International Journal of Engineering & Technology, 7 (4.

44) (2018) 203-209

International Journal of Engineering & Technology

Website: www.sciencepubco.com/index.php/IJET

Research paper

State-feedback control with a full-state estimator for a

cart-inverted pendulum system
Indrazno Siradjuddin1, Zakiyah Amalia1*, Erfan Rohadi2, Budhy Setiawan1, Awan Setiawan1, Ratna Ika Putri1, and
Erni Yudaningtyas3
1State Polytechnic of Malang, Electrical Engineering Department, Malang, Indonesia
2State Polytechnic of Malang, Information Technology Department, Malang, Indonesia
3
Brawijaya University, Electrical Engineering Department, Malang, Indonesia
*Corresponding author E-mail: zakiyah amalia@polinema.ac.id

Abstract

A Cart Inverted Pendulum System is an unstable, nonlinear and underactuated system. This makes a cart inverted pendulum system used
as a benchmark for testing many control method. A cart must occupy the desired position and the angle of the pendulum must be in an
equilibrium point. System modeling of a cart inverted pendulum is important for controlling this system, but modeling using assumptions
from state-feedback control is not completely valid. To minimize unmeasured state variables, state estimators need to be designed. In this
paper, the state estimator is designed to complete the state-feedback control to control the cart inverted pendulum system. The mathemat-
ical model of the cart inverted pendulum system is obtained by using the Lagrange equation which is then changed in the state space
form. Mathematical models of motors and mechanical transmissions are also included in the cart inverted pendulum system modeling so
that it can reduce errors in a real-time application. The state gain control parameter is obtained by selecting the weighting matrix in the
Linear Quadratic Regulator (LQR) method, then added with the Leuenberger observer gain that obtained by the pole placement method
on the state estimator. Simulation is done to determine the system performance. The simulation results show that the proposed method
can stabilize the cart inverted pendulum system on the cart position and the desired pendulum angle.

Keywords: Cart Inverted Pendulum; Lagrange Equation; LQR; State Estimator; State-Feedback Control

optimal response with a minimum control signal. Mathematical

1. Introduction models using mechanical transmission and dc motors are added to
approach the real cart inverted pendulum system. The modeling of
The cart inverted pendulum system is a very common system for cart inverted pendulum system with the Lagrange equation and the
testing many control methods such as PID[1], LQR[2], pole state space representation of the inverted pendulum system cart
placement[3], fuzzy logic[4], genetic algorithms[5], LQG[6], etc. are shown in chapter 2. The selection of state feedback parameters
A normal pendulum that is a pendulum facing downward is a sta- with Linear Quadratic Regulator and full-state observers with pole
ble system because there is a gravitational force that will force the placement is shown in chapter 3. The simulation results are shown
normal pendulum back to its starting position. Otherwise, the in- in chapter 4, and the last in chapter 5 there is a conclusion.
verted pendulum or pendulum facing upwards is an unstable sys-
tem. This system also has many problems such as nonlinear, un-
deractuated, and complex systems. Controlling the inverted pen-
2. System Modeling
dulum system is about the position and balance point of the pen-
dulum[7], such as on two-wheeled self-balancing scooters, single- A cart inverted pendulum system is shown in Figure.1. The cart of
wheeled electric unicycles, humanoid robots[8], etc. This makes cart inverted pendulum is moved along c-axis direction by DC
the researchers interested in developing this system to get better motor. Belt and pulley are used in mechanical transmission to gain
system performance. One of the control methods that commonly the torque of the actuator. The angular acceleration and the angu-
used to control this cart inverted pendulum system is the state lar velocity of DC motor are controlled to get the acceleration and
feedback controller[9]. However, the use of the state feedback the velocity of the cart.
controller is not enough to represent all the states in the cart in-
verted pendulum system. A state observer[10] needs to be added 2.1. Lagranges Equation
to find out all the states in this system. This paper discusses the
control of the inverted pendulum system cart using state feedback In this cart inverted pendulum system, there is a force by DC mo-
control with a full-state estimator. The parameters of state feed- tor that applied to the cart, 𝑓. This force affects the system output
back control are obtained by the Linear Quadratic Regulator. The (𝑥, 𝜃). To control the system, Lagranges equation used to deriving
selection of the 𝑸 and 𝑅 weighting matrix is done to get the best a mathematical model of the cart inverted pendulum. The La-
response. The addition of full-state estimator parameters with granges Equation is the equation that combines the kinetic energy
Leuenberger observer gain is done by the pole placement method 𝑲 and potential energy 𝑷 of the system. Using the equalised coor-
with selecting the desired pole. The selection is done to get an dinate 𝒒 = {𝑞1 , … , 𝑞𝑖 , … , 𝑞𝑛 } of the system, the mathematical
Copyright © 2018 Authors. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
204 International Journal of Engineering & Technology

model of the system can be described. 𝑛 is the total established The potential energy is only affected by the mass of the pendulum
equalised coordinate and 𝑞𝑖 is an exempt degree of freedom from 𝑚
the system.
𝑃 = 𝑚𝑔ℎ = 𝑚𝑔𝑙 cos 𝜃 (12)

At last, Eq.(1) can be determined using Eq.(11) and Eq.(12), as the

Lagrangian function shown as
1 1
ℒ = 𝑀𝑥̇ 2 + 𝑚(𝑥̇ 2 + 2𝑥̇ 𝑙 𝜃̇ cos 𝜃 + 𝑙 2 𝜃̇ 2 ) (13)
2 2

−𝑚𝑔𝑙 cos 𝜃
The dynamics of the system must be declared in terms of 𝜃 and 𝑥
as the two degrees of freedom of the system. The equation for
each generalised coordinate can be derivated as follows
Fig. 1: System of a cart inverted pendulum
𝑑 𝜕ℒ 𝜕ℒ
( )− =𝑓 (14)
The Lagrangian equation 𝓛 can be described as 𝑑𝑡 𝜕𝑥̇ 𝜕𝑥

𝑑 𝜕ℒ 𝜕ℒ
𝓛 = 𝑲(𝐪, 𝐪̇ ) − 𝑷(𝐪) (1) ( ) − =0 (15)
𝑑𝑡 𝜕𝜃̇ 𝜕𝜃

where the equalised coordinate 𝐪 and the derivation of it 𝐪̇ are The equation shows that the DC motor or external force only af-
used in the function of kinematic energy. The equalised coordinate fect the cart in 𝑖̂ direction. The external force does not affects the
𝐪 used in the function of potential energy. The term of 𝑥 and 𝜃, pendulum. The differential equation from Eq.(14) and Eq.(15) can
thus 𝐪 = [𝑥, 𝜃] are the cart inverted pendulum system coordi- be derived as
nate.The desired position can be derived as follows

𝑑 𝜕ℒ 𝜕ℒ 𝜕ℒ
(
𝑑𝑡 𝜕𝑞𝑖̇
)−
𝜕𝑞𝑖
= 𝑄𝑖 (2) ( ) = (𝑀 + 𝑚)𝑥̇ + 𝑚𝑙𝜃̇ cos 𝜃 (16)
𝜕𝑥̇

𝑑 𝜕ℒ
where 𝑄𝑖 is the external force that applied in term of 𝑞𝑖 coordinate. ( ) = (𝑀 + 𝑚)𝑥̈ + 𝑚𝑙𝜃̈ cos 𝜃 − 𝑚𝑙𝜃̇ 2 sin 𝜃 (17)
𝑑𝑡 𝜕𝑥̇
The Lagranges equation shown in Eq.(2). The total kinetic energy
of the cart obtained with the movement of the cart is only in hori- 𝜕ℒ
=0 (18)
zontal axis 𝑖̂ direction as 𝜕𝑥
1
𝐾𝑀 = 𝑀𝑥̇ 2 (3) and
2
where 𝑀 is the mass of the cart and 𝑥̇ is the translational velocity
𝜕ℒ
of the cart. The total kinetic energy of the pendulum obtained with ( ̇ ) = 𝑚𝑙𝑥̇ cos 𝜃 + 𝑚𝑙 2 𝜃̇ (19)
the movement of the pendulum in the vertical and the horizontal 𝜕𝜃

axis (𝑗̂ and 𝑖̂) directions. 𝑑 𝜕ℒ

( ) = 𝑚𝑙(𝑥̈ cos 𝜃 − 𝜃̇𝑥̇ sin 𝜃) + 𝑚𝑙 2 𝜃̈ (20)
𝑑𝑡 𝜕𝜃̇
1
𝐾𝑚 = 𝑚(𝑥̇ 𝜃2 + ℎ̇2 ) (4) 𝜕ℒ
2
= −𝑚𝑥̇ 𝑙 𝜃̇ sin 𝜃 + 𝑚𝑔𝑙 sin 𝜃 (21)
𝜕𝜃
where 𝑚 is the mass of the pendulum, ℎ̇ is the projected pendulum
therefore
position rate on the vertical axis, and 𝑥̇ 𝜃 is the projected pendulum
position rate on the horizontal axis. Figure.1 shows that
𝑓 = (𝑀 + 𝑚)𝑥̈ + 𝑚𝑙𝜃̈ cos 𝜃 − 𝑚𝑙𝜃̇ 2 sin 𝜃 (22)
𝑥𝜃 = 𝑥 + 𝑙 sin 𝜃 (5) 0 = 𝑚𝑙(𝑥̈ cos 𝜃 − 𝜃̇ 𝑥̇ sin 𝜃) + 𝑚𝑙 2 𝜃̈ − (23)
ℎ = 𝑙 cos 𝜃 (6) (−𝑚𝑥̇ 𝑙 𝜃̇ sin 𝜃 + 𝑚𝑔𝑙 sin 𝜃)
where 𝑙 is the length of the pendulum rod. Eq.(5) and Eq.(6) can
be derived as
The equations contain nonlinear terms: terms 𝑠𝑖𝑛(), 𝑐𝑜𝑠(), and 𝜃̇ 2 ,
therefore from Eq.(22) and Eq.(23) it can be seen that the equation
𝑥𝜃̇ = 𝑥̇ + 𝑙𝜃̇ cos 𝜃 (7)
represent the nonlinear of the system. The linearised Lagranges
equation can be determined as
ℎ̇ = −𝑙 𝜃̇ sin 𝜃 (8)
𝑓 = (𝑀 + 𝑚)𝑥̈ + 𝑚𝑙𝜃̈ (24)
By substituting Eq.(7) and Eq.(8) into Eq.(4), the kinetic energy of
the pendulum can be derived as
0 = 𝑥̈ + 𝑙𝜃̈ − 𝑔𝜃 (25)
1 2 2
𝐾𝑚 = 𝑚 ((𝑥̇ + 𝑙𝜃̇ cos 𝜃) + (−𝑙 𝜃̇ sin 𝜃) ) (9) The pendulum assumed always rotates near the equilibrium,𝜃 ≈ 0.
2
Hence, the non linear terms can be sin 𝜃 ≈ 𝜃, cos 𝜃 ≈ 1, 𝜃̇ 2 = 0
1
𝐾𝑚 = 𝑚(𝑥̇ 2 + 2𝑥̇ 𝑙 𝜃̇ cos 𝜃 + 𝑙 2 𝜃̇ 2 ) (10) and 𝜃̇𝜃 = 0.
2

The cart and pendulum has a total kinetic energy as 2.2 DC Motor

1 1 The equivalent circuit of DC motor armature shows in Figure.2.

𝐾 = 𝐾𝑀 + 𝐾𝑚 = 𝑀𝑥̇ 2 + 𝑚(𝑥̇ 2 + 2𝑥̇ 𝑙 𝜃̇ cos 𝜃 + 𝑙 2 𝜃̇ 2 ) (11)
2 2
International Journal of Engineering & Technology 205

𝑲𝒓
𝒄𝟐 = (38)
𝒓𝟏

from the Eq.36), the term 𝒍𝜽̈ and Eq.(25) into Eq.(36), the differ-
ential equation in a cart of the system can be derived as
𝟏
𝒙̈ = (𝒄𝟐 𝒗𝒂 − 𝒄𝟏 𝒙̇ − 𝒎𝒈𝜽) (39)
𝑴
Fig. 2: Equivalent Circuit of DC Motor Armature
Rearrange Eq.(25), the equation derived as
The equation for the DC motor armature circuit can be derived as
𝟏
𝒅𝒊𝒂 𝜽̈ = (𝒈𝜽 − 𝒙̈ ) (40)
𝒊𝒂 𝑹𝒂 + 𝑳𝒂 + 𝒗𝒃 = 𝒗𝒂 (26) 𝒍
𝒅𝒕
The next differential function in the pendulum of the system can
𝒗𝒃 , the generated emf (back electromotive force) is proportional to be derived from Eq.(39) and Eq.(40), as
the rotor rotational speed
𝟏 𝟏
𝒅𝜽𝟏 𝜽̈ = (𝒈𝜽 − (𝒄𝟐 𝒗𝒂 − 𝒄𝟏 𝒙̇ − 𝒎𝒈𝜽)) (41)
𝒗𝒃 = 𝑲𝒃 (27) 𝒍 𝑴
𝒅𝒕
𝒄𝟐 𝒄𝟏 (𝑴+𝒎)𝒈
𝒅𝜽 =− 𝒗𝒂 + 𝒙̇ + 𝜽 (42)
where 𝑲𝒃 is the constant of back emf and 𝝎𝟏 = 𝟏. The motor 𝑴𝒍 𝑴𝒍 𝑴𝒍
𝒅𝒕
torque 𝝉𝟏 with the armature current 𝒊𝒂 are proportional , hence Lastly, the function of the differential cart and pendulum move-
ment using Lagranges equation has been obtained, and the dynam-
𝝉𝟏 = 𝑲𝒕 𝒊𝒂 (28) ic actuators have been respected in the development of mathemat-
ical models.
where 𝑲𝒕 is the constant of motor torque. Eq.(27) and Eq.(28) (in
term of 𝒊𝒂 ) substituted into Eq.(26), then 2.3 State Space Representation
𝝉𝟏 𝒅𝟐 𝝉 𝟏 𝒅𝜽𝟏
𝑹𝒂 + 𝑳𝒂 + 𝑲𝒃 = 𝒗𝒂 (29) Figure.3 represents the system in state space model.
𝑲𝒕 𝒅𝒕𝟐 𝒅𝒕

The inductor 𝑳𝒂 can be ignored along with the small size of the
inductor in the rotor, and the function of DC motor armature cir-
cuit in term of 𝝉𝟏 becomes

𝑲𝒃 𝑲𝒕
𝝉𝟏 = −𝑲𝒕 𝝎𝟏 + 𝒗𝒂 (30)
𝑹𝒂 𝑹𝒂

Fig. 3: Open loop system representation in state space

In pulley and geared systems there is the mechanical transmission
applied, The representatives of state space equation can be derived as
𝝉𝟐 𝑵 𝝎 𝒓
= 𝟐= 𝟐= 𝟐 (31)
𝝉𝟏 𝑵𝟏 𝝎𝟏 𝒓𝟏
where 𝑵 is the total amount of gear teeth and 𝒓 is the pulley radius. 𝒙̇ = 𝑨𝒙 + 𝑩𝒖 (43)
The DC motor angular velocity 𝝎𝟏 can be shown by 𝒙̇ as
𝟐𝝅𝒓𝟐 𝒓 𝟐𝝅 𝒚 = 𝑪𝒙 + 𝑫𝒖 (44)
= 𝟐 (32)
𝒙̇ 𝒓𝟏 𝝎𝟏
where 𝟐𝝅𝒓𝟐 is the pulley circle that links using a belt to the cart. where 𝒙 ∈ 𝑹𝒏×𝒏 is the vector of the state and 𝒏 is the state varia-
Hence, ble total number. 𝒙̇ is the vector of state derivation. A vector 𝒖 ∈
𝑹𝒌 is the control input or control vector that has 𝒌 elements of
𝒙̇
𝝎𝟏 = (33) control variables. The 𝑨 ∈ 𝑹𝒏×𝒏 , 𝑩 ∈ 𝑹𝒏×𝒌 and 𝑪 ∈ 𝑹𝒑×𝒏 are
𝒓𝟏
the system, input and output matrices, respectively, where 𝒑 is the
number of output. The output of the vector is derived as 𝒚 ∈ 𝑹𝒑 .
Substituting the Eq.(33) into Eq.(30) generates
The state vector, the derivation of the state vector and the control
𝒙̇ input of the state vector for the cart inverted pendulum system, t
𝝉𝟏 = 𝑲𝒓 (−𝑲𝒃 + 𝒗𝒂 ) (34) are declared as 𝒙 = [𝒙𝜽𝒙̇ 𝜽̇]𝑻 , 𝒙̇ = [𝒙̇ 𝜽̇𝒙̈ 𝜽̈]𝑻 , 𝒖 = 𝒗𝒂 consecutively.
𝒓𝟏
The state variable has 𝒏 = 𝟒 total number in this case. Regulating
The force that caused by the torque 𝝉𝟐 moves the cart, so, the Eq.(39) and Eq.(42) into state-space form, the matrices of the
transform 𝝉𝟏 → 𝝉𝟐 is needed. Hence, it can be proved that system derived as
𝟎 𝟎 𝟏 𝟎 𝟎
𝒇=
𝑲𝒓
(−𝑲𝒃
𝒙̇
+ 𝒗𝒂 ) (35) 𝟎 𝟎 𝟎 𝟏 𝟎
𝒓𝟏 𝒓𝟏 𝒎𝒈 𝒄 𝒄𝟐 𝟏 𝟎 𝟎 𝟎
𝑨= 𝟎 − − 𝟏 𝟎 ,𝑩= ,𝑪=[ ],
𝑴 𝑴 𝑴 𝟎 𝟏 𝟎 𝟎
𝑲𝒕 (𝑴+𝒎)𝒈 𝒄𝟏 𝒄𝟐
where 𝑲𝒓 =
𝑹𝒂
. Eq.(35) and Eq.(24), makes a new term of equal- [𝟎 𝑴𝒍
𝟎]
𝑴𝒍 [− ] 𝑴𝒍
ised coordinate 𝒙 in Lagranges function,that declared as 𝒚 = [𝒙𝜽]𝑻 , 𝑫 = ∅
in assumption only from sensors, the actual value of 𝒙 and 𝜽 can
𝑲𝒓 𝑲𝒓 𝑲𝒃 be observed directly. Practically, the alteration from actual to the
𝒗𝒂 = (𝑴 + 𝒎)𝒙̈ + 𝒙̇ + 𝒎𝒍𝜽̈ (36)
𝒓𝟏 (𝒓𝟏 )𝟐 value that used for calculations is required.

To simplify derivation, the abbreviations are used as follows

𝑲𝒓 𝑲𝒃
𝒄𝟏 = (37)
(𝒓𝟏 )𝟐
206 International Journal of Engineering & Technology

3. State Feedback Control with Full-state Ob- where (𝑞1 , 𝑞2 , ⋯ 𝑞𝑗 ) are greater than 0. The Laplace transfer func-
tion of the closed-loop system in Figure.4 can be used to found the
server denominator.
3.1. State Feedback Control 𝐷(𝑠) = 𝑠𝑰 − (𝑨 − 𝑩𝑲) (52)
Figure.4 shows a diagram of the control system using state feed-
where identity matrix is 𝑰. Therefore, the stability and characteris-
back control. The feedback gain 𝑲 in state feedback control is
tics of the transient response of the closed-loop system can be
used to position the closed loop eigenvalues in desired locations.
determined with all eigenvalues of (𝑨 − 𝑩𝑲). The choosing of the
The main purpose of feedback control is to bring the output of the
feedback gain design is an effort to do, 𝑲 so that the eigenvalues
system 𝒚 to tracks the desired position even though there is inter-
of (𝑨 − 𝑩𝑲) have negative real parts. By solving the following
ference.
Equation of Riccati for a positive definite matrix 𝑷, the optimal
feedback gain 𝑲 can be obtained as

𝑨𝑻 𝑷 + 𝑷𝑨 − 𝑷𝑩𝑅−1 𝑩𝑇 𝑷 + 𝑸 = 0 (53)

and the optimal feedback gain 𝑲 can be derived as

Fig. 4: Closed loop system 𝑲 = 𝑅 −1 𝑩𝑇 𝑷 (54)

The system dynamic equation using state feedback control is de- For a unit step reference input, the transfer function 𝑇(𝑠) of a
rived as follows closed-loop has dc gain 𝑁, 0 < 𝑁 < 1. The closed loop system
transfer function shown in Figure.4 is declared as
𝒙̇ = (𝑨 − 𝑩𝑲)𝒙 + 𝑩𝒖 (45)
𝑇(𝑠) = 𝑪(𝑠𝑰 − (𝑨 − 𝑩𝑲))−1 𝑩 (55)
𝒚 = 𝑪𝒙 (46)
where the dc gain 𝑁 can be derived by 𝑇(𝑠)|𝑠=0 . 𝑁̅ in the system
𝒖 = 𝒓 − 𝑲𝒙 (47) is added as shown in Figure.5 can be the settlement to repair the
̅ is calculated
performance of the steady state error. The pregain 𝑁
where 𝑲 ∈ 𝑹𝒋 . Therefore, the controllability and observability of as 𝑁̅ = 1.
𝑁
the system model checking are necessary to do before the control-
ler is implemented. The controllability(Eq.48)) of the system
means that for each initial state value, control efforts 𝒖 can direct
the state to each final state value in a limited time window if only
𝑪 is equal to the number of state variables. Observability (Eq.49))
of the system is a property of the plant without pay attention to the
selection of the actuator with suitable sensor selection. The rank of
matrix O must be same with the number of the state to test that the Fig. 5: Closed loop system with pregain
system is a system with observability.
The equation of state space can be derived as follow, where 𝑲 ∈
𝑪 = [𝑩|𝑨𝑩|𝑨𝟐 𝑩| ⋯ |𝑨𝒋−𝟏 𝑩] (48) 𝑹𝑗 .

𝑶 = [𝑪|𝑪𝑨|𝑪𝑨𝟐 | ⋯ |𝑪𝑨𝒋−𝟏 ] (49) 𝒙̇ = (𝑨 − 𝑩𝑲)𝒙 + 𝑩𝒖 (56)

3.2. Linear Quadratic Regulator (LQR) 𝒚 = 𝑪𝒙 (57)

̅ − 𝑲𝒙
𝒖 = 𝒓𝑵 (58)
Linear Quadratic Regulator Controller as shown in Figure.4 is a
method based on full state-feedback control. By minimising the
performance cost function, the control gain K can be obtained, it 3.3. Full-state Observer
described as
In systems with many states, it is impossible to know all infor-
∞ mation about the state. However, not all states can be sensed by
𝑱 = ∫𝟎 (𝒙𝑻 𝑸𝒙 + 𝒖𝟐 𝑹) 𝒅𝒕 (50)
sensors, and can be reversed. The system with state feedback re-
quires an ideal sensor that has an infinite bandwidth. In fact there
where 𝑹 is a scalar 𝑸 ∈ 𝑹𝒋×𝒋 is a real symmetric matrix which
are no ideal sensors and sensors also have limited bandwidth. The
must be chosen. 𝑹 and 𝑸 establish the relative concern of errors
actual observer concept is that if we do not have all the states, it is
and energy costs. Hence, the matrix 𝑹 and 𝑸 signifies a trade-off possible to estimate these states by using the system inputs and
both in control efforts and performance. If 𝑸 is smaller than 𝑹, outputs. This concept can be done if observability conditions are
control systems have become expensive since then 𝑱 especially the met. The concept of the full-state observer is as shown in Figure.6.
use of the control energy. If not, when 𝑸 is bigger than 𝑹, control
systems become cheap because small arbitrary control efforts can
be used to stabilize the system, but the system response will be
slow. It is a general training to let 𝑹 > 𝟎 and 𝑸 be a diagonal
matrix in the form of

𝑞1 0 ⋯ 0
0 𝑞2 ⋯ 0
𝑸=[⋮ ⋮ ⋱ 0] (51)
0 0 ⋯ 𝑞𝑗
International Journal of Engineering & Technology 207

move the unstable poles to the stable poles in the left side of the
complex plane such that the obtained control gain 𝑳 can satisfy the
designed criteria. The characteristic polynomial of the system can
be derived as Eq.(68). 𝒂𝒊 are the polynomial coefficients and 𝑰 is
an identity matrix. If the desired eigenvalues 𝝁𝟏 , ⋯ , 𝝁𝒏 are con-
sidered, the desired characteristic polynomial of the system be-
comes as Eq.(69). Then, the output feedback gain matrix that re-
quired 𝑳 can be calculated using Eq.(70).
|𝒔𝑰 − 𝑨| = 𝒔𝒏 + 𝒂𝟏 𝒔𝒏−𝟏 + ⋯ + 𝒂𝒏−𝟏 𝒔 + 𝒂𝒏 (68)
∏𝒏𝟏=𝟏(𝒔 − 𝝁𝒊 ) = 𝒔𝒏 + 𝜶𝟏 𝒔𝒏−𝟏 + ⋯ + 𝜶𝒏−𝟏 𝒔+𝜶𝒏 (69)
𝑳 = [𝜶𝒏 − 𝒂𝒏 |𝜶𝒏−𝟏 − 𝒂𝒏−𝟏 | ⋯ |𝜶𝟐 − 𝒂𝟐 |𝜶𝟏 − 𝒂𝟏 ]𝑻−𝟏 (70)
where 𝑻 is the transformation matrix that changes the state equa-
tion of the system into the controllable canonical form. The trans-
formation matrix 𝑻 can be calculated by

𝑻 = 𝑪𝑾 (71)
Fig. 6: Closed loop system with full-state observer
Where
For the systems as in Eq.(43) and Eq.(44), with the state estima-
tion of 𝒙 defined as follow 𝒂𝒏−𝟏 𝒂𝒏−𝟐 ⋯ 𝒂𝟏 𝟏
𝒂𝒏−𝟐 𝒂𝒏−𝟑 ⋯ 𝟏 𝟎
̂̇ = 𝑨𝒙
𝒙 ̂ + 𝑩𝒖 + 𝑳(𝒚 − 𝒚
̂) (59) 𝑾= ⋮ ⋮ ⋮ ⋮ ⋮ (72)
𝒂𝟏 𝟏 ⋯ 𝟎 𝟎
̂ = 𝑪𝒙
𝒚 ̂ (60) [ 𝟏 𝟎 ⋯ 𝟎 𝟎]
The value of 𝒚 ̃ = 𝒚−𝒚 ̂ is attempted to be as small as possible,
which indicates that the estimator works well. To make 𝒙 ̂ as close
4. Simulation and Result
as possible 𝒙, state error estimator is defined.
Before doing a simulation, the parameters of the system defined in
̃=𝒙−̂
𝒙 𝒙 (61) Table 1 are used. The open loop system has the eigenvalues that
shown in Figure.3 are (0, 5.3622, -5.4316, -0.3696) which can be
̂̇
̃̇ = 𝒙̇ − 𝒙
𝒙 (62) obtained by det(𝒔𝑰 − 𝑨). That eigenvalue is shown in pole-zero
mapping in Figure.7.
̃̇ = (𝑨𝒙 + 𝑩𝒖) − (𝑨𝒙
𝒙 ̂ 𝒙))
̂ + 𝑩𝒖 + 𝑳(𝑪𝒙 − 𝑪 (63)

̃̇ = 𝑨(𝒙 − ̂
𝒙 𝒙) − 𝑳𝑪(𝒙 − ̂
𝒙) (64)

̃̇ = (𝑨 − 𝑳𝑪)𝒙
𝒙 ̃ (65)

One of the controller design is Output Feedback Control. By using

Output Feedback Control, the existing states can be estimated
using only system inputs and outputs. One method used in Output
Feedback Control is to use Leuenberger estimation. By calculation
Fig. 7: Pole-Zero Mapping for a Cart Inverted Pendulum System
𝑳 = (−𝑺−𝟏 𝑪𝑷)𝑻 (66)
The open loop of cart inverted pendulum system has a positive
where 𝑷 is the solution of the Riccati equation real part or has a pole in the right side of the complex plane.
Hence, the system is a system that unstable. Nevertheless, using
𝑨𝑷 + 𝑷𝑨𝑻 + 𝑸 − 𝑷𝑪𝑻 𝑹−𝟏 𝑪𝑷 = 𝟎 (67) the system state space described in Eq.(43) and Eq.(44), the cart
inverted pendulum system is a completely observable and control-
In this case, the determination of the 𝑳 matrix is done by the pole lable. The ranks of the matrix 𝑪 and 𝑶 are equal to the number of
placement control method. the state respectively, rank(𝑪)= rank(𝑶) = 𝟒, that verified using
the result that shown in Eq.(73) and Eq.(74).
3.4. Pole Placement

In the design of the controller, the pole location of the system on Table.1: The cart inverted pendulum parameters
the complex plane can verify the system stability . The system is No Parameter Value Units
unstable if it has a pole on the right side of the complex plane. In 1 M 1.04 Kg
other words, the system is said to be stable when all the poles are 2 Ra 1 Ω
on the left side of the complex plane[11]. Therefore, a system that 3 l 0.4 m
has a pole in the right side of the complex plane requires a control 4 Kt 0.01 N.m/A
method that can carry the pole that located in the right side of the 5 r1 0.1 M
complex plane to the left side of the complex plane. As with the 6 Kb 0.407 V.s/rad
previous LQR method, controllability and observability of the 7 m 0.19 Kg
system must be fulfilled. The suitable gain matrix can be selected
𝟎 𝟎. 𝟐𝟎𝟎 −𝟎. 𝟎𝟎𝟖 𝟎. 𝟕𝟖𝟓
for the state feedback as shown in Figure.4, to have closed-loop 𝟎 −𝟎. 𝟖𝟎𝟎 𝟎. 𝟎𝟑𝟐 −𝟑𝟒. 𝟓𝟑𝟑
poles at the desired locations, it is possible to force the system, 𝑪=[ ]
𝟎. 𝟐𝟎𝟎 −𝟎. 𝟎𝟎𝟖 𝟎. 𝟕𝟖𝟓 −𝟎. 𝟎𝟔𝟑
provided that the original system completely states controllable. −𝟎. 𝟖𝟎𝟎 𝟎. 𝟎𝟑𝟐 −𝟑𝟒. 𝟓𝟑𝟑 𝟏. 𝟓𝟎𝟕
All the variables of state must be available and measurable in pole
placement control method. The pole placement method is used to
208 International Journal of Engineering & Technology

𝟏 𝟎 𝟎 𝟎
𝟎 𝟏 𝟎 𝟎
𝟎 𝟎 𝟏 𝟎
𝟎 𝟎 𝟎 𝟏
𝑶=
𝟎 −𝟎. 𝟗𝟖𝟏 −𝟎. 𝟎𝟒𝟎 𝟎
𝟎 𝟒𝟑. 𝟏𝟔𝟒 𝟎. 𝟏𝟔𝟎 𝟎
𝟎 𝟎. 𝟎𝟑𝟗 𝟎. 𝟎𝟎𝟐 −𝟎. 𝟗𝟖𝟏
[𝟎 −𝟎. 𝟏𝟓𝟕 −𝟎. 𝟎𝟎𝟔 𝟒𝟑. 𝟏𝟔𝟒 ]

4.1. State Feedback Control with Linear Quadratic

Regulator
Fig. 11: Pendulum Angle Response
The simulation of cart inverted pendulum system of state feedback
control has been obtained with a Linear Quadratic Regulator. The In this section, the simulation of the state feedback controller has
weighting matrix 𝑸 and 𝑹 that used in 3 variations of the simula- been obtained.
tion are 𝒒𝟏 = (𝟏. 𝟓, 𝟏𝟓) with the value of 𝒒𝟐 = 𝟏 and 𝑹 = 𝟏 in all
combination. Figure.8 and Figure.9 shows the response of the cart 4.2. A Full-state observer with Pole Placement
position and angle pendulum respectively.
The simulation of cart inverted pendulum system using state feed-
back control with the full-state observer. The parameter that used
in full-state observer has been obtained with pole placement con-
trol method. The pole that used for a gain matrix in Leuenberger
in this simulation are (-40, -41, -42, -43). The result of the simula-
tion also shown in Figure.10 and Figure.11. Figure.10 and Fig-
ure.11 shows the response of the cart position and angle pendulum
respectively. The simulation shows that the system can reach the
desired position respectively. The response with full-state observer
is similar to the response with state feedback control. It because all
Fig. 8: Cart Position Response of the state variables had been included in the mathematical model
of the cart inverted pendulum system. The observer state also
identical with the actual plant.

5. Conclusion
State feedback control with a full-state estimator for a cart invert-
ed pendulum system has been successfully conducted. The gain
state feedback 𝑲 has been obtained using Linear Quadratic Regu-
lator. The gain of full-state observer 𝑳 also been obtained using
pole placement control method. The simulation result shows that
the cart inverted pendulum system can reach the desired position
Fig. 9: Pendulum Angle Response and angle with the optimal responses. The simulation also shows
that the system using state feedback control and state feedback
The cart position response shows that the cart can't reach the de- control with full state observer have similar responses. Its because
sired position, otherwise, the pendulum can reach the equilibrium. the initial state and the observer state have same state variables in
The bigger 𝒒𝒊 that is given to the system, the responses become mathematical modeling. The implementation can be done to know
faster. Because the response of the cart that shown in Figure.8 the response of the cart inverted pendulum system on a real im-
̅ added to the system for
can't reach the desired position, pregain 𝑵 plementation.
a better performance. Figure.10 and Figure.11 shows the response
of the cart position and angle pendulum respectively. It can be
seen that both of the cart and pendulum can reach the desired posi-
Acknowledgment
tion.
The authors would like to thank Kemenristekdikti and State Poly-
technic of Malang for their support.

References
[1] I. Siradjuddin, Z. Amalia, B. Setiawan, F. Ronilaya, E. Rohadi, A.
Setiawan, S. Adhisuwignjo, ”Stabilising a cart inverted pendulum
with an augmented PID control scheme,” In MATEC Web of Con-
ferences, EDP Sciences, 2018, Vol. 197, p. 11013.
[2] I. Siradjuddin, B. Setiawan, A. Fahmi, Z. Amalia, E. Rohadi, ”State
space control using LQR method for a cart-inverted pendulum line-
arised model,” International Journal of Mechanical & Mechatronics
Fig. 10: Cart Position Response
Engineering IJMME-IJENS, 2017, Vol:17, No:01
[3] I. Siradjuddin, Z. Amalia, B. Setiawan, R. P. Wicaksono and E.
Yudaningtyas, ”Stabilising a cart inverted pendulum system using
pole placement control method,” 2017 15th International Confer-
ence on Quality in Research (QiR) : International Symposium on
Electrical and Computer Engineering, Nusa Dua, 2017, pp. 197-
203.
[4] Zhenkai Guo, Jianqin Mao and Ping Wang, ”An Approach of
Adaptive Fuzzy Modeling and Application to Inverted Pendulum
International Journal of Engineering & Technology 209

Control,” 2006 6th World Congress on Intelligent Control and Au-

tomation, Dalian, 2006, pp. 1594-1598.
[5] L. Ye, ”Secondary Pendulum Control System Based on Genetic
Algorithm and Neural Network,” 2006 Chinese Control Confer-
ence, Harbin, 2006, pp. 1152-1155.
[6] E. S. Sazonov, P. Klinkhachorn and R. L. Klein, ”Hybrid LQG-
neural controller for inverted pendulum system,” Proceedings of the
35th Southeastern Symposium on System Theory, Morgantown,
WV, USA, 2003, pp. 206-210.
[7] Y. Kado, Y. Pan and K. Furuta, ”Control System for Skill Acquisi-
tion- Balancing Pendulum based on Human Adaptive Mechatron-
ics-,” 2006 IEEE International Conference on Systems, Man and
Cybernetics, Taipei, 2006, pp. 4040-4045.
[8] I. Park, J. Kim and J. Oh, ”Online Biped Walking Pattern Genera-
tion for Humanoid Robot KHR-3(KAIST Humanoid Robot - 3:
HUBO),” 2006 6th IEEE-RAS International Conference on Hu-
manoid Robots, Genova, 2006, pp. 398-403.
[9] B. A. Sharif and A. Ucar, ”State feedback and LQR controllers for
an inverted pendulum system,” 2013 The International Conference
on Technological Advances in Electrical, Electronics and Computer
Engineering (TAEECE), Konya, 2013, pp. 298-303.
[10] Z. Jie and R. Sijing, ”Sliding mode control of inverted pendulum
based on state observer,” 2016 Sixth International Conference on
Information Science and Technology (ICIST), Dalian, 2016, pp.
322-326. doi:10.1109/ICIST.2016.7483432\
[11] C. Mahapatra and S. Chauhan, ”Tracking control of inverted pendu-
lum on a cart with disturbance using pole placement and LQR,”
2017 International Conference on Emerging Trends in Computing
and Communication Technologies (ICETCCT), Dehradun, 2017,
pp. 1-6.