Int
ern
at
i
onal
Journ
al of
R
obot
ic
s
and
Autom
ati
on (I
JR
A)
Vo
l.
7
, No
.
3
,
Septem
ber
201
8
, p
p.
185
~
196
IS
S
N:
20
89
-
4856,
DOI: 10
.11
591/
i
jra
.
v
7
i
3
.
pp18
5
-
196
185
Journ
al h
om
e
page
:
http:
//
ia
escore.c
om/j
ourn
als/i
ndex.
ph
p/IJRA/i
ndex
Linea
r a
nd
No
n
-
l
inear Co
ntrol D
es
ign
of
S
kid
Steer Mobil
e
Robot
on
a
n Em
bedd
ed
Board
Jharn
a Maju
mda
r
,
Su
dip
C
G
upta
,
B P
rass
an
n
a
P
ras
at
h
Cent
re
for
Robo
t
ic
s Re
se
arc
h
,
Ni
t
te
Me
ena
kshi
In
stit
ute of Te
chno
log
y
,
P.
B.
No.
6
429
,
Yel
aha
nk
a,
Banga
lor
e, I
ndia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Feb
12, 201
8
Re
vised
Ju
l
18
,
201
8
Accepte
d Aug
13
, 201
8
A
det
ailed
app
roa
ch
for
a
l
in
ea
r
Proportion
a
l
-
Inte
gr
al
-
Der
ivative
(PID
)
cont
roller
and
a
non
-
l
ine
ar
controlle
r
-
Li
ne
ar
Quadra
t
ic
Regu
la
to
r
(LQR)
i
s
discussed
in
this
pape
r
.
B
y
an
aly
z
ing
sev
era
l
m
at
hemat
ic
a
l
designs
fo
r
the
Skid
St
ee
r
Mobile
Robot
(
SS
MR),
the
con
trol
lers
ar
e
impl
emente
d
in
an
embedde
d
m
ic
roc
ontroller
-
Mbed
LPC1768.
To
v
eri
f
y
the
cont
rol
le
rs,
M
atlab
-
Sim
uli
n
k
is
used
for
the
sim
ulatio
n
of
b
oth
the
controll
ers
invol
ving
m
otor
s
-
Maxon
RE40.
Thi
s
pape
r
compare
s
be
twee
n
PI
D
and
LQR
cont
roller
al
ong
with
th
e
p
erf
orm
anc
e
compari
son
bet
we
en
Hom
ogenous
and
Non
-
Hom
ogeno
us L
QR c
ontro
llers.
Ke
yw
or
d:
LQR
Ma
xon
RE
40
Mbed
LPC
17
68
PI
D
SSMR
Copyrigh
t
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Jh
ar
na
Ma
jum
dar
,
Ce
ntre fo
r
R
obotics R
esearc
h
,
Nitt
e Me
enak
s
hi Insti
tute
of
Tech
no
l
og
y
,
P.B.No.
6429,
Yelaha
nk
a
, Ba
ng
al
or
e
, In
dia
.
Em
a
il
:
j
harna.
m
aju
m
dar
@
n
m
it
.ac.in
1
, s
idi
sh
ere
.s
ud
i
p@g
m
ai
l.co
m
2
, p
ra
ssan
na4r
obo@gm
ail.co
m
3
1.
INTROD
U
CTION
The
i
nterest
i
n
the
dom
ai
n
of
m
ob
il
e
r
obot
s
an
d
it
s
a
utom
at
ion
has
gr
own
i
n
t
he
pa
st
few
ye
ars
.
Ther
e
are
l
ots
of
resea
rc
h
on
the
kin
em
at
ic
s
and
dynam
ic
m
od
el
li
ng
,
a
nd
f
or
t
he
co
ntr
ol
syst
e
m
fo
r
the
m
ob
il
iz
at
ion
of
the
r
obot,
w
hich
sho
uld
sat
isfy
it
s
pe
r
form
ance
base
d
on
ti
m
e
and
accu
racy.
Althou
gh
sever
al
researc
h
we
re con
du
ct
ed
on the c
ontr
ol syst
e
m
[1]
, th
e im
ple
m
entat
ion
on a
n
em
bed
de
d bo
a
r
d
ha
s n
ot
been
m
entione
d
i
nvolv
i
ng
m
em
or
y
m
anag
e
m
ent
an
d
lo
w
redu
nd
a
ncy.
T
her
e
wer
e
se
ve
ral
r
esearc
h
rel
at
ed
t
o
var
i
ou
s
ap
proa
ches
t
o
ac
hiev
e
the
op
ti
m
u
m
co
ntr
ol
syst
em
and
their
re
sp
ect
ive
sim
ulati
on
resu
lt
s
pr
ov
e
d
t
o
be op
ti
m
u
m
[1
-
6
]
, b
ut whe
n u
sed
i
n
a syst
em
or a
rob
ot,
the
ou
t
pu
t
de
te
rm
i
nes
al
l t
hese
m
od
el
s
to be s
pa
rse.
Sk
id
Stee
r
Mo
bile
Ro
bo
ts
(
S
SMR
)
are
the
veh
ic
le
s
t
hat
c
an
tract
on
al
l
te
rr
ai
n
co
ndit
ion.
T
o
m
ov
e
the
SSMR
on
al
l
te
rr
ai
n
ther
e
m
us
t
be
a
f
eedb
ac
k
deter
m
ining
the
e
rror
s
i
n
it
s
pr
e
di
ct
ed
par
am
et
e
rs
an
d
the
outp
ut
pa
r
a
m
et
ers,
w
hich
m
us
t
be
do
ne
a
utono
m
ously
by
the
em
bedde
d
c
on
tr
ol
le
r,
w
hich
prov
i
de
s
the
act
uation
s
ign
al
s
to
al
l
th
e
wh
eel
s
in
te
r
m
s
of
pulse
-
width
m
od
ulati
on
(PWM
)
.
Thi
s
can
be
ac
hie
ved
by
i
m
ple
m
enting
the tw
o
c
ontr
ol
al
gorithm
s (
PID a
nd L
QR)
i
n t
he
em
bed
de
d
con
t
ro
ll
er M
be
d
L
PC1
768.
2.
SSMR
MO
DE
LING
The
S
kid
Stee
rin
g
Mo
bile
Robot
is
a
f
our
-
wh
eel
e
d
hi
gh
tract
ion
m
ob
il
e
rob
ot,
w
hich
are
ste
ere
d
by
dif
fe
ren
ti
al
dr
i
ve.
A
str
ong
co
ntr
oller
is
r
equ
i
red
to
c
ompu
te
al
l
the
val
ues
of
the
m
obil
e
robo
t
a
nd
he
nce
Mbed
L
PC1
76
8
is
prefe
rr
e
d
t
akin
g
co
st
eff
e
ct
iveness
a
nd
s
peed
of
operati
on
i
nto
c
onsid
erati
on.
T
he
S
SMR
Mod
el
is
obser
ved th
rou
gh 3D CA
D
C
AT
I
A
is s
how
n
i
n Fi
g
ure
1.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
7
,
No.
3
,
Septem
ber 2
01
8
:
185
–
196
186
Fig
ure
1
.
CAT
IA
m
od
el
of
S
SMR
A
m
at
he
m
at
ic
a
l
desc
riptio
n
of
the
dynam
ic
s
of
a
n
SSMR
m
ov
ing
on
a
pl
anar
surface
i
s
intr
oduce
d
in
this
sect
io
n.
T
he
m
at
he
m
atical
m
od
el
of
t
he
ve
hicle
ca
n
be
div
i
de
d
i
nto
th
ree
pa
rts:
ki
nem
atics,
dy
na
m
ic
s
and
dr
i
ve
s
ubs
yst
e
m
s.
Her
e
,
we
f
ocu
s
on
th
e
fir
st
tw
o
bl
oc
ks
,
i.e.
,
t
he
dr
ive
a
nd
dynam
ic
s
subsyst
em
s
,
a
nd
we
us
e
them
fo
r
ref
e
ren
ce
t
r
ackin
g
co
ntr
ol
of
both
the
li
ne
ar
an
d
a
ngula
r
vel
ociti
es.
SS
MR
m
od
el
as
sh
ow
n
in Figu
re
2.
Fig
ure
2
.
SSM
R M
od
el
2.1.
D
ynamic
m
od
el
li
ng
This
proces
s
a
ll
ow
ed
a
th
oro
ugh
stu
dy
for
the
desi
gn
a
nd
the
loa
d
car
ryi
ng
ca
pacit
y
of
the
SSM
R
and
the
optim
i
zat
ion
in
it
s
w
ei
gh
t,
w
hich
is
ver
y
be
nef
ic
ia
l
for
it
s
co
ntr
ol.
I
n
Fi
gure
3
is
schem
at
ic
m
o
del
of
SSMR
in
glob
al
co
ordinate
s
yst
e
m
.
The
m
a
in
e
quat
ion
tha
t
desc
ribes
the
dy
nam
ic
su
bsy
stem
of
the
S
SMR
m
ov
ing
on a
planar su
rf
ace
as
shown i
n fig
ure is gi
ven b
y
(
1)
:
M (
q)
̇
η
+
C
(̇
q)
η +
R
(̇
q) =
B (q)
τ
(1)
Fig
ure
3
.
Sc
he
m
at
ic
Mod
el
of S
SMR
i
n Glo
bal Co
ordinate
Syste
m
2.2.
Dri
ve
m
odel
In
Fi
gure
4
is
dr
i
ve
m
od
el
.
F
our
DC
m
oto
r
s
c
oupled
with
m
echan
ic
al
ge
ars
finall
y
t
o
the
w
heels
dr
i
ve
the
SSM
R.
Con
si
der
i
ng
on
ly
one
of
the
dr
i
ves
(M
otor)
a
nd
ass
um
ing
rem
ai
nin
g
al
l
m
oto
rs
to
ha
ve
sam
e p
ara
m
et
e
rs,
t
he rel
at
ion
betwee
n
to
r
qu
e
τ an
d v
oltage
u
va
can
be
writ
te
n
as
sh
i
wn in (2
-
3)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
4856
L
inea
r
a
nd N
on
-
Li
ne
ar
C
on
tr
ol D
e
sig
n
o
f Sk
id S
te
er
Mobil
e Rob
ot
o
n
… (
Jharn
a
M
ajum
da
r)
187
Fig
ure
4
.
D
rive
Model
τ = n
K
i
i
a
(2)
u
va
=
L
a
d
dt
(
i
a
)
+
R
a
i
a
+
n
K
e
wi
(3)
Wh
e
re,
i
a
is t
he
arm
at
ur
e
curre
nt,
K
i
is t
he
m
oto
r
t
orqu
e
consta
nt,
N
is t
he gear
ra
ti
o
(
n
>
1),
L
a
and
R
a
de
note
the
series in
duct
an
ce an
d resist
an
ce o
f
the
roto
rs
r
es
pecti
vely
,
K
e
is t
he
el
ect
r
omoti
ve fo
rce c
oe
ff
ic
ie
nt, a
nd
W
=
[
w
l
w
r
]
T.
The
le
ft
w
l
an
d
r
igh
t
w
r
si
des
a
ng
ular
vel
ociti
es
can
be
obta
ine
d
from
the
fo
l
lowing
f
or
m
ulas
as
sh
ow
n
in
(4
-
5)
w
l
=
vx
−
pr
e
sent
e
rr
or
r
;
(4)
w
r
=
vx
−
pr
e
sent
e
rr
or
r
;
(5)
The
Ma
tl
ab
m
od
el
li
ng
of the
dr
i
ve
is s
how
n belo
w
in
the
Figure
5
Fig
ure
.
5 D
rive
m
od
el
in
Ma
tl
ab
3.
CONTR
OLL
ER A
L
GO
RI
THM
3.1.
PI
D
c
ont
rolle
r
PI
D
is
t
he
basi
c
li
near
co
ntr
ol
al
gorithm
us
e
d
i
n
a
syst
em
,
wh
ic
h
i
nvolv
e
d
a
co
ns
ta
nt
fe
edb
ac
k
gai
n
fed
to
the
in
put. A
PID
c
ontr
ol
le
r
dete
rm
ines
er
ror
val
ues
a
s
a d
if
fer
e
nce
be
tween
re
quire
d
value
a
nd
m
e
asur
e
process v
a
riabl
es.
P
ID
in
vo
l
ve
s
three
pa
ram
et
ers
i.e. Pro
portion
al
(
P
),
I
nt
egr
al
(
I
)
a
nd
D
eriva
ti
ve
(D) wher
e
P
is
acco
un
ta
ble
for
pr
e
sent
val
ues,
I
f
or
past
values
an
d
D
f
or
f
uture
value
s.
Bl
oc
k
diag
ra
m
of
PID
co
nt
ro
ll
er
as sho
wn in Fi
gure
6.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
7
,
No.
3
,
Septem
ber 2
01
8
:
185
–
196
188
Fig
ure
6
.
Bl
oc
k
d
ia
gram
o
f
P
ID
c
ontr
oller
The basic
gove
rn
i
ng equati
on
of PID c
ontr
oller is
giv
e
n by
(
6)
:
u (t)=
K
p
e
(
t
)
+
K
i
∫
e
(
t
)
dt
+
K
p
de
dt
(6)
The
val
ues
of
P
,
I,
D
we
re
m
anu
al
ly
tun
e
d
to
obta
in
t
he
best
re
sp
onse
with
t
he
help
of
m
at
la
bs
i
m
ulati
on,
wh
ic
h
was
then
fe
d
i
nto
t
he
act
ual
c
ontr
oller
desig
n
for
the
SSMR
as
C
pro
gr
am
.
Ma
tl
ab
si
m
ula
ti
on
m
o
del
of
the
c
ontr
oller
desi
gn
an
d
it
s
r
es
pons
e can
be
obse
rv
e
d
in
t
he
F
ig
ur
e
7
an
d
8
belo
w
wh
e
re
PI
D
contr
oller
is at
ta
ched
t
o d
rive
m
od
el
, w
hi
ch
is
der
ive
d e
arli
er.
Figure
7.
Flo
w
D
ia
gram
of
P
I
D
Co
ntr
oller
Fig
ure
8.
Re
spon
s
e
of
PID C
on
t
ro
l
le
r
3.1.1
.
I
mple
m
e
ntation
of LQR on
N
X
P L
PC 1
768
m
ic
r
o
co
nt
r
oller
LPC 1
768 co
nt
ro
ll
er
has
a
32
-
bit ARM
C
or
te
x
-
M
3
c
or
e
runnin
g
at
96
M
Hz
. I
t al
s
o has
51
2K
B
flas
h,
32KB R
AM and lots
of
I/O pi
ns
, w
hich
are r
equ
i
red f
or the
op
e
rati
on
of
S
SMR
.
Pin
d
ia
gram
o
f
Mb
e
d
LPC1
768
a
s s
how
n
i
n
Fi
gure
9.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
4856
L
inea
r
a
nd N
on
-
Li
ne
ar
C
on
tr
ol D
e
sig
n
o
f Sk
id S
te
er
Mobil
e Rob
ot
o
n
… (
Jharn
a
M
ajum
da
r)
189
Figure
9.
Pin
Diag
ram
o
f
M
bed LPC
1768
Algori
th
m
St
ep
1
In
it
ia
li
ze the
pin
s
of th
e m
ic
ro
co
ntr
oller.
St
ep
2
De
fine param
et
ers
li
ke
gain
s, K,
P, I,
D
St
ep
3
I
nput th
e d
esi
re
d spee
d
St
ep
4
-
Ca
lc
ul
at
e the c
ount
pe
r
sec
ond
“t
”
f
or the
desire
d
s
peed in
one
revoluti
on
St
ep
5
-
Assign
interru
pts fo
r
t
he
c
ounts
St
ep
6
-
Re
cei
ve
the
feedbac
k from
the en
co
de
r
f
or ti
m
e “t”
St
ep
7
-
C
om
par
e the
d
esi
red s
peed an
d fee
dback
for
ti
m
e “t
”
St
ep
8
-
Ca
lc
ul
at
e the err
or
usi
ng
t
he
P
I
D ge
ner
al
form
ula
St
ep
9
-
c
om
par
e the
er
ror wit
h
the
set
gain
St
ep
10
-
Ge
ne
rate P
W
M
usi
ng
resp
ect
ive
ga
ins a
nd er
rors
a
s b
ase
P
WM
St
ep
11
-
Ru
n
t
he
e
ntire ste
p
i
n
lo
op.
3.2.
LQ
R
c
ont
rolle
r
Althou
gh
the
PI
D
co
ntr
oller
wor
ked
on
t
he
SSMR
a
nd
showe
d
a
n
a
ccepta
ble
m
obil
ity
,
it
was
flawe
d.
The
i
dea
of
li
nea
r
feedbac
k
pro
ve
d
t
o
m
ake
th
e
SSMR
una
bl
e
to
m
ov
e
unde
r
va
ryi
ng
t
err
ai
n
conditi
ons.
T
o
ove
rco
m
e
this
pr
ob
le
m
,
LQ
R
co
ntr
oller
w
as
im
ple
m
ente
d
on
t
he
SSM
R.
L
QR
is
a
n
op
ti
m
al
con
t
ro
ll
er
as
it
pro
vid
es
th
e
sm
al
le
s
t
poss
ible
er
ror
in
it
s
ou
t
pu
t
w
he
n
com
par
e
d
t
o
the
in
put.
The
la
te
r
con
t
ro
ll
er
desi
gn
wa
s
op
ti
m
a
ll
y
in
pa
per
[1]
.
By
ta
king
al
l
the
par
am
et
ers
nee
de
d
for
fe
edb
ac
k
m
od
el
,
dri
ve
m
od
el
and
dy
nam
ic
m
od
el
as
de
rive
d
in
pa
per
[1
]
[
2]
f
urt
her
a
naly
sis
wer
e
im
bu
e
d
i
n
the
Mbe
d
L
PC176
8
m
ic
ro
co
ntro
ll
e
r.
T
he bloc
k di
agr
am
o
f LQR
con
t
ro
ll
er
as s
how
n
i
n
Fi
gure
10.
Fig
ure
10.
Bl
oc
k Diag
ram
o
f LQR C
on
tr
oller
3.2.1
.
D
ynami
c
m
odel
The dynam
ic
m
od
el
co
ns
ist
s
of the
stat
e spa
ce eval
uation
of M, C, R
, a
nd B a
re
form
ul
at
ed
(7
-
10)
̅
=
[
0
0
2
]
(
7)
̅
=
[
0
̇
−
̇
̇
]
(8)
̅
=
[
(
̇
)
(
̇
)
+
]
(9)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
7
,
No.
3
,
Septem
ber 2
01
8
:
185
–
196
190
̅
=
[
1
1
−
]
(10)
Her
e
,
m
is
the
m
ass
of
th
e
body,
is
t
he
or
ie
ntati
on
of
th
e
r
obot
with
r
espect
t
o
t
he
global
c
oor
dina
te
s.
Using all
these
v
al
ue
s, we s
ub
sti
tute t
hem
in
the stat
e eq
uation
(
11)
.
̇
=
̅
−
1
(
)
̅
(
)
−
̅
−
1
(
)
̅
−
̅
−
1
(
)
̅
(
̇
)
(11)
Wh
e
re,
M
is
the
i
ner
ti
al
f
or
c
e
bei
ng
e
xp
e
r
ie
nced
by
the
body
of
the
r
obot,
B
is
the
input
tr
ans
form
at
ion
m
at
rix,
is t
he
s
ta
te
m
at
rix
for a
ny insta
nce a
nd
is t
he
torq
ue
on th
e
syst
em
.
3.2.2
.
Ov
er
all
state s
pa
ce
m
od
el
Althou
gh
t
he
dy
nam
ic
an
d
dri
ve
m
od
el
are
re
al
iz
ed,
t
he
ent
ire
syst
em
cal
culat
ion hav
e
t
o be don
e
b
y
involvin
g
bo
t
h
the
dynam
ic
and
dr
i
ve
e
quat
ion
s
.
This
idea
was
well
e
xpla
ined
in
the
paper
[1
]
,
wh
il
e
tr
yi
ng
to d
e
vice a
n o
pt
i
m
u
m
stat
e sp
ace f
or
sim
ulatio
n
a
s s
how
n
i
n (12)
.
[
̇
̇
1
̇
2
̇
]
=
1
[
1
2
]
+
1
[
1
2
]
+
(12)
Her
e
,
A
1
is
a
4x4
m
at
rix
representin
g
the
Wh
eel
L
oa
d
Con
sta
nts,
B
1
is
a
4x2
m
at
rix
represe
nting
V
oltag
e
Conver
sio
n
C
on
sta
nt
a
nd
D
is
a
dist
urba
nc
e,
wh
ic
h
is
of
th
e
orde
r
4x1.
T
he
c
on
sta
nt
s
are
in
the
f
or
m
of
a stat
e sp
ace
, whic
h keep
cha
ng
i
ng w
it
h re
sp
ect
to
x
ICR
as
shown i
n (13
-
15)
.
1
=
[
0
−
2
+
0
−
(
2
+
)
(
2
+
)
−
−
−
−
0
0
−
]
(13)
1
=
[
0
0
0
0
1
0
0
1
]
(14)
=
[
−
−
−
2
+
0
0
]
(15)
These 4xN
m
atr
ic
es
are r
ed
uc
ed
to 2xN
m
at
rix
by
ass
um
ing
to
be
z
er
o
sin
ce
it
do
es not
m
ake
any
cha
nge
s
for
the
ar
gum
e
nts
as s
how
n
i
n (
16
-
20)
.
=
+
(16)
(
̇
̇
)
=
(
)
+
(
1
2
)
+
−
(17)
=
[
2
2
2
−
2
+
−
2
2
2
2
(
2
+
)
]
(18)
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
4856
L
inea
r
a
nd N
on
-
Li
ne
ar
C
on
tr
ol D
e
sig
n
o
f Sk
id S
te
er
Mobil
e Rob
ot
o
n
… (
Jharn
a
M
ajum
da
r)
191
=
[
−
(
2
+
)
(
2
+
)
]
(19)
=
[
−
−
−
2
+
]
(20)
Gen
e
ral f
orm
o
f
the
stat
e sp
ac
e re
pr
ese
ntati
on is as
sho
wn i
n (21)
:
X=
A(X) X
+
BU +
D(X)
(21)
Wh
e
re t
he
syst
e
m
stat
es are repr
ese
nted
b
y
(
22)
:
X
=
[vx
w] T
(22)
The
c
ontrol i
nput
vecto
r U is
r
e
pr
ese
nted
b
y
(
23)
:
U
=
[V1 V
2] T
(23)
D(X) i
s consi
de
red as a
distu
r
ban
ce
.
3.
2.3
.
Desi
gn
of LQ
R
c
ontr
ol
le
r
The
LQR
co
nt
ro
ll
er
ca
n
be
desi
gn
e
d
f
or
a
syst
em
in
two
dif
fer
e
nt
ways.
H
om
og
enous
L
Q
R
Con
tr
oller:
T
hi
s
ty
pe
of
LQ
R
co
ntr
ol
does
not
ta
ke
distu
rb
a
nce
i
n
fact
or.
T
his
is
hel
pful
as
it
gi
ve
s
faste
r
respo
ns
e
wh
e
n
the
syst
e
m
is
des
ig
ne
d
to
m
ov
e
e
ve
n
te
r
ra
in.
N
on
H
om
og
en
ous
L
QR
c
on
t
ro
ll
er:
T
his
LQR
con
t
ro
ll
er
c
ons
iders
distu
rb
a
nc
es
an
d
it
is
a
appr
oach
t
hat
is
m
or
e
pr
act
ic
al
.
It
hel
ps
the
SSMR
or
a
s
yst
e
m
m
ov
e in une
ve
n
te
r
rains
.
I
n F
igure
11 is d
e
sign o
f
c
ontrolle
r.
Figure
1
1
.
Des
ign
of
c
on
tr
oll
e
r
3.2.4
.
Hom
oge
no
us
LQ
R
des
i
gn
Con
si
der the st
at
e sp
ace
re
pr
e
sentat
ion o
f
a
s
yst
e
m
(24
-
25):
x
˙
=
A
x
+
Bu,
(24)
y
=
Cx
(25)
W
it
h
x
(
t
)
∈
R
n,
U
(
t
)
∈
R
m
a
nd
the init
ia
l co
ndit
ion
is
x
(
0).
Assum
ing
that
al
l t
he
sta
te
s are
m
easur
able a
nd see
k
to
f
in
d a st
at
e
-
va
riabl
e fee
db
ac
k
c
on
trol law
as
s
ho
wn in
(26)
:
u
=
−
Kx
(26)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
7
,
No.
3
,
Septem
ber 2
01
8
:
185
–
196
192
Her
e
, K is the
fe
edb
ac
k gai
n v
ect
or
.
Th
e
qua
dr
at
ic
c
os
t f
un
ct
ion
def
i
ned by
(27)
:
J=
(
1
2
)
∫
(
x
T
Qx
∞
0
+
u
T
Ru
) dt
(27)
The o
pti
m
al
f
eedb
ac
k g
ai
n v
e
ct
or
ca
n be cal
culat
ed by (
28)
:
K
=
R
−
1
B
T
P
(28)
Wh
e
re
P
is t
he
so
l
ution o
f
the
A
lge
brai
c Ri
ccat
i Equ
at
io
n (
ARE)
d
e
fine
d by
(2
9)
:
P
+
P
A
+
Q
–
PB
R
−
1
B
T
P
= 0
(29)
3.2.5
.
N
on
hom
ogen
ou
s L
Q
R
desi
gn
The
cl
os
ed
lo
op syst
em
eq
uation o
f
the
contr
oller is
̇
=
A
x
+
Bu.
̇
=(A
-
B
K)x
.
T
he
f
unct
io
n i
s g
ive
n
by
(30
-
34):
V(x)
=
x
T
Px
(30)
V
̇
=
x
̇
T
Px
+
x
T
P
x
̇
̇
(31)
V
̇
=
[
(
A
−
BK
)
x
]
Px
+
[
(
A
−
BK
)
x
]
(32)
V
̇
=
x
T
[
(
A
T
P
+
P
A
+
Q
–
PBR
−
1
B
T
P
)
-
Q
–
PBR
−
1
B
T
P
]x
(33)
V
̇
=
x
T
[
-
Q
–
PBR
−
1
B
T
P
]x
(34)
Fr
om
̇
it
is cle
ar th
at
R >
0, P
>0
a
nd h
e
nce
PBR
−
1
B
T
P
>0.
Als
o
Q
>0.
He
nce
(Q
+
PBR
−
1
B
T
P
) >
0.
Ther
e
f
or
e
̇
< 0.
Hen
ce
the cl
ose
d
lo
op syst
em
is stable.
C
ontrol
So
l
ution
(35
-
37)
:
u
=
-
R
−
1
B
T
λ
(35)
u
=
-
R
−
1
B
T
(P
x+
K)
(36
)
u=
-
R
−
1
B
T
Px
-
R
−
1
B
T
K
(37)
It
is
obse
rv
e
d
from
the
e
qu
at
ion
that
eve
n
a
fter
t
he
prese
nt
sta
te
x
te
nd
s
to
ze
ro
the
re
is
a
resi
du
al
con
t
ro
ll
er
act
in
g wh
ic
h
te
nds
to red
uce t
he dist
urban
ce
.
3.2.6
.
I
mple
m
ent
at
io
n
of LQR on
N
X
P L
PC 1
768
mi
cr
ocontr
oller
LPC
1768
co
nt
ro
ll
er
has
a
32
-
bit
ARM
Corte
x
-
M
3
co
re
r
unning
at
96MHz
.
It
al
so
ha
s
51
2K
B
flash,
32KB R
AM and lot
s
of
I/O pi
ns
, w
hich
are r
equ
i
red f
or the
op
e
rati
on
of
S
SMR
.
Algori
th
m
St
ep
1
In
it
ia
li
ze the
pin
s
of th
e m
ic
ro
co
ntr
oller.
St
ep
2
De
fine param
et
ers
li
ke
gain
s, A,
B, C
, D,
K
St
ep
3
I
nput th
e d
esi
re
d spee
d
St
ep
4
Ca
lc
ulate
the cou
nt
pe
r
sec
ond
“t
”
f
or the
d
es
i
red spe
ed
St
ep
5
Assi
gn
interr
up
ts
for t
he
c
ounts
St
ep
6
Re
cei
ve
the
feedbac
k f
ro
m
the en
c
od
er fo
r
ti
m
e “t”
St
ep
7
C
om
pare t
he desire
d
s
peed an
d fee
dback
for
ti
m
e “t
”
St
ep
8
Ca
lc
ulate
the g
ai
n
m
atr
ic
es A, B,
C,
D,
K, Q
, R
St
ep
9
Ca
lc
ulate
the contr
ol
unit
u= (f+
v)
-
Kx
St
ep
10
Ge
nerat
e P
W
M
us
in
g resp
ect
ive
ga
ins a
nd er
rors
a
s b
ase
P
WM
St
ep
11
R
un th
e entire ste
p
i
n l
oop.
*
N
ote
-
D
is
ta
ken
as
z
ero in
case
of h
om
ogenou
s
LQ
R
c
ontr
oller
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
4856
L
inea
r
a
nd N
on
-
Li
ne
ar
C
on
tr
ol D
e
sig
n
o
f Sk
id S
te
er
Mobil
e Rob
ot
o
n
… (
Jharn
a
M
ajum
da
r)
193
4.
RESU
LT
S
Desig
n
Re
s
ults
of
t
he
S
kid
St
eer
Mo
bile
Ro
bo
t.
Table
1
s
hows
t
he
fin
al
Ele
ct
rical
and
Me
chan
ic
al
par
am
et
ers
of
the
S
SMR
r
obot.
Detai
l
Ca
lc
ulat
ion
of
SSMR
is
give
n
in
AP
PE
N
DIX
–
1.
Ta
ble
2
show
s
the f
i
nal r
es
ults o
f
the
p
a
ram
e
te
rs
of
SSMR
.
Table
1
.
Ele
ct
r
ic
al
an
d
Me
c
ha
nical
Par
am
et
e
rs of
t
he
R
obot
Para
m
e
ters
Va
lues (Di
m
e
nsio
n)
R (r
ad
iu
s o
f
wheel
)
0
.10
7
5
(
m
)
M
(
Mass o
f
Ro
b
o
t
)
4
6
(
k
g
)
I
(
Mo
m
en
t of
I
n
ert
ia abo
u
t COM)
0
.00
6
2
8
8
(
Kg
m
2
)
R
a
(Ar
m
atu
re
Resi
stan
ce)
0
.31
7
(
Oh
m
s
)
K
i
(
Torq
u
e Co
n
stan
t)
0
.03
(
N
m
/A
)
K
e
(E
MF
Co
n
stan
t)
0
.03
(
Vs/rad
)
a (
L
atera
l
dis
tan
ce
o
f
r
ear
wh
eel
cent
e
r
f
ro
m
CO
M)
0
.22
7
5
(
m
)
b
(
Late
ral
d
istan
ce
of
f
ron
t wheel center f
ro
m
C
OM
)
0
.22
7
5
(
m
)
c (
L
atera
l
dis
tan
ce
b
etween COM
and
the sid
e wheel)
0
.22
7
5
(
m
)
x
I
C
R
(
X coo
rdin
ate
o
f
I
n
stan
tan
eo
u
s C
en
tre
o
f
Ro
tatio
n
)
-
0
.15
Table
2
.
Final
cal
culat
ed
Pa
ra
m
et
ers
of SSM
R
Para
m
e
ters
Va
lue
A
1
.24
8
B
1
0
.83
C
1
D
0
F
6
.17
3
N
K
1
.54
V
2
4
V
4.1.
Perf
orm
ance
co
mp
ariso
n r
esult
In
Fig
ure
11
is
(a)
,
(
b)
and
(c
)
sho
w
s
the
res
pons
e
of
PID
a
nd
L
QR
-
H
om
og
ene
ous
an
d
non
-
H
om
og
en
eous Co
ntr
oller.
(a)
Res
ponse
of P
I
D
Co
ntr
ol
(b)
Re
s
pons
e
of Hom
og
e
nous LQR C
ontroll
er
c)
Re
s
pons
e
of
Non H
om
og
e
nous
LQR
cont
ro
ll
er
Fig
ure
11
.
(a),
(b)
a
nd (
c
)
s
ho
ws
th
e re
spo
nse
of PID
and
L
QR
-
Ho
m
og
e
ne
ou
s
a
nd no
n
-
H
om
og
ene
ous
Con
tr
oller
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
7
,
No.
3
,
Septem
ber 2
01
8
:
185
–
196
194
Fr
om
the
res
ul
ts
in
Table
3,
it
isob
se
r
ved
t
hat
the
L
Q
R
with
Ho
m
og
ene
ous
C
on
tr
oller
gi
ve
s
the b
et
te
r
p
e
rfo
rm
ance in
set
tl
ing
ti
m
e, p
eak
a
m
plit
ud
e a
nd
ov
e
rs
hoot.
Table
3
.
Per
for
m
ance Com
pari
so
n
5.
HARD
WA
RE
IM
PLE
MEN
TATION
OF
LQR
ON SS
MR U
SING
N
X
P
LP
C 1
768
Both
PID
an
d
LQR
Co
ntr
ol
le
rs
are
im
plem
ented
on
the
N
XP
LPC
1768
f
or
the
S
kid
Stee
ri
ng
Mob
il
e Ro
bot
and the
pe
rform
ance co
m
par
ison i
s m
ade.
T
he result i
n Ta
ble
4 dem
on
str
at
es the foll
ow
ing
:
a.
The
L
QR C
on
t
ro
ll
er
of the
SSM
R R
obot
has l
arg
er
v
el
ocity
than
t
he
tra
diti
on
al
PID c
ontr
oller
.
b.
The
re
fer
e
nce
velocit
y
of
the
syst
e
m
was
set
at
20
m
/s,
Th
e
LQR
co
ntr
ol
gav
e
pr
act
ic
al
ly
ver
y
le
ss
loss in
co
m
par
i
so
n t
o t
he PI
D C
on
t
r
oller.
c.
Fu
rt
her o
bs
e
rvat
ion
s
hows
th
at
the v
el
ocity
o
f
all
the
wh
ee
ls are the
sam
e
for
LQR C
ontrolle
r.
Hen
ce
ba
sed
on
our
t
heoreti
cal
m
od
el
in
g
s
uppo
rted
by
E
xperim
ental
resu
lt
s,
it
can
be
con
cl
ud
e
d
that
LQR
Con
tr
oller
is
a
n
O
pti
m
al
Con
tr
oller.
Per
for
m
ance
C
om
par
iso
n
with
Em
bedde
d
B
oard
N
XP
LPC
17
68
as
sh
ow
n
in
Ta
ble 4
.
Table
4
.
Per
for
m
ance Com
pari
so
n wit
h Em
bed
de
d
B
oard
N
XP
L
PC
1768
Q4
E
n
co
d
ers
CONTRO
LL
ER
S
W
H
EE
L
S
LE
FT
FRONT
LE
FT
REAR
RIGHT
FRONT
RIGHT
REAR
VEL
OC
IT
Y
PID
13
1
3
.3
1
4
.6
1
4
.7
LQR
1
5
.9
1
5
.9
16
16
COUNTS
PID
2
9
0
0
0
2
9
0
0
0
2
9
0
0
0
2
9
0
0
0
LQR
2
9
0
0
0
2
9
0
0
0
2
9
0
0
0
2
9
0
0
0
ACKN
OWLE
DGE
MENT
Our
since
re
th
anks
goes
to
t
he
Visi
on
Gro
up
of
Scie
nce
and
Tech
nolo
gy
(VGS
T),
K
arn
at
a
ka
t
o
ackno
wled
ge
our
re
searc
h
a
nd
pr
ov
i
de
us
the
fi
nan
ci
al
s
upport
to
car
ry o
u
t
the
resea
rc
h
at
NM
IT. We
e
xpress
our
sincere
tha
nk
s
to
our
c
ollea
gu
e
s
at
t
he
Roboti
cs
Re
se
arch
Ce
nte
r,
N
MIT
for
pro
vid
in
g
s
uppo
rt.
Finall
y
,
our
si
ncer
e
gr
at
it
ud
e
goes
t
o
Prof
.
N
R
S
he
tt
y,
Director
NMIT
a
nd
Dr
.
H
C
Nagara
j
,
Pr
inci
pal
NM
IT
f
or
pro
vid
in
g
t
h
e i
nfrastr
uctu
re supp
or
t a
nd
wholehea
rted
e
nc
oura
gem
ent to car
ry
ou
t t
he re
search
at NM
I
T.
REFERE
NCE
S
[1]
Os
ama
El
shaz
l
y
,
Ahm
ed
Abo
-
Ism
ai
l,
Hos
sam
S.
Abbas
and
Za
k
ar
y
aZy
a
daD,
“
Skid
-
steering
m
obil
e
ro
bo
t
Modeli
ng
and
C
ontrol
,
”2014
UK
A
CC.,
978
-
1
-
47
99
-
5011
-
9/1
4
©
2014.
[2]
K.
Kozlowski
a
nd
D.
Pazde
rski
,
“
Modeli
ng
and
control
of
a
4
-
whee
l
Skid
-
steering
m
obil
e
rob
ot,
”
Int
.
J.
App
l
.
Math.
Com
put
.
Sci.
,
vol
.
14
,
no
.
4,
pp
.
477
–
496
,
2004
.
[3]
A.
Mandow,
J.
L.
Mar
ti
n
ez
,
J.
Morale
s,
J
.
L
.
B
l
anc
o,
A.
G
.
Cer
e
zo,
and
J.
Gonza
le
z
,
“
Expe
r
iment
al
k
ine
m
at
i
cs
for
whee
le
d
skid
-
ste
er
m
obile
robo
ts,”
in
In
Proc.
of
the
IEEE/RSJ
In
te
rna
ti
ona
l
Conf
ere
nc
e
onIntelli
g
ent
Robots
an
d
S
y
stems
,
US
A, 2007,
pp
.
1222
–
1227.
[4]
L.
Car
ac
c
iol
o
,
A
.
D.
Luca,
and
S
.
I
anni
t
ti,
“
Tr
aj
e
ct
or
y
tracki
ng
c
ontrol
of
a
f
our
-
whee
l
diff
ere
nt
i
al
l
y
driv
en
m
obil
e
robot,
”
in
In
Pro
c.
of
IE
EE
In
te
rn
at
ion
al
Conf
ere
n
ce
on
Roboti
cs
a
nd
Autom
at
ion,
US
A,
1999,
pp.
2
632
–
2638.
[5]
K.
Ogat
a, Mode
r
n
Control Engi
n
ee
ring
,
5
th
ed
.
Pear
son Prent
ice
-
Hall
,
2010.
[6]
O.
D.
I
.
Nw
okah
and
Y
.
Hurm
uzlu,
Th
e
Me
cha
n
ical
S
y
s
te
m
s Desi
gn
Handbook, C
RC Pre
ss
,
2002,
ch
26.
Ti
m
e
Speci
ficatio
ns
PID
LQR(
H
)
LQR
(N
H
)
Settlin
g
T
i
m
e
1
.09
0
.59
6
0
.69
2
Peak
A
m
p
litu
d
e
1
.95
0
.71
0
.99
6
Ov
ersh
o
o
t
1
3
.1%
0
.13
6
6
6
.13
6
6
%
Evaluation Warning : The document was created with Spire.PDF for Python.