Professional Documents
Culture Documents
by Nonlinear Controller
Qiguo Yan
FlyingTiger Technologies, Inc.
2123 W 7th St
Brooklyn, NY 11223, USA
qiguoyan@yahoo.com
Abstract A novel tracking control law is de-
veloped for underactuated nonminimum inverted
pendulum on a rotor arm by a combinated appli-
cation of nonlinear backstepping, dierential at-
ness, and small gain theorem. It has been shown
that through proper input transformation, the
whole system can be modied to be weak min-
imum phase with the external subsystem being
dierentially at, and be further modied into 2
exp-ISS subsystems. Simulation and experimen-
tal results for tracking a reference signal are pre-
sented to verify the eectiveness of the proposed
methodology.
1. Introduction
The stabilization and output tracking of underac-
tuated mechanical systems has been a challenging
task for many years [1, 2]. The diculty encoun-
tered in controlling this kind of system arised from
the nonminimum phase characteristics.
The Rotary Inverted Pendulum (RIP) is a non-
trivial simplied model unifying the pendulum on a
cart and Euler-Poincare system such as a satellite
with momentum wheels. Consisting of two rigid
link with an actuator at only the arm link, this RIP
system was rst described in [3]. A sliding-mode
controller for regulation was given in [4]. A smooth
stabilization controller was derived in [5] using the
controlled Lagrangian method, unfortunately, this
framework provided no basis for output tracking.
The current output tracking methods are mainly
based on the standard output regulation theory
[69] etc. The fundamental limitation of this ap-
proach is that it depends on a locally exponentially
stable nonlinear closed-loop system through linear
approximation theory and a reference signal gen-
erated by a neutral stable exosystem, and involves
solving a set of nonlinear PDEs. Inversion-based
output tracking schemes [10] for minimum and non-
minimum phase nonlinear systems still require lin-
ear approximation and local closed loop exponen-
tial stability. Output tracking by recursive back-
stepping method [11] depends on the assumption
if the system dynamics can be successfully trans-
formed into strict-feedback form, or pure-feedback
form, therefore, is not applicable to the class of non-
linear nonminimum systems for which such a trans-
formation is impossible.
The purpose of this paper is to overcome these
drawbacks by a combinated application of nonlinear
backstepping, the dierential atness [12], and non-
linear small gain theorem [16] to the output tracking
problem of the nonlinear underactuated nonmini-
mum system. In this paper, rst, a new output is
dened to transform the nonminimum system into a
weak minimum one. Next, an input transformation
is utilized to transform the external subsystem into
a dierentially at one while preserving the weak
minimum phase characteristics. Finally, the whole
system is further transformed into 2 exp-ISS sub-
systems [14, 16], and a controller is proposed which
guarantees uniformly ultimately bounded output
tracking errors within the domain of attraction.
2. the RIP Dynamics and Dynamics Transfor-
mation
2.1. the RIP Dynamics
A schematic representation of the RIP system is
given in Figure 1 where l
p
denotes the pendulum
length and m
p
the pendulum mass. Let be the
pendulum angle. The total eective moment of base
inertia is J
b
. The pivot arm O
A has a length of
r, and is mounted on the load shaft of the SRV-
02 DC-motor which applies a torque to drive the
arm rotate on the horizontal plane. The plane of
pendulum is orthogonal to the radial arm. The RIP
dynamics is governed by
(A +B sin
2
)
+ (C cos) (C sin)
2
+(2 B sincos)
= , (1)
B + (C cos)
(B sincos)
2
D sin = 0, (2)
where A
=
m
p
r
2
+ J
b
, B
=
1
3
m
p
l
2
p
, C
=
1
2
m
p
r l
p
,
D
=
1
2
m
p
g l
p
. The natural output is . For state-
ment convenience, we refer to (1) as the external
dynamics (subsystem), and (2) as the internal dy-
namics (subsystem).
Proceedings of the 42nd IEEE
Conference on Decision and Control
Maui, Hawaii USA, December 2003 WeM12-3
0-7803-7924-1/03/$17.00 2003 IEEE 2395
Figure 1: Schematic representation of RIP system
The numerical values of the mechanical subsys-
tem parameters for a laboratory RIP-model are pro-
vided in Table 1.
Physical quantity Symbol value Units
Pivot arm length r 8.750.0254 m
Base inertia moment J
b
0.005 Kg-m
2
Pendulum length l
p
13.1250.0254 m
Pendulum mass m
p
0.126 Kg
Gravitational const g 9.8 m/sec
2
Table 1: Parameters of the experiment RIP System
We have A = 0.0112, B = 0.0047, C = 0.0047, and
D = 0.2058. Obviously, (2) is uncontrollable and
unstable when =
2
, so all considerations shall
be conned within the range || <
2
.
2.2. Dene New Output to Make the System
Weak Minimum Phase
In the sense of stability of the zero dynamics [17],
the (1)-(2) with as output is nonminimum. One
approach to circumvent this barrior is to redene
output to stabilize the internal dynamics [18]. Now,
imagine there is a massless pendulum with an arm
of r
+ h sin, (3)
and then
Z as
Z = r
+h cos h
2
sin. (4)
With the new output variable Z, the internal dy-
namics (2) can be represented as
C
r
cos
Z+(
C h
r
cos
2
B) (
C h
r
sincos)
2
+B(
Z
r
h
r
cos)
2
sincos+Dsin = 0. (5)
Proposition 2.1. The system (1)-(2) is weak
minimum phase with Z as output if
Ch
r
cos
2
B > 0
|| < tan
1
C h Br
Br
<
2
. (6)
Proof. The new zero dynamics can be obtained
from (5) as
(
C h
r
cos
2
B) (
C h
r
sincos)
2
+B(
h
r
cos )
2
sincos +D sin = 0, (7)
A Lyapunov function candidate can be chosen as
V (, ) =
1
2
_
C h
r
cos
2
B
_
2
+B
_
0
_
h
r
a cos a
_
2
sina cos a da
+D( 1 cos ), (8)
The 1st term in Lyapunov candidate is positive
denite and radially unbounded in direction if
Ch
r
cos
2
B > 0 ; the 2nd term can be shown
to be positive denite and strictly increasing for
within the range || <
2
; the 3rd term is also
positive denite and strictly increasing within this
range. Also, V (0, 0) = 0. This proved that within
the range || < tan
1
_
C hBr
B r
<
2
, (8) which
is positive denite is a Lyapunov function having
arbitrarily small upper limit [19], and can be well
interpreted in the sense of energy. Next, dierenti-
ate (8), we obtain
V (, ) = (Eq.(7)) = 0. (9)
This concluded the proof.
It can be shown by (6) the stable range of ||
2
as
h
r
= 2.25 .
Denition 2.1. Domain of Stability The set
of all points (, ) in the periodic trajectories in the
phase portrait of the zero dynamics (7), e.g., as in
Figure 2, denoted as , is the domain of stability of
the RIP system for a specic
h
r
.
2396
50 40 30 20 10 0 10 20 30 40 50
800
600
400
200
0
200
400
600
800
a (deg)
D
e
riv
a
tiv
e
o
f a
(d
e
g
/s
e
c
)
Figure 2: Phase portrait for
h
r
= 2.25
2.3. RIP Dynamics Transformation
Dierential atness [12, 13] of systems has been
proven to be very useful and particularly easy for
output trajectory tracking. To exploit this prop-
erty, were going to transform the external subsys-
tem into such dierentially at one with Z as at
output.
Proposition 2.2. Suppose AB > C
2
, then
(, ) , the RIP dynamics can be transformed
into (15)-(16) which is weak minimum phase, and
the external dynamics (15) is dierentially at with
Z as the at output if the control input is chosen as
= [(M() h C) sin]
2
+(B sin2)
+
_
BG() sin cos
2
2
+DG() sin cos
K
1
M()
Z K
2
M() Z +M()v, (10)
where K
1
, K
2
> 0, M(), G() are determined by
(12) and (13) respectively.
Proof: we solve for from (5), and substituting ,
2
+DG() sincos = , (11)
where
M()
=
1
r
_
(A+Bsin
2
) G()C cos
2
)
_
, (12)
G()
=
(A +Bsin
2
) h C r
(C hcos
2
Br
)
. (13)
When cos
2
=
AB+B
2
B
2
+C
2
, M() = 0, (11) is con-
trollable. Specically, when
AB > C
2
, 0 > M() / 0. (14)
Then substitute (10) into (11), the new external dy-
namics becomes
Z +K
1
Z +K
2
Z = v, (15)
which is dierentially at with the at output y =
Z. Let (15)
C
r
cos
2
B) (
C h
r
sincos)
2
+B(
Z
r
h
r
cos)
2
sincos+Dsin
+(
K
1
C
r
cos)
Z + (
K
2
C
r
cos) Z
= (
C
r
cos) v. (16)
The zero dynamics of (15)-(16) is (7), which is stable
at the origin (, ) .
3. Output Tracking Controller Design
Now the RIP system is weak minimum with the
external subsystem being dierentially at. Let Z
d
be the desired output trajectory. Our objective is
to design control input such that Z(t) Z
d
(t) as
t while ensuring internal stability on the basis
of the transformed dynamics (15)-(16).
3.1. Asymptotic Output Trajectory Tracking
Controller Design for Z
In this subsection, we are going to make use of dif-
ferential atness of (15) to design an output tracking
controller. First, we make one assumption.
Assumption 1. Suppose AB > C
2
, and
Z
d
,
Z
d
,
Z
d
are bounded such that the time response
of the dynamics
(
C h
r
cos
2
B) (
C h
r
sincos)
2
+B(
Z
d
r
h
r
cos)
2
sincos
+Dsin =
C
r
cos
Z
d
(17)
is within the domain .
This assumption sets up a uplimit for the RIP
system to resist a tracking acceleration impact.
Proposition 3.1. Suppose assumption1 holds,
then (, ) , the whole system is stable and
the output tracking error
lim
t
Z
e
(t),
Z
e
(t) = 0 (18)
where Z
e
=
Z Z
d
, the control input is given by
(10), and v is determined by
v =
Z
d
+K
1
Z
d
+K
2
Z
d
. (19)
2397
Proof: Substitute (19) into (15) to get the closed
loop external error dynamics and internal dynamics,
respectively, as
Z
e
+K
1
Z
e
+K
2
Z
e
= 0, (20)
(
C h
r
cos
2
B) (
C h
r
sincos)
2
+B(
Z
e
+
Z
d
r
h
r
cos)
2
sincos+D sin
+(
K
1
C
r
cos)
Z
e
+(
K
2
C
r
cos) Z
e
= (
C
r
cos)
Zd. (21)
With proposition2.1 and assumption1, the zero
dynamics of (20)-(21) is BIBO. According to [15],
an extension of nonlinear block backstepping theo-
rem in [11], we can conclude the proof.
3.2. Bounded Output Trajectory Tracking Con-
troller Design
The objective of this subsection to make both Z
e
and ultimately uniformly bounded around the ori-
gin.
It can be shown that even the time response of
of (17) or (21) is bounded within the domain ,
a small extra disturbance can easily drive out of
this domain. Therefore, further modication of the
internal dynamics (21) to make it exp-ISS by giving
up exact output tracking in (19) is essential for the
output tracking design. [10] did the similar thing
to remove the nonhyperbolicity of nonlinear system
with nonhyperbolic internal dynamics when apply-
ing stable inversion. The dierence between the pro-
posed technique and [10] is that our framework is
totally based on nonlinear approach.
To modify (21), an extra term K
3
, K
3
> 0 is
added to (19),
v =
Z
d
+K
1
Z
d
+K
2
Z
d
K
3
, (22)
accordingly,
= [(M() h C) sin]
2
+(B sin2)
+
_
BG() sin cos
2
2
+DG() sin cos
+M()
Z
d
K
1
M()
Z
e
K
2
M() Z
e
K
3
M() (23)
substitute (22) into (15) and (16) to obtain closed
loop tracking error external subsystem and internal
subsystem, respectively, as
Z
e
+K
1
Z
e
+K
2
Z
e
+ K
3
= 0,
y
1
= [Z
e
,
Z
e
]
T
, (24)
(
C h
r
cos
2
B) (
C h
r
sincos)
2
+B(
Z
e
+
Z
d
r
h
r
cos)
2
sincos+Dsin
+(
K
3
C
r
cos) + (
K
1
C
r
cos)
Z
e
+(
K
2
C
r
cos) Z
e
= (
C
r
cos)
Z
d
,
y
2
= [, ]
T
. (25)
Proposition 3.2. Suppose assumption1 holds,
then (, ) , the interconnection system (24)-
(25) is exp-ISS with
Z
d
as input, or the output tra-
jectory tracking errors Z
e
, are ultimately, uni-
formly bounded around the origin within this do-
main if K
1
, K
2
, K
3
> 0 and meet the small gain
constraint (29).
Regretfully we omitted the denition of exp-ISS
system as in [14, 16], and the small gain theorem for
2 exp-ISS subsystem [16], due to space limitation.
Now we prove proposition3.2.
Proof: Let
Z
=
[
Z
e
, Z
e
]
T
,
=
[ , ]
T
. First we
prove that (24) is exp-ISS. By treating = constant
as input, choose V
1
(
Z) =
Z
T
P
Z,
P =
P
T
> 0, such
that
P
_
0 1
K
2
K
1
_
+
_
0 1
K
2
K
1
_
T
P = Q,
Q = Q
T
> 0.
The derivative of V
1
with respect to time along so-
lution of (24), for any
z
> 0 satisfying
z
<
min
,
we have
V
1
(
Z)
min
(Q)|
Z|
2
+K
3
z
|
Z|
2
_
(
min
z
)|
Z|
2
K
3
| | |
Z|
_
z
|
Z|
2
,
whenever |
Z|
z
| |, where
z
=
K
3
(
min
z
)
, we
establish that, t > 0,
1
,
C
1
> 0,
|
Z|
C
1
e
1
t
|
Z(0)| +
z
. (26)
In other words, (24) is exp-ISS. Similarly, consider
(25), by treating
Z,
Z
d
,
Z
d
as constant, we choose
V
2
( ) =
1
2
(
C h
r
cos
2
B)
2
+B
_
0
(
Z
e
+
Z
d
r
h
r
a cos a)
2
sina cos a da
+D( 1 cos ). (27)
2398
Let K
=
K
3
C
r
cos||
max
,
K
=
C
r
(K
1
+ K
2
). The
derivative of V
2
with respect to time along solution
of (25), for any
Z
,
Z
d
>0 satisfying
Z
+
Z
d
<K
,
V
2
( )
K
3
C
r
cos||
max
2
+
C
r
(K
1
+K
2
) | | |
Z| +
C
r
| | |
Z
d
|
Z
2
[
Z
d
2
C
r
| | |
Z
d
|]
[(K
Z
d
)
2
K| | |
Z|]
Z
2
,
whenever | |
z
|
Z| +
C
Z
d
r
Z
d
|, where
z
=
K
(K
Z
d
)
, so V ( ) is bounded, is
bounded. By applying LaSalles invariant theorem,
we have, t > 0,
2
,
C
2
> 0,
| |
C
2
e
2
t
| (0)| +
z
Z +
C
Z
d
r
Z
d
. (28)
According to small gain theorem [16], if
z
z
<
1, that is,
K
3
(
min
z
)
C (K
1
+K
2
)
(
K
3
C
r
cos||
max
Z
d
)r
<1, (29)
then the whole system is exp-ISS with
Z
d
as input,
Z (1
z
z
)
1
[
C
1
e
1
t
|
Z(0)|
+
C
2
z
e
2
t
| (0)|+
C
Z
d
r
Z
d
], (30)
(1
z
z
)
1
[
C
2
e
2
t
| (0)|
+
C
1
z
e
1
t
|
Z(0)| +
C
Z
d
r
Z
d
]. (31)
The proof concluded.
Remark 3.1. If Z
e
, 0, (, ) , then
(
d
) 0 where
d
=
1
r
Z
d
; if Z
e
, are ultimately
uniformly bounded around the orign (, ) ,
then (
d
) is also ultimately uniformly bounded.
Remark 3.2. For the case
Z
d
= 0, Z
e
,
Z
e
, , ,
(
d
), (
d
) all approach to zero exponentially
as t .
4. Simulation and Experimental Verication
For laboratory implementation, the voltage input
for SRV-02 DC-motor is determined by
V
i
=
R
a
K
g
K
T
+ K
b
K
g
, (32)
where DC-motor constants R
a
= 2.6(ohm), K
b
=
0.00767(volt-sec/rad), K
T
= 0.00767(N-m/amp),
and gear ratio K
g
= 14 5. Optical encoders and
tachometers were mounted to pick up , , , and
+ h sin) track
Z
d
= a
0
sinw(t +t
0
), where a
0
, w, t
0
are constants,
and track 0 (deg), simultaneously, would track
d
=
180
Z
d
r
= 1.6632 l
p
, h =
1.8125 l
p
. The stable range for was determined
by (6) as || < 16.6788
o
(deg).
For the experiment, Z
d
= 0.7sin{
2
10
(t + 0.18)}.
Gains were picked as K
1
= 5, K
2
= 6, and K
3
= 2.
The initial conditions were
0
= 13.64
o
(deg),
0
=
103.64
o
(deg), and
0
=
0
= 0.00 (deg/sec). In
Figure 3, the solid and dashed lines were the exper-
iment and simulation trajectories respectively for ,
and the dotted line for
d
. Figure 4 show the per-
formance for to track 0, the sawed and smooth
curves were for the experiment and simulation re-
sults respectively. Figure 5 showed the control in-
puts.
Except physical experiment with limitations by
the capacity of SRV-02 DC-motor, numerous nu-
merical simulation can be easily testied. Results
are omitted here.
5. Conclusion
A novel nonlinear approach was proposed such
that the nonminimum RIP system was modied to
be weak minimum phase with the external subsys-
tem being dierentially at, and an asymptotically
tracking controller for that at output was derived
rst while ensuring the internal pendulum dynam-
ics marginally stable at the origin (, ) . This
controller was then modied such that the RIP sys-
tem was further transformed into 2 exp-ISS sub-
systems and all tracking errors were uniformly ul-
timately bounded around the origin within this do-
main of stability. A new generalized strategy is pos-
sible for the output tracking designs of nonlinear
underactuated nonminimum systems.
References
[1] R. Gurumoorthy and S.R. Sanders (1993).
Controlling Nonminimum Phase Nonlinear
Systems- the Inverted Pendulum on a Cart
Example. In: Proc. of American Control Con-
ference, San Francisco, CA, pp. 680685.
[2] Ohsumi A., T., Izumikawa (1995). Nonlin-
ear Control of Swing-up and Stabilization of
2399
an Inverted Pendulum. In: Proc. of the 34th
CDC, New Orlean, LA, pp. 38733880.
[3] IWASHIRO M., K. FURUTA, and K. J.,
Z
d
r
0 5 10 15
15
10
5
0
5
time (sec)
a
(d
e
g
)
Figure 4: Tracking of to 0 deg
0 5 10 15
15
10
5
0
5
time (sec)
V
i (v
o
lt)
Figure 5: Control input V
i
volts
2400