the ge
netic trust
regi
on, th
e glo
bal
opti
m
al p
a
th
i
s
f
o
rme
d
with multiple sub
-
goal point by
the
positio
ning
o
f
the goal a
nd the
robot
itsel
f. The
movement
speed V a
n
d
orientatio
n
angle
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 66
82 – 669
0
6688
veloc
i
ty
ω
of the variable
s
of the moti
on co
ntrol i
s
obtaine
d fro
m
the com
p
u
t
er by the m
obile
robot to adju
s
t its movement state and
movement
m
o
tion and real
ize the free
-collision m
o
tio
n
.
5.2. Experiment
In order to ve
rify the effe
cti
v
ity of the sy
st
em con
s
tru
c
t
,
the exi
s
ting
con
d
ition
s
sh
ould
be
made use of:
robot
、、
、
b
o
d
y se
nsor
device
of t
he Mi
crosoft Kinect
c
o
m
puter
dyna
mic
obsta
cle
s
、
static ob
stacl
e
s. The effective experim
ent
system is
co
nstru
c
ted
and
the experim
ent
of the path plannin
g
is de
si
gned. The ex
perim
ental scene is
sho
w
n
as in Figu
re 5
.
Figure 5. Experime
n
tal Sce
n
e
The initial pa
rameters of the algorith
m
is sele
cted a
s
:
00
0.0
5
(
)
Uz
,
1
0.1
,
2
0.0
3
,
3
0.0
5
,
0.5
ab
,
1.5
M
,
1
0.15
,
2
0.
3
,
02
2
BI
,
1
0.3
5
,
2
0.
75
,
3
1.2
5
, N=20,
0.
99
c
P
,
0.05
m
P
,
max
10
0
T
,
0.
1
, length of the cod
e
l
=32,
max
V
0.3m/s,
0
t
3s,
0m
a
x
0
(5
/
6
,
0
)
zV
t
,
k
1,
2,
0.1,
n
2,
m
2,
s
0.3,
π
/2,
0
1m,
a
0.15m, the o
r
ig
inal point of t
he ro
bot
(1.00,0.25
), the go
al poi
nt (1.
35,3.6
5
),
the unit is
m. Accordi
ng t
o
the e
s
tabli
s
he
d coordin
a
te
system, m
u
ltiple typical p
o
i
nts in
the
proce
s
s of
th
e
mobile
ro
bot’
s
m
o
tion i
s
reco
rde
d
by
the
comp
uter. T
h
e fitting path
curve
of the
multiple poi
nts is
sh
own a
s
Fig
u
re
6. T
he pote
n
tial field
stren
g
th and t
he algo
rithm’
s execute time of the ty
pical point in the motion traje
c
tory is sho
w
n
as
Table 1.
Figure 6. The
Motion Traje
c
tory of the Mobile Robot
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Kinect and O
p
tim
i
zation Algorithm
Base
d Mobile Rob
o
t Path Planning in… (Z
he
nzh
ong Yu
)
6689
Table 1. The
Typical Point’
s Execute Ti
me and t
he P
o
tential Field
Strength in th
e Traje
c
tory
points x/m
y/m
time/ms
field
strength
1 1.00
0.25
17
43.635
2 0.98
0.83
19
41.729
3 0.99
1.17
20
40.109
4 1.01
1.42
16
38.987
5 1.07
1.57
15
35.679
6 1.17
1.97
17
30.250
7 1.25
2.09
10
27.629
8 1.30
2.31
21
24.873
9 1.42
2.42
14
20.298
10 1.57
2.65
19
19.234
11 1.62
2.80
22
16.378
12 1.64
3.05
16
12.394
13 1.61
3.33
19
9.234
14 1.57
3.52
9
4.983
15 1.35
3.63
12
0.002
From the an
alysis of the
mobile rob
o
t
’s
trajecto
ry, the experim
ent sho
w
s that the
behavio
r of t
he mo
bile
ro
bot ha
s g
ood
stability,
co
n
s
iste
ncy, an
d
contin
uity. The defe
c
t of t
h
e
method
of tra
d
itional
artifici
al pote
n
tial fi
eld
ca
n
be
ov
ercome
by th
e integ
r
ation
of the m
odel
of
improve
d
arti
ficial potentia
l field and g
enetic
al
go
rithm and it ha
s a faste
r
co
nverge
d spe
e
d
whe
n
the su
b-go
al point i
s
solve
d
. It can be
s
een f
r
om the expe
riment the m
obile ro
bot can
quickly avoid
the static a
nd dynami
c
obsta
cle
s
an
d rea
c
h the t
a
rget lo
catio
n
quickly wit
hout
colli
sion.
Thi
s
archite
c
ture can better achieve
t
he p
a
th plan
ning ta
sk
of the mo
bile ro
bot in t
h
e
dynamic e
n
vironm
ent.
6. Conclusio
n
This p
ape
r p
r
opo
se
s the
path plan
nin
g
met
hod of
the mobile
robot in the
dynamic
environ
ment
based o
n
the
Kinect b
ody
sen
s
o
r
d
e
vi
ce with th
e lat
e
st visio
n
se
nsin
g technol
ogy.
Obtainin
g the
real
-time current 3
D
terrai
n inform
ation by
usin
g
the body
sen
s
or device of
Kinect
can p
e
rceive
the dynamic environme
n
t informati
on
of the robot’s surrou
nding
s effectively and
usin
g the me
thod of improved artificial
potent
ial field based on the geneti
c
trust regi
on ca
n
overcome the
defect of the traditional a
r
tificial
potential
field method and re
alize the optimizatio
n
of the path
pl
annin
g
. The
system
of the mobile
ro
b
o
t based o
n
t
he Kine
ct an
d the meth
od
of
improved artificial
potential
field has been verified
that
it has a bett
e
r St
ability and practicabil
i
ty,
can
better m
eet the
dema
nd of th
e
real
-time
cont
rol,
reali
z
e th
e p
a
th plan
ning
ta
sk of the
mo
b
ile
robot in the
dynamic e
n
vironm
ent and
suppl
y a ne
w app
roa
c
h f
o
r the path
planni
ng of the
mobile robot.
Referen
ces
[1]
Z
hu DQ,
Y
an MZ
. Surve
y
on
techno
lo
g
y
of
mobil
e
ro
bot p
a
th pl
an
nin
g
.
Contro
l an
d D
e
cisio
n
. 2
010;
25(7): 96
1-9
6
7
.
[2]
Z
hou LF
, Xu F
.
A
method for p
a
th pla
n
n
i
ng of
mobil
e
robot a
ccounti
ng for u
n
certai
nt
y
.
Micr
oel
ectronics
& Computer
. 2
010; 27(
7): 86-
89.
[3]
Che
n
LB,
Y
o
u
B. Robot’
s
d
y
n
a
mic tracki
n
g
an
d
obst
a
cl
e-avo
i
di
ng b
a
s
ed o
n
evo
l
uti
o
nar
y artifici
al
potenti
a
l fie
l
d.
T
e
chni
qu
es of auto
m
ati
on a
n
d
app
licati
ons
200
7; 26(4): 8-
10.
[4]
Kuan
g F
,
W
a
n
g
YN. Robot p
a
th pl
ann
in
g b
a
sed
on h
y
b
r
i
d
artificial
pote
n
t
ial fiel
d/ ge
net
ic alg
o
rithm.
Journal of system
simulation
.
200
6; 18(3): 77
4-77
7.
[5]
Park M, Jeon
J, Lee M. Obstacle
avoi
da
nce
for m
obil
e
ro
b
o
ts usin
g artifi
cial p
o
tenti
a
l fi
eld
appr
oac
h
w
i
t
h
sim
u
late
d
ann
ea
lin
g.
Pr
oceedings of
the 2001 IEEE Inte
rnational Symposium
on Industrial
Electron
ics
, Pusan: Institute of Electrical a
nd
Electron
ics En
gin
eers. 20
01: 153
0-15
35.
[6]
Ravari
A
R
N,
T
aghira
d H D
and
T
a
mjidi
A
H. V
i
si
on-b
a
s
ed fuzz
y n
a
vi
gatio
n of mo
bi
le ro
bots
i
n
grassl
and e
n
vir
onme
n
ts[C].
in
Advanc
ed Intel
lige
n
t Mechatr
onics
.
AIM 2009. IEEE/ASME
Internation
a
l
Confer
ence
on
. 2009: 14
41-1
446.
[7]
Xu
e H,
T
a
o
y
i, MHX. Re
al-tim
e
path pl
an
nin
g
base
d
on
w
i
r
e
less se
nsor n
e
t
w
o
r
k for mob
ile ro
bot un
de
r
unkn
o
w
n
envir
onme
n
t.
Applic
ation res
earch
of computers
. 200
8; 25(7): 20
29-2
032.
[8]
Z
hu QD, W
u
YB. Path pla
n
n
in
g of mob
i
l
e
robots b
a
se
d
on the
w
i
re
les
s
sensor
net
w
o
rk.
Journ
a
l of
Canc
er Rese
ar
ch and
Cli
nica
l Oncolo
gy
. 201
0; 38(12): 1
13-
116.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 66
82 – 669
0
6690
[9]
Li W
F
. W
i
reless sensor net
w
o
rk and mob
ile r
obot contro
l. Beiji
ng:
Scie
nce press
. 200
9.
[10]
Sheu JP
, Pras
an KS. Ef
ficie
n
t path
pla
nni
ng a
nd
data
g
a
theri
ng
proto
c
ols for the
w
i
reless s
ens
o
r
net
w
o
rk.
Co
mp
uter Co
mmu
n
ic
ation
. 20
10; 33
(5): 398-4
08.
[11]
Patrick Ben
a
vi
dez. Mob
ile
Ro
bot
Nav
i
gati
on
and T
a
rget T
r
a
cking
S
y
stem.
Internatio
na
l C
onfere
n
ce
o
n
Systems En
gin
eeri
n
g
,
Albu
qu
erqu
e, Ne
w
Me
xic
o
. 201
1: 299
-304.
[12]
Jan S, Micha
l
J,
T
o
mas P
.
3D
w
i
th Ki
nec
t.
IEEE
International Confer
ence on Computer V
i
s
i
on
W
o
rkshops
. Ba
rcelo
na, Spa
i
n.
201
1: 1
1
54-1
1
60.
[13]
Csab
a G, Vámoss
y
Z
.
F
u
zz
y
Based Obstac
l
e
A
v
oid
anc
e fo
r Mobil R
o
b
o
ts
w
i
t
h
Kin
e
ct Sensor
.
IEEE
Internatio
na
l Symp
osi
u
m
on L
ogistics a
nd In
dustria
l Informatics
, Smolen
ic
e, Slovaki
a
. 20
12: 101-
10
5.
Evaluation Warning : The document was created with Spire.PDF for Python.