TELKOM
NIKA
, Vol. 11, No. 12, Decem
ber 20
13, pp.
7745
~77
5
3
e-ISSN: 2087
-278X
7745
Re
cei
v
ed
Jul
y
14, 201
3; Revi
sed Aug
u
st
25, 2013; Accepted Sept
em
ber 6, 201
3
Control Structure Design for Man-Function Humanoid
Robot
Yifeng Cui
1
, Su Goog Shon*
2
, Hee Ju
ng B
y
un
2
1
School of Infor
m
atics, Lin
y
i U
n
iversit
y
(
X
U), T
he
Middle Par
t
of Shuang
lin
g
Road, Li
n
y
i 2
7
600
0, Chi
na,
Ph./F
ax: +
86-1
396
99
537
50
2
Departme
n
t of IT
Engineeri
n
g
,
T
he Universit
y
of
Su
w
o
n (
X
U), IT
Building,
445-7
43, San
2-2 W
au-ri,
Bong
dam-e
up,
H
w
a
s
e
ong-s
i
, Korea, Ph./F
ax: +
82-010-6
3
3
2
426
0
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: cui
y
ifen
g2
01
1@gma
il.com
1
, sshon@s
u
w
o
n.ac.kr*
2
,
h
e
e
j
u
ng
by
un
@su
w
on
.a
c.kr
2
A
b
st
r
a
ct
T
h
is pap
er pre
s
ents a new
hu
ma
no
id rob
o
t cont
rol structure
-Man-F
uncti
on
hu
ma
noi
d rob
o
t. T
h
e
sensi
ng d
e
vice
s w
o
rn on the
hu
man
body,
these dev
ices
w
ill produc
e
sign
als of jo
int
s
’
ch
an
ge w
h
e
n
people
m
o
v
e
. Com
p
uter of
the control system
rec
e
iv
ing the signals and processing them
, then issue
control
sig
nals
to the s
e
rvos
o
f
the ro
bot
at th
e sa
me
ti
me, c
ontrol
the r
o
b
o
t’
s
b
ehav
ior. F
o
r this re
aso
n
, a
control structur
e of hu
ma
n
’
s b
ehav
io
r to dete
r
mi
ne the ro
bot
’
s
b
e
h
a
vior for
m
e
d
. T
he hu
mano
id ro
bot ha
s
17 s
e
rvos
an
d
tw
o pressur
e
sens
ors, the
rotation
of
these
se
rvo
s
’
stee
ri
ng
ge
a
r
s
l
e
a
d
to th
e ro
bot’
s
beh
avior
cha
n
ges, a
nd
12 s
e
rvos corres
p
o
n
d
in
g to th
e h
u
m
a
n
b
ody
sens
ing
dev
ices, ot
her 5
serv
os u
s
ed
for the stab
ility
control
of
the
robot co
mbi
n
e
d
w
i
th the
pres
sure se
nsors. Based on
t
h
is
control structur
e,
some pi
lot tests of the sensi
n
g devic
e or se
rvo hav
e
be
en
don
e, the clos
ed-l
oop positi
o
n
control mode
has be
en ch
o
s
en an
d the
Kal
m
a
n
filter smo
o
thi
ng o
p
ti
mi
z
a
t
i
o
n
meth
od be
en us
ed
, the initial st
atic
w
a
lking co
ntrol
of the robot be
en rea
l
i
z
e
d
.
Ke
y
w
ords
:
ma
n-functio
n
, con
t
rol structure, human
oi
d ro
bot,
kalman filter, s
t
atic w
a
lking
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
A humanoi
d robot or a
n
an
thropo
morphi
c ro
bot
is a robot with its
overall ap
pea
ran
c
e,
based
on th
at of the
h
u
man
body,
allo
wing
in
t
e
ra
ction
with
made
-for-h
uman to
ols
or
environ
ment
s. In general, human
oid ro
bots h
a
ve
a torso with
a h
ead, two a
r
m
s
and t
w
o le
g
s
,
althoug
h
som
e
form
s of h
u
m
anoid
robot
s m
a
y mo
del
only
part
of t
he b
ody, for
example, fro
m
the wai
s
t u
p
. Some hu
man
o
id robot
s m
a
y also
have
a “fa
c
e”,
with
“eye
s”
and
“mouth”. Al
so
a
human
oid
ca
n do m
o
tion
s
simila
r to hu
mans. A
s
we
all kn
own, ASIMO [1] is the
first hu
manoi
d
robot to wal
k
on its own, it can
walk free
ly, co
mplete some
com
p
le
x actions such as “8
” sh
ap
e
wal
k
ing, up a
nd down stairs, bendin
g
, it
can even
d
a
n
c
e with the m
u
si
c, runni
ng in a spee
d of
six kilom
e
ters pe
r h
our
a
nd sha
k
ing
with pe
ople,
it seem
s li
ke
the plot of
a scien
c
e fi
ction
movie be
co
me a re
ality. ASIMO is a symbol
fo
r advanced
step in innov
ative mobility of
human
oid
ro
bot. But we
find that
huma
noid
rob
o
t
like ASIMO is li
mited to the
behavio
r of t
h
e
robot
own
control, ju
st t
o
im
itate h
u
m
an b
ehavio
r, but thi
s
n
eed to
sim
u
l
a
te the
com
p
lex
function
an
d l
engthy im
ple
m
entation
proce
s
s.
Th
ese
ro
bots can
o
n
ly do th
e
progra
mmed
fixed
action
o
r
the
activities u
n
der certain
ci
rcu
m
st
a
n
ces now. The
r
efo
r
e, we wa
nt
t
o
u
s
e
an
othe
r
way to desi
g
n
our ro
bot whi
c
h can sim
u
la
te human’
s m
o
tion more ea
sily and un
co
nstrai
ned.
In this pap
er,
we p
r
op
ose a ne
w type o
f
humanoi
d robot control
st
ru
cture. Th
e main
parts
of our
resea
r
ch are
the peopl
e and the
r
obo
t, people take part in the
control of th
e
human
oid
ro
bot a
s
the m
a
in control reason, an
d t
he comp
uter as th
e
controller li
nked t
h
e
peopl
e
a
nd
th
e
robot
to
get
her. Using th
e sen
s
o
r
device
s
whi
c
h wo
rn on the joint
s
of the huma
n
body, the
pe
rson’
s
state
of motion
is tra
n
smitted
to th
e servo
s
of th
e corre
s
po
ndi
ng joi
n
t of the
robot by
cont
roller in
a cert
ain of
control
method, so, the ro
bot in
ful
l
accordan
ce
with the mo
d
e
of action of the person’
s behaviors [2]. This
cont
rol method
can achi
eve the unrestri
c
t
ed
movement of
the ro
bot in
different e
n
vironm
ent
s, b
e
ca
use its
brain is
the h
u
m
an b
r
ain, t
h
e
robot as anot
her “o
wn
”
of the
peo
ple
in
imitation
of t
heir
own a
c
tivities. We
ca
ll this
kind
of
robot
control
stru
cture for “Man-F
u
n
c
tion
” co
ntrol
stru
cture. Of
cou
r
se,
fo
r the reason that we
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 774
5 – 7753
7746
can’t b
u
ilt a robot a
s
sa
me as
huma
n
body, in the Man
-
Fu
nction huma
noi
d rob
o
t co
ntrol
research, to maintain the
stability of robot in
motion, ensure the
consi
s
te
ncy of the robot wi
th
human b
ehav
ior, both are the found
ation
and
difficulty of our re
se
arch work.
The rest
of th
is pa
pe
r is organi
zed
as fo
llo
ws: in the
next se
ction i
s
ab
out the
control
stru
cture de
sign of o
u
r M
a
in-F
un
ction
robot i
n
cl
udi
ng pe
ople,
r
obot an
d
co
ntrolle
r pa
rts.
In
se
ction
3, is the
system
id
entif
ication
a
nd control
a
n
a
lysis
which we use
fo
r ch
oosi
ng a suit
able
control mo
de
for ou
r
control structu
r
e.
After t
hat, se
ction 4
prese
n
ts ab
out the
appli
c
ation
o
f
Kalman filter which
can
optimize o
u
r control
syst
em. Section
5 gives the static wal
k
i
n
g
experim
ental
results that proved
the feasibility of the
desi
gned ro
bot in the robotics
research.
Eventually paper is finali
z
e
d
with sh
ort concl
u
si
on
s an
d mention
s
a
bout future re
sea
r
ch plan
s.
2.
Man-F
unc
tion Humanoid
Robo
t Con
t
r
o
l Structur
e
The rob
o
t co
ntrol stru
ctu
r
e
we de
sign
ed
i
s
m
a
inly
comp
osed
of
thre
e p
a
rt
s:
peopl
e,
controlle
r, an
d robot. Peop
le is the “ori
gi
nal edi
tion
”, people’
s motio
n
deci
de the robot’
s
motio
n
prima
r
ily. Ro
bot looks like
a little
“copy” of the huma
n
in bodily fo
rm and
stru
ct
ure, it copy the
motion a
s
sa
me a
s
th
e p
eople. Bet
w
e
en the
pe
opl
e an
d the
ro
bot is the
re
al time
cont
roller
whic
h us
ed for realiz
ing the c
o
mmunic
a
tion and c
o
ordination of t
hem, c
ontroller is
the s
e
c
ond
“brain” of the
robot he
re [3]
.
2.1. People
In traditio
nal
human
oid
ro
bot control, p
eople
ju
st
co
ntrol th
e
rob
o
t
by control
rods an
d
buttons, o
r
by
voice a
nd to
uch, all th
e m
o
tions
of
the
robot ju
st un
d
e
r the
simple
orde
rs. But in
our
control structu
r
e, peo
ple as a p
a
rt
of the
cont
r
o
l sy
st
em,
o
n
it
s sho
u
lde
r
s,
elbo
w
s
,
h
i
p
joints,
kne
e
joints
and
a
n
kle
joint
s
t
o
set up
12
motion
sen
s
ors in
ord
e
r
to
obtain
the
informatio
n of
the moveme
nt cha
nge
s of
the huma
n
a
s
sho
w
n in Fi
gure
1. Each
sen
s
o
r
ma
de
by the Dyna
mixel AX-12
A actuato
r
which i
s
u
s
e
d
for the
servo
of the ro
bot
too, usin
g t
h
is
modified
gea
r
we
ca
n
get
the i
n
form
ation of
cu
rvatu
r
e
ch
ang
e a
n
d
the
rotatio
n
a
l velo
city of
each moving
part of the hu
man and
cont
rol the ro
bot easily [4].
Figure 1. The
Con
c
ept of Man-F
u
n
c
tion
2.2. Robot
The
con
s
id
ered robot, i
s
a sm
all h
u
m
anoid
ro
bot
kit made
by servos an
d ri
g
i
d ro
ds.
The robot
ha
s 17
de
gre
e
s of freed
om
(DOF
s):
5 in e
a
ch l
eg, 3 i
n
each a
r
m a
n
d
1 in th
e wai
s
t.
These DOFs
are p
r
e
s
ent b
y
the servo
s
of Dynamix
el
AX-12A actu
ator. This
rob
o
t is 25 cm ta
ll
and h
a
s
a total wei
ght of a
bout 90
0g. Each
se
rvos
a
nd sen
s
ors h
a
s its
FLASH-ROM an
d RAM,
they can com
m
unication wi
th the controll
er by digi
tal p
a
ckets. In the
feet of the robot, there are
2 pre
s
sure
se
nso
r
s
whi
c
h combine
d
with
5 servos
(on
e
in waist, 2 in hip joints a
nd 2 in ankle
s)
to remai
n
sta
b
le of the rob
o
t. Other 12
servo
s
in
ro
b
o
t’s shoul
der,
elbo
w, hip jo
ints, kn
ee joi
n
ts
and a
n
kl
e joi
n
ts a
r
e o
ne t
o
one
co
rrespond
en
ce
with the
sen
s
o
r
s which a
r
e
worn o
n
hum
an
body. 12 j
o
int
sen
s
o
r
s p
r
o
duce control
sign
al a
s
e
n
coders, 1
2
join
t servo
s
drive
rob
o
t to mov
e
as de
co
ders. The po
sition
and ap
plicati
on of all serv
os list in Tabl
e 1 [5-6].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Control Struct
ure Desi
gn fo
r Man-Fu
n
c
ti
on Hum
anoi
d
Robot (Yifen
g Cui)
7747
Table 1. Serv
os Desi
gn in
Rob
o
t
Se
rv
os Po
s
i
t
i
on
A
ppl
i
c
a
t
i
on
1,2
Ankel (horizontal)
Keep balance
3,4
Ankel (vertial)
Cop
y
h
u
man mot
i
on
5,6
Knee
Cop
y
h
u
man mot
i
on
7,8
Hip (vertial)
Cop
y
h
u
man mot
i
on
9,10
Hip (horizontal)
Keep balance
11 Waist
Keep
balance
12,13
Shoulder (ho
r
izontal)
Cop
y
h
u
man mot
i
on
14,15
Shoulder (vertica
l)
Cop
y
h
u
man mot
i
on
16,17
Elbow
Cop
y
h
u
man mot
i
on
2.3. Contr
o
ller
The cont
rolle
r
is
the co
re of
the control
sy
stem. A
s
shown in
Figu
re 2, the
hum
an-side
is 12 sen
s
ors as the h
u
m
an joints, t
he rob
o
t-si
de
is 17 se
rvo
s
as the ro
bot
joints, betwe
en
them is th
e
comp
uter
as the controll
er, an
d thre
e pa
rts i
s
conne
cted
by USB2dyna
mixel
device. All sensors and
servos
m
ade
by AX-12A actuators, the
s
e a
c
tuators
con
n
e
c
ted wi
th
USB2dynami
x
el device usi
ng TTL cable
in daisy ch
ain
style, and co
ntrolled by th
e comp
uter. In
our
controller, we cho
o
si
n
g
half-du
plex
commu
nication mode to
receive and
sen
d
the dat
a
betwe
en com
puter an
d out
side n
ode.
Whe
n
p
eopl
e
moving, th
e
sen
s
o
r
s p
r
od
uce
initial
sig
nals,
and
the
s
e
sig
nal
s
re
ceived
by the co
mp
uter, after p
r
oce
s
sing to t
he sen
s
or
sig
nals, the
co
m
puter
sen
d
th
e co
ntrol
sign
al
to the
servo
s
of th
e rob
o
t, then the
robot
moving
like
the
peo
ple. In the
same time, th
e
feedba
ck info
rmation
of th
e robot i
s
se
nd b
a
ck to
th
e comp
uter b
y
the
same
cable. Ba
se
d
on
the feed
ba
ck
informatio
n, compute
r
cal
c
ulates
the ce
nter of
grav
ity
of
the ro
bot whi
c
h
i
s
ca
n be
use
d
as the b
a
si
s of kee
p
in
g balan
ce of the rob
o
t [7].
Figure 2. Man-Fu
nctio
n
Robot Co
ntrol
Structu
r
e
3.
Sy
stem Identifica
tion an
d Control
The
we
ara
b
l
e
devi
c
e
s
o
n
hu
man
bo
dy and
the
servo
s
in th
e robot
both
are the
Dynamixel A
X
-12A a
c
tuat
or (Rob
otics, Co. Ltd), th
i
s
kind of a
c
tuat
or is li
ght wei
ght, small
size,
low
co
st, a
n
d
direct
editing
of the
digital
sig
nal
s. Th
ere a
r
e t
w
o typ
e
of
ope
ratio
n
mo
de fo
r th
e
steeri
ng
gea
r of the
actu
ator, o
ne i
s
wh
eel mo
de, a
n
d
the
othe
r i
s
joint mod
e
. T
he
whe
e
l mo
de
can
be used
t
o
wh
eel
-type operation rob
o
ts sin
c
e
m
o
to
rs of the
ro
b
o
ts
spin i
n
fini
tely. The joint
mode
ca
n be
use
d
to m
u
lti-joints ro
bot
since th
e
robo
ts
can
b
e
con
t
rolled with specifi
c
an
gle
s
.
For the re
ason of most of the
servos’
motion of th
e robot is se
ems like peo
ple’s joint, we
cho
o
se the jo
int mode for t
he actu
ator o
peratio
n mod
e
. Joint mod
e
has the limit
ed rotate
ran
ge
from 0 de
gre
e
to 300 de
g
r
ee a
nd ha
s
two directio
n
s
of clo
c
kwi
s
e and
counte
r
clo
c
kwi
s
e, a
s
sho
w
n in Fig
u
re 3. No
w le
t us test the actuator
in joi
n
t mode and choo
se the be
st cont
rol styl
e.
USB2dy
na
m
i
xel
TT
L
TT
L
USB
Sensor 1
Sensor 2
Sensor 12
TT
L
Hu
m
a
n
Servo 17
Servo 2
Servo 1
TT
L
R
obot
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 774
5 – 7753
7748
Figure 3. Dyn
a
mixel AX-12
A Actuator an
d its Joint Rotating Mode
3.1. Descrip
tion of the M
easur
e
d Signals from th
e Actuators
A set of output signal
s can be retri
e
ved fr
om the AX-12A actu
ators. Th
ese
signa
l
s
provide info
rmation re
garding the a
c
tual se
rv
os
angul
ar po
sit
i
on, angula
r velocity, lord,
temperature
and volta
ge.
The th
eory
b
ehind
cont
rol system
s and
how
to co
ntro
l
actu
ators an
d
other d
e
vice
s is th
e fou
ndation
of al
l mode
rn
me
cha
n
ical sy
stems. Usin
g
control sy
ste
m
mathemati
c
s and th
eory,
we
ca
n d
e
sign sy
stem
s
that do n
e
a
r
ly anything
we wa
nt, to th
e
gran
ula
r
ity that you d
e
si
re,
in the
amo
u
n
t
of ti
me that
you de
sire. Cont
rol syste
m
theory ca
n
b
e
broa
dly bro
k
e
n
up into two
major
categ
o
ries:
ope
n loo
p
control and
clo
s
ed lo
op control.
3.2. Open Lo
op Con
t
rol (OLC)
Open lo
op
co
ntrol is by fa
r the simple
r
of t
he two types of control theory. In op
en loop
control,
the
r
e
is som
e
sort of
input sign
a
l
,
wh
i
c
h the
n
passe
s th
rou
gh am
plifiers
to pro
d
u
c
e th
e
prop
er
output
, and is t
hen
passe
d out of
the syste
m
. For o
u
r
co
ntrol structu
r
e in
Figure 4, wh
en
peopl
e move
the a
c
tuato
r
(sen
so
r) on
p
eople’
s j
o
in
t
will p
r
od
uce
a
n
gula
r velo
cit
y
, the controll
er
get the d
a
ta
of spe
ed i
n
a
sam
p
le of ti
ming a
nd p
u
t it into the a
c
t
u
ator i
n
the
robot, by a d
e
l
a
y
time the robot
move.
Figure 4. Ope
n
Loop
Control Structu
r
e
Open
loo
p
co
ntrols h
a
ve n
o
feed
ba
ck a
nd
requi
r
e
th
e inp
u
t to
ret
u
rn
to
zero
b
e
fore
the
output will ret
u
rn to ze
ro. We do the ex
perim
ent
of one actuato
r
’
s
open loo
p
sp
eed co
ntrol, as
sho
w
n
in
Fig
u
re
5.
We
ca
n
find
that
servo’s
in
p
u
t sp
eed (a
ngula
r
velocity) is a con
s
tant (3
00
),
the value
of
amplifier is 1.
Althoug
h the
output
spee
d which
relati
ve to the
inp
u
t sp
eed
h
a
s
a
certai
n delay
that is can b
e
tolerated but,
in the c
han
g
e
of the spee
d dire
ction, the value of loa
d
has a relative
ly sharp b
r
e
a
k
.
Figure 5. Servo Followin
g
a Referen
c
e
Speed (OL
C
)
I
nput speed
Output speed
Contr
o
ller
Ser
vo
Sensor
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Control Struct
ure Desi
gn fo
r Man-Fu
n
c
ti
on Hum
anoi
d
Robot (Yifen
g Cui)
7749
3.3. Closed
Loop Co
ntro
l (CLC)
In closed loo
p
control, the system is ad
just
ed by itse
lf. Data does
not flow one
way; it
may pass ba
ck from a sp
ecific a
m
plifie
r (such as
po
sition feed
ba
ck in Fi
gure
6) to the sta
r
t of
the co
ntrol
sy
stem, telling i
t
to adju
s
t itself ac
co
rdin
gl
y.
For
the rea
s
on of
ou
r se
rvos have
the
feedba
ck of positio
n information, we can get
the data of servo
s
’ po
sition in
every timing
sampl
e
. Ba
sed o
n
the
s
e
data
we
ca
n let the
servo follow the
de
sire
d tim
e
varying
cu
rve
referen
c
e p
o
s
ition
whi
c
h
we g
e
t from t
he sen
s
or
by cha
ngin
g
its
actual
sp
eed
or loa
d
cha
r
g
e
through time.
Figure 6. Clo
s
e Lo
op Cont
rol Stru
cture
In Figu
re
7,
the servo
tri
e
s to
follo
w
the de
si
red
time varyin
g
sinu
soi
dal
ref
e
ren
c
e
positio
n by ch
angin
g
its lo
a
d
ch
arg
e
thro
ugh time,
a
n
d
the value of
output po
sitio
n
is ve
ry clo
s
e
to the input
positio
n as
we ca
n se
e. Compa
r
ed
with ope
n loop
control meth
od, clo
s
e
d
lo
op
control m
e
tho
d
ne
ed m
o
re
compl
e
x de
si
gn of
syst
em
architectu
re
a
nd p
r
og
ram
m
i
ng, but it
can
reali
z
e a sta
b
l
e and control
l
able statu
s
in
our Man
-
Fu
n
c
tion ro
bot co
ntrol structu
r
e
.
Figure 7. Servo Followin
g
a Referen
c
e
Speed (CL
C
)
3.4. Voltage Supply
Another
para
m
eter
with rel
e
vance to the
behav
io
r of the sy
stem is
the voltage
supplie
d
to the servo
s
. The effe
ctive voltage
ran
ge of AX-
12
A actuato
r
i
s
9-1
2
V, but
we
want
ro
b
o
t’s
motion follow the people’
s motion in a
quit simila
r d
egre
e
, or
we
can’t re
alize
the accurate
control of th
e
rob
o
t by p
e
o
p
le. Becau
s
e
open
l
oop
co
ntrol m
e
thod
have a
big
error bet
wee
n
of
input and
out
put sign
al, h
e
re
we d
o
the voltage
ex
perim
ent of o
pen loo
p
spe
ed control. O
u
r
experim
ents
sho
w
that th
e output velo
city erro
r
is p
r
opo
rtion
a
l to the voltage sup
p
lied to the
servo. In fact,
good outp
u
t velocity estim
a
tion is
a
c
hie
v
ed only if th
e power i
s
ch
arge
d aroun
d
10.5 V, as ca
n be se
en fro
m
Figure 8.
Position feedback
Servo
A
Position m
e
asur
ing
Speed control
Position co
m
p
are
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 774
5 – 7753
7750
Figure 8. Effects of the Sup
p
lied
Voltage
(9v, 10.5v, 12v) to the Se
rvos in the Out
p
uts Velo
city
Re
spo
n
s
e
4.
The Optimization of
Con
t
rol Algorith
m
Previou
s
exp
e
rime
ntal tests of the
act
uat
or
param
eters prove t
hat the
clo
s
e
d
loop
positio
n control method i
s
f
i
t for the Man
-
Fun
c
tion
rob
o
t control st
ru
cture. Ba
se
d
on the
control
stru
cture, the
co
ntrolle
r first get
s the
chang
es from
the bo
dy sen
s
or when
bo
dy form
ch
an
ge,
and then
adj
usted the
da
ta as the se
rvo co
ntro
l
comman
d
inp
u
t to the rob
o
t. Our control
system
ha
s a
more
step th
an the
traditi
onal
rob
o
t co
ntrol that i
s
t
he d
a
ta a
c
qu
isition from th
e
body se
nsor.
For the
rea
s
on of i
rre
gu
lar bo
dy
mo
vements a
n
d
the sen
s
o
r
of the AX-12
A
actuato
r
whi
c
h we
u
s
ed, t
he po
sition
d
a
ta
we
have
obtaine
d a
c
t
ually have
some e
r
ror
an
d
mutation rela
tive to
the real movement
of human,
so we ch
ose the Kalman filter algorith
m to
deal with o
b
tained fro
m
the human
bod
y. After the
data are p
r
o
c
e
s
sed by the Kalman filter, th
e
pro
c
e
s
sed
si
gnal
will be
co
me sm
ooth a
nd sta
b
le,
u
s
i
ng the
s
e p
r
o
c
essed
data a
s
control si
gn
al
of the robot’s
movement.
Kalman filter i
s
p
r
op
osed
b
y
R.E. Kalma
n
’s
fam
o
u
s
p
aper which d
e
scribi
ng
a re
cursive
solutio
n
to th
e discrete d
a
ta linea
r filteri
ng p
r
oble
m
in
1960. Sin
c
e t
hat time, due
in larg
e p
a
rt t
o
advan
ce
s in
digital compu
t
ing; the Kal
m
an filter
h
a
s
be
en the
subje
c
t of extensive
re
sea
r
ch
and
appli
c
ati
on. Kalma
n
f
ilter is a to
ol
that c
an
est
i
mate the va
riable
s
of a
wide
ra
nge
of
pro
c
e
s
ses. In
mathematica
l
terms
we would say
that a Kalman filter e
s
timates
the state
s
of a
linear system
[8].
Combi
ned
wit
h
the
clo
s
e lo
op control me
thod,
we
get t
he Fig
u
re
9,
whe
n
pe
ople’
s joint
act a
s
a vari
a
b
le of u, after the interfere
n
ce
noise w
and the m
e
a
s
ureme
n
t noi
s
e v, the out
put
value of the
sen
s
o
r
is
z, b
u
t this mea
s
u
r
eme
n
t is not
smooth
as
we ca
n se
e in
the Figu
re 10
.
Via the Kalm
an filter
we
can get the
estimate
value
x which is
sm
ooth than th
e
mea
s
ure val
u
e
z by expe
rim
ent test ba
se
d on the p
r
og
ram (i
n Ma
tla
b
) bel
ow. In
Figure 10, th
e star
dot lin
e is
the mea
s
ure
m
ent value of
z, it is
the re
al output of a
c
tuator; the
c
i
rc
le
dot line is
the es
timate
value x after Kalman filter, it is smoothl
y than z.
We use x as the
control
singl
e
of the robo
t
servo
s
, it must let the robo
t move stable
and ea
sily co
ntrol [9].
Figure 9. The
CLC Stru
ctu
r
e based on K
a
lman Filter
function p
o
siti
o
n
=
SimpleKa
l
m
an(z)
persistent A H Q R
persistent x
P
persistent firstRun
if isempt
y
(
first
R
un)
A=1;
H=1;
Q
=
0.6;
u
x
ou
t
+
+
Sensor
Interfer
e
nc
e noise
w
z
+
+
Kal
m
a
n
Filter
Meas
urem
ent nois
e
v
x
-
+
CLC
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Control Struct
ure Desi
gn fo
r Man-Fu
n
c
ti
on Hum
anoi
d
Robot (Yifen
g Cui)
7751
R=4;
x=
500;
P=6;
firstRun=1;
end
x
p
= A*x
;
% I
.
a priori estimate
Pp = A*P*A' +
Q;
% E
s
timate error covariance
K = Pp*H'*inv(H*Pp*H' + R);
% II. Compute the Kalman gain
x =
xp +
K*(z-H
*xp);
% III. Update estim
a
te
w
i
th meas
u
r
ement z
P =
Pp - K*H*Pp'
;
% IV. Update th
e err
o
r covari
ance
positi
on =
x;
Figure 10. Th
e Optimizin
g
of Control Sig
nal by Kalma
n
Filter
5.
Experiment
of Sta
t
ic Wal
k
ing
Usi
ng b
r
ain,
nervou
s
syst
em an
d mu
scle, hu
man
can
co
ntrol t
heir
own ce
nter of
gravity in the
wal
k
ing
pro
c
ess, and
mai
n
tain t
he b
o
d
y
’s stability. Although the
frame
stru
ctu
r
e
of our rob
o
t is simila
r with
human and i
t
s behavio
r is controll
ed b
y
the people, but the robo
t
doe
s not hav
e huma
n
ad
vance
d
brain
and self-con
trol functio
n
to ke
ep a g
o
od bala
n
ce i
n
motion, so
we joined th
e COM
(Ce
n
ter of mass) al
g
o
rithm de
sig
n
in the cont
rol
pro
c
e
ss.
Wh
en
peopl
e move, the chan
ge
s of 12 joints’ o
f
people will b
e
sent to the 12 joints’ of the rob
o
t by the
comp
uter, th
e robot will
move and th
e COM of
robot will ch
a
nge. At the
same time, the
comp
uter cal
c
ulate the ne
w COM an
d sent anot
h
e
r 5 joint’s con
t
rol sign
al together the 1
2
joint’s for kee
p
ing the sta
b
le of gravity center
of the robot. For exa
m
ple, whe
n
o
ne leg put up
and go forwa
r
d, the COM
sha
d
o
w
on the gro
und
wi
ll move to th
e sup
portin
g
leg. Figure 11
sho
w
e
d
the static wal
k
ing
by our co
ntrol
method [10-11].
Figure 11. Static Wal
k
in
g of MS-rob
o
t
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 774
5 – 7753
7752
Figure 12 i
s
the co
mpa
r
iso
n
of usin
g an
d not
usi
ng t
he Kalman fil
t
er in stati
c
walkin
g
test of Man-F
unctio
n
rob
o
t. Lines a
r
e se
rvos’ f
eed
ba
ck data of wal
k
ing robot’
s
thigh an
d kn
e
e
we have
con
t
rolled. Fou
r
lines p
r
e
s
ent
the left thig
h, right thigh
,
left knee and right
kne
e
sep
a
rately by
different col
o
rs. Th
e first
picture
is the
normal
control results wit
hout sol
u
tion
of
control sig
nal
; the seco
nd
picture is the
feedba
ck
re
sults afte
r op
timizing of co
ntrol sig
nal b
y
Kalman filter.
Obviously, we can find that the app
lication of certain
filter
ing algori
thm will make
the movem
e
nt of ea
ch
jo
int of the
ro
bot be
com
e
s flat and
sm
ooth, an
d
co
ndu
cive to th
e
stability of the robot du
ring
wal
k
ing to ma
intain and a
c
curacy imp
r
o
v
ed [12].
a) Static wal
k
ing witho
u
t Kalman
b) Static wal
k
ing usi
ng Kal
m
an
Figure 12. Th
e Comp
ari
s
o
n
of Robot Po
sition Fee
dba
ck fo
r usi
ng o
r
Not u
s
ing K
a
lman Filter i
n
Static Wal
k
in
g Control Experime
n
t
6. Conclu
sion
A ne
w robot
control
stru
ct
ure–
Man
-
Fu
n
c
tion
huma
n
o
id
robot
wa
s e
s
tabli
s
h
e
d
in o
u
r
pape
r. This ki
nd of stru
cture reali
z
e
s
a true se
nse of the huma
noid
movement
s’ simulatio
n
tha
n
traditional ro
bot
co
ntrol st
ructu
r
e.
T
he contro
l of M
an-F
u
n
c
tion
robot ne
ed a
s
cl
ose a
s
t
he
action
state of the huma
n
and it can
keep the
st
able by itself
in motion. To this end,
we
con
d
u
c
ted th
e analysi
s
of
the control mode, the
cl
ose
d
loop p
o
s
ition control is sel
e
cte
d
, and
the Kalman filter algo
rithm to ensu
r
e a stable and
effi
cient ro
bot se
rvo cont
rol. Throu
gh si
mpl
e
static
wal
k
ing
test, it prove
d
that
we
can
reali
z
e
the
b
a
si
c
control o
f
our
Man
-
Fu
nction
ro
bot.
Of
cou
r
se, this kin
d
of
co
ntrol
stru
ctures
nee
d mo
re intelli
gent
and
sophi
sticated
co
ntrol
algorith
m
s to
achieve the
combi
nation
of human
co
ntrol an
d rob
o
tics
cont
rol
to let the rob
o
t
forward a
step better. This
will
be the next step in our
research.
Ackn
o
w
l
e
dg
ement
This
wo
rk was
sup
p
o
r
ted
by the G
R
RC Pro
g
ram o
f
Gyeonggi P
r
ovince
Repu
blic o
f
Korea, [(GRRC SUWOM
2012-B5), Situational
technology and System
development.] and A
Proje
c
t of Shando
ng Provi
n
ce
High
er E
ducational S
c
ien
c
e a
nd T
e
ch
nolo
g
y Progra
m
, Chin
a
(
J
13
LN
8
4
)
.
Referen
ces
[1]
ASIMO by Honda
. 201
3, http://
w
o
rl
d.ho
nd
a.com/ASIMO/
[2]
Patrick Robl
er,
U
w
e
D Han
e
beck.
T
e
lepr
es
ence tech
ni
qu
es for excepti
on ha
ndl
in
g in
house
hol
d
robots, Syste
m
s, Man
an
d Cy
bern
e
tics
. IEE
E
Internati
o
n
a
l
Confer
ence
o
n
Date
of
Co
nfer
ence.
20
04;
1: 53-58.
[3] Cai
Zix
i
ng.
Ro
botics
. T
s
inghu
a Univ
ersit
y
Pr
ess. Beiji
ng, C
h
in
a. 200
9.
[4]
Green J, Ch
un
g M, Cha
ng
L,
Scherert R, S
m
ith J, Rao
R
P
N.
An ad
apti
v
e bra
i
n-co
mp
uter interfac
e
for hu
man
o
i
d
robot contro
l
. 1
1
th IEEE-RAS Internatio
na
l C
onfere
n
ce o
n
, 201
1; 199-
204.
[5]
Ismoilovich EP, Min D.
Be
hav
i
o
ral sy
nchro
n
i
z
ation
of h
u
man
and
hu
man
i
od
robot
, U
b
iq
uito
us Ro
bot
s
and Amb
i
e
n
t Intellig
enc
e (UR
A
I). 8th Internat
iona
l Conf
eren
ce on. 20
11; 6
55-6
60.
[6]
Pedro T
eod
or
o, Morio M
a
r
ques, Jor
ge
Martins,
Car
l
o
s
Card
eira,
Migu
el A
y
a
l
a-
Botto, Limor
Sch
w
eitzer, Jose Sa Da Costa.
Hu
ma
noi
d R
obot C
ontro
l Ar
chitecture
. RO
BOT
I
CA - 7th Confer
enc
e
on Mob
ile R
o
b
o
ts and Com
p
e
t
itions Ce
ntro Pa
roq
u
ia
l de P
ader
ne, Alb
u
fei
r
a, Portugal. 2
007.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Control Struct
ure Desi
gn fo
r Man-Fu
n
c
ti
on Hum
anoi
d
Robot (Yifen
g Cui)
7753
[7] Nishika
w
a
Masao.
Overview
of robotics
. K
y
or
itsu Shup
pa
n Co., Ltd. 2007.
[8]
Xi
n Jin
g
, Liu
Ding, Ba
n Jia-
an.
App
licati
o
n
Researc
h
of Adaptiv
e
Kal
m
an F
ilter in
Ro
bot Contr
o
l
.
Journ
a
l of Xi’a
n Univ
ersit
y
of
T
e
chnolog
y. 2
007; 23(
2): 136
-139.
[9]
Cho
o
mua
n
g
R,
Afzulp
urkar
N. H
y
bri
d
K
a
lma
n
filt
er/fuzz
y l
o
g
i
c b
a
sed
p
o
siti
on c
ontrol
of
a
u
tonom
ou
s
mobile robot.
Internati
o
n
a
l Jo
urna
l of Advan
c
ed Ro
botic Sy
stems
. 20
05; 2
(
3): 197-2
08.
[10] F
ang
S
h
e
n
w
e
i.
T
he Dy
na
mic
An a
l
ysis
of the
Rob
o
t Bas
e
d
on A
D
AMS Virt
ual
Co
nsol
e.
M. S. T
hesis.
Beih
ang U
n
iv
e
r
sit
y
. 20
09.
[11]
Cui Y
i
fen
g
, Sh
on S
u
Goo
g
,
B
y
u
n
H
ee J
u
n
g
. T
he Kinem
a
t
ics Rese
arch
and
Co
ntrol
of Quadr
upe
d
Rob
o
t Based
on Scil
ab.
Inte
rnatio
nal J
our
n
a
l of Di
gita
l C
ont
ent T
e
ch
no
l
ogy a
nd its A
pplic
atio
ns
.
201
2; 6(19): 11
3-12
0.
[12]
Ning S
huro
ng, Guo Xi
ju
a.
T
he Applic
ation of
Matlab i
n
Simu
latio
n
of the Paralle
l Mech
anis
m
.
Journa
l
of System Si
mulati
on
. 20
04; 16(1
0
): 227
3-2
275.
Evaluation Warning : The document was created with Spire.PDF for Python.