See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283659634

Attitude Control and Stabilization of a Two-Wheeled Self-Balancing Robot

Article in Control Engineering and Applied Informatics · September 2015

CITATIONS READS

19 3,397

3 authors, including:

Omer Saleem Muhammad Anas Imtiaz

National University of Computer and Emerging Sciences Boston University
46 PUBLICATIONS 467 CITATIONS 11 PUBLICATIONS 124 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Omer Saleem on 03 February 2020.

The user has requested enhancement of the downloaded file.

CEAI, Vol.17, No.3 pp. 98-104, 2015 Printed in Romania

Attitude Control and Stabilization of a Two-Wheeled Self-Balancing

Robot
Omer Saleem Bhatti*, Khalid Mehmood-ul-Hasan**,
Muhammad Anas Imtiaz*

* Electrical Engineering Department, FAST - NU,

Lahore, Pakistan,( e-mail: [email protected], [email protected])
** Electrical Engineering Department, UET,
Lahore, Pakistan, (e-mail: [email protected])

Abstract: The paper demonstrates development of attitude control and stabilization technique of a self-
balancing robot. The main aim is to ensure its vertical stability, even in the presence of an external
bounded impulsive force. By electronically programming a hard-coded vertical reference position, the
proposed robotic system can be balanced at the desired set-point angle. The orientation and the extent of
inclination of the robot body in either direction are measured with inertial-sensor based feedback. The
proposed system uses a combination of first-order spatial filters in order to remove the noise and to merge
the analog sensor readings. Comparative performance analysis is also done between a simple PID
controller and an auto-tuned PID controller for the optimization of attitude control and stabilization of the
self-balancing platform.
Keywords: Attitude control, auto-tuned PID, inertial sensors, filters, PID, self-balancing robot.

where,
1. INTRODUCTION M = moment arm (perpendicular distance between center of
Two-wheeled balancing robots have immense significance in mass and distance from pivot)
g = acceleration due to gravity
the area of robotics and control systems engineering. They
θ = inclination (angle with the vertical)
offer to develop an intricate control system that is capable of
maintaining stability of an otherwise unstable system. This
balancing robotic system imitates the behavior of an inverted
pendulum and in effect works on the same principle as the
Pole and Cart theory. Hence, these principles are taken into
account while designing a robot that is capable of balancing
upright on its two wheels that are aligned on the same axle.
The two wheels are situated below the base and allow the
robot chassis to maintain an upright position by moving in
the direction of tilt, either forward or backward, in an attempt
to keep the centre of the mass above the wheel axles. These
robots are highly non-linear and under-actuated. Since they
are able to balance themselves on only two co-axial
motorized wheels, it is very easy for them to maneuver on
various terrains. Without active control, these systems
become unstable and collapse. Apart from balancing the
posture in a stable upright fashion, they are also able to Fig. 1. Conceptual view of the robot.
regain their posture and stand erect, even when a bounded When θ = 0 degree, the robot is in balanced position and no
external force is applied to them. This force acts as a balancing torque in needed. With θ> 0 or θ< 0, the balancing
disturbance to the system. These robots sense their inclination torque moves the robot in the direction against falling torque.
(rotational pitch angle) continuously, compare it with the set- In this way, the robot tries to retain its balanced position.
point reference provided by the user and correct their Development of a flexible self-balancing robotic platform
orientation by keeping it at the desired pitch angle. The comprises of several essential units. These units include a
system also keeps track of the maximum recovery pitch angle reliable systems model, sensors, signal processors, a stable
(the threshold angular displacement of the robot from the control scheme and actuators. These units have been properly
vertical before it collapses). Inverted pendulum being an discussed by (Nawawi et al., 2007). In the past, the
inherently unstable system tends to fall in either direction. A researchers have extensively used MATLAB toolbox to
conceptual view of the proposed robotic system is shown in efficiently model and control a self-balancing robot (Araghi,
Fig. 1. The balancing torque is given by (1). et al., 2011).Interactive software tools and virtual prototyping
T Mgsin θ (1) techniques, such as ADAMS, can be used to build and
CONTROL ENGINEERING AND APPLIED INFORMATICS 99

simulate a stable mechanical system model (Qian Hao et al., regarding the robot’s orientation. Using the basic
2007). Several non-linear control schemes have been trigonometric relations, this acceleration is used to compute
proposed and their performance has been verified by rigorous angular displacement (θ) of the robot body along with the
experimentation. Neural network controllers have been used direction of tilt, as shown in Fig.3. If one axis (x-axis) is used
in mobile inverted pendulum experiments to control the to calculate the tilted angle of the accelerometer, the
pendulum angle and the position of the cart (S. Jung et al., trigonometry relationship of (2) is used (Analog Devices (a),
2007). Simulation results with PID backstepping control n.d.).
algorithms have proven that with three control loops, this
algorithm can offer a quicker response to balance the two- θ sin (2)
wheeled platform (Nguyen Gia Minh Thao et al., 2010).
where,
State-feedback controllers and Linear Quadratic Regulators Vout= Accelerometer Output (Volt)
(LQR) have also been experimentally validated to provide Voffset= Accelerometer Offset = 1650 mV
robustness for the balance control of a self-balancing robot S = Accelerometer Sensitivity = 800 mV/g
(Solis and Takanishi, 2010; Junfeng and Wanyang, 2011). θ = Angle of Tilt (radians)

2. EXPERIMENTAL SETUP

The voltage signal is the input and the rotational pitch angle
serves as the output. The high level block diagram of the
robot is shown in the Fig.2.

Fig. 3. Single Axis used for Tilt Sensing (Analog Devices (b),
n.d.).

Fig. 4. Complementary Filter for Sensor Integration.

Once these values are read, they are fed directly to the
Fig. 2. High Level Block Diagram. microcontroller (PIC18F452) which initially converts them to
equivalent digital values via the internal 10 bit ADC of PIC.
The inertial sensors (gyroscope and accelerometer) are used The readings are stored in a 16 bit integer variable, namely
to provide analog signals regarding the attitude and “gyro_reading” and “accel_reading”. The samples are taken
orientation of the robot to the microcontroller, PIC18F452 and updated at a regular interval of 10 msec. This helps to
(Microchip, 2006). The microcontroller processes them, ensure a reliable performance of robot.
compares them with the hard-coded equilibrium set-point,
and then issues appropriate motor commands to actuate the 2.2 Signal Conditioning
DC geared motors via the power electronic motor driver
circuit. The accelerometer readings have noise while the gyroscope
readings have an inherent drift. Therefore, before further
2.1 Sensors & Measurements processing, the digital signals corresponding to θ and dθ/dt
are fed to the first order Median-Filter and Mean-Filter, to
The information regarding the orientation and the attitude of remove the random additive noise from the individual sensor
the robot is measured with the aid of analog gyroscope, readings. The simplest procedure was to take 100 samples,
LPR550AL (ST Micro Electronics n.d.), and accelerometer, remove the upper and lower 15 samples and then take the
ADXL335 (Analog Devices (a), n.d.), sensors. The average of the remaining 70 samples. The pseudo-code is as
gyroscope tells us about the rate of change of angle (dθ/dt) of follows.
the robot body in the forward/backward direction. The
accelerometer tells us about the acceleration along the desired 1. START;
axis (d2x/dt2). The purpose of using both of these sensors, 2. TAKE 100 SAMPLES;
3. SORT VALUES IN ASCENDING ORDER;
instead of one, is due to the fact that the accelerometer 4. REMOVE FIRST 15 VALUES;
readings have noise while the gyroscope reading has an 5. REMOVE LAST 15 VALUES;
inherent drift. Hence, in order to overcome the individual 6. SUM REMAINIG 70 VALUES;
short-comings of the two sensors, they are fused 7. DIVIDE BY 70;
appropriately. This feedback provides reliable information 8. END.
100 CONTROL ENGINEERING AND APPLIED INFORMATICS

These filtered values are then combined together via a first The mathematical relationship of PID is shown in (4).
order digital complimentary filter, represented by (3).
u t K e t K e t dt K e t (4)
f a y 1 a x (3)
where,
where,
a = τ / (τ + dt), KP = Proportional Gain
y = gyroscope reading, KI = Integral Gain
x = accelerometer reading, KD = Derivative Gain
f = filtered-output The output of the PID controller is checked continuously, so
A properly implemented filter would combine the raw angle that it may stay between a minimum and maximum value.
with the gravity angle, as shown in Fig.4. It fixes the inherent This helps in avoiding the wind-up state. The tuning of KP,
problems of sensor noise, drift and horizontal acceleration KD, KI constants is usually done with the aid of simulations
dependency. It helps rejecting all the short-term fluctuations. and rigorous experimentation. Now, the variation of the PID
The time interval between successive program loops is control output magnitude between these bounds helps in
known as the Sample Period, dt. The time constant, τ, is the deciding the variation in the duty cycle of the Pulse-Width-
time interval on which it operates on a given signal. So the Modulated (PWM) signal. The resultant PWM signal helps
signals that have time period smaller than this time constant controlling the speed and direction of rotation of the DC
are filtered out, while the longer signals stay unaltered. To geared motors, via an H-Bridge motor driver circuit. This
reduce the gyroscopic drift, a lower time constant should be way, if the robot body tilts in a given direction, the motors
implemented. But on the other hand, this leads to a lot of respond immediately by moving in the direction of
horizontal acceleration noise. Hence a compromise is made inclination at an appropriate speed, in order to bring the
by experimentally tweaking its value and adopting the best wheels exactly below the centre of mass of the robot body.
one for the application. The 'Derivative' term amplifies higher frequency noise that is
generated by the sensors. Thus, the higher values of the
2.3 Closed Loop Control derivatives lead to large changes in the output of the PID
controller. A practical solution that has been adopted to
This digital value is used henceforth for the purpose of remove these high frequency components of noise is by
comparison and correction of robots orientation and attitude putting a first order low pass filter programmatically on the
to put it in its stable upright posture. The filtered output of the derivative term. Consequently, the poles of the derivative
sensors, when the robot body is exactly in stable upright term are tuned such that the noise does not affect the output.
position, is taken as the equilibrium reference or equilibrium
set-point by the control scheme. Once the robot is set into 2.4 Motor Control
action, it continuously checks and compares its current state
with the equilibrium set-point. The difference of these two The motor control is probably the simplest of all the tasks.
entities generates the error signal, e(t). The sign of this error For driving the motor, L298 based dual H-Bridge motor
signal denotes whether the robot is leaning forward (if e(t) > driver circuit is used as shown in Fig.6. It can fully control
0) or backward (if e(t) < 0). The magnitude of the e(t) and drive two motors simultaneously. Also the motors require
specifies the extent to which it has fallen. These error signals, a unique PWM signal. This signal is fed to the motor driver
once computed, are stored. The current error is fed to the P circuit in order to control the speed of the motor rotation. The
controller after being multiplied with KP. The I controller pulse length can be varied to change the speed of the motor.
takes the sum of recent errors. Hence the ten recent errors are Generally, the PWM frequency is about 1000 Hertz, with a
added over the time interval (between successive error period cycle of 1.0 msec.
readings) and sent to the I controller, where they are
2.5 Power Source
multiplied with KI. The rate of change/difference between
two recent errors is subjected to the D controller where they To provide DC power to all the electronic devices explained
are multiplied with KD. Eventually all these three terms are earlier, a DC battery has been utilized.
added and the output u(t) of PID control scheme is obtained,
as illustrated in Fig.5.

Fig. 6. H-Bridge Motor Control Circuit.

A Sealed-Lead-Acid (SLA) battery has been used. The
specifications of the battery are 12V and 1300mAh.Each of
Fig. 5. Closed Loop Control Architecture. the two motors require at most 500mA of current while
CONTROL ENGINEERING AND APPLIED INFORMATICS 101

operating, whereas the sensors and the digital circuit requires

approximately 100mA of current. Since the total current
required by the system is roughly 1100mA, thus the battery
utilized can provide a standby time of 70 minutes, before
discharging completely.

2.6 Robot Structure

The material used in the design is able to offer durability,

strength, maintainability, energy efficiency and operating
capability of the robot. These parameters are also responsible
to contribute in the final size and hence the weight of the
robot. The weight is an important factor in designing the
wheel and base structure of the robot. If the weight applied at
the base is very high, the wheels and the hubs holding them
would bend outward, making it quite difficult to maintain the
robot in upright posture. The total weight of the robot is
1.513 Kg. The robot contains two horizontal plates. One is
very near to the floor level and other is right above it. The
two plates are separated by spacers. The battery of the robot
is installed beneath the lower plate, geometrically placed in
between the co-axial wheels. The power electronic circuitry
is placed right above this lower plate. The upper plate
contains the microcontroller circuitry on it. The computer
aided design of the proposed robot structure is shown in
Fig.7.

Fig. 9. Flow-chart of sub-routines.

3. CONTROL SYSTEM DESIGN
Fig. 7. Computer aided design of the robot.
For a better understanding of the firmware that is responsible
for the control of robots position, it is divided in a number of
sub-routines.

3.1 PID Control

Before implementing the PID routine in software, it is

mandatory to learn a couple of basics. A PID controller has
basically three main components: Proportional controller,
Integral controller and Derivative controller. Each of these
terms is multiplied with a coefficient, namely KP, KD and KI.
The variation in controller gain ‘KP’ has a direct impact at the
robot’s behavior. A higher value of KP usually ensures a
Fig. 8. Fabricated Structure of Robot.
faster controller response. However, a fairly large value of KP
Plexi-Glass of 0.75mm thickness is used to build the chassis leads to undesirable oscillations along with system overshoot.
of the robot because it is strong, durable and light in weight. The derivative term helps speeding up the P controller’s
A larger moment of inertia enhances the static stability. After response to a change of input. Consequently, D controller
experimentation, a mechanically balanced structure is causes the robot to reach the equilibrium state faster. Finally,
fabricated that is easier to control via the embedded system. It the I controller serves to reduce the steady-state error of the P
is shown in Fig.8. The final dimensions of the robot are controller. The steady-state error is made negligible, but the
18.5cm × 10.5cm × 10.0cm. equilibrium state is reached somewhat slowly. The flow chart
of the software routine of PID control is shown in Fig. 10.
The flow of sub-routines is illustrated in Fig. 9.
102 CONTROL ENGINEERING AND APPLIED INFORMATICS

Zeiger-Nicholas Frequency Domain (ZNFD) method. As

shown in Fig.11, the PID controller is being replaced by the
relay right before the plant. While tuning the parameters via
ZNFD method, the KI and KD are made zero, whereas KP is
manually adjusted to a point such that the closed-loop system
starts oscillating in a periodic manner, neither decaying nor
growing in magnitude. This value of KP is recorded as KU.
The time period of these oscillations is measured and
recorded as TU. Then using the mathematical relation in
Table 1, the three parameters are calculated. Unlike the
ZNFD method, the relay method attempts to find the KU and
TU on its own.

Fig. 11. Relay for auto-tuning of PID parameters (Dew,

2014).

Fig. 10. Flow-chart of PID control routine (Microchip, n.d.).

Initially, the PID controller was implemented using the
manual tuning, also known as the simple hand-tuning,
technique. Adjusting the parameters of KP, KD and KI is an
important task. Hence the KP, KD and KI constants are set to
zero in the program. The robot/system is powered up using a
12V power supply. If the base does not move and the robot
topples over freely, it validates that all constants are read as
zeros. The KP constant is gradually increased in small steps, Fig. 12. Input and Output of Relay (Arduino PID Autotune
until there is a little oscillation in the base. As a rule of Library, 2012).
thumb, KI must not exceed 10% of the KP. The KI is then
increased until the robot platform balances itself for a long Once found, this auto-tuner then performs back-calculations
duration, while still oscillating about its mean position. When to tune the parameters. Referring to the Fig. 12, as we start
the KI has been optimized, the robot body will not only be from the steady state, the input to the relay is the error signal,
balanced, but also its oscillations about the equilibrium e(t). The relay outputs a step response. The amplitude of this
position will be small and smooth. Eventually, the KD is step response is ‘D’ and is fixed for a given relay. After each
increased in the similar fashion as the other two constants subsequent zero-crossing of the input, the output step changes
until the platform becomes stable. Once these KP, KD and KI its direction as shown in the Fig. 12. TU is the time period of
co-efficient are found experimentally, they are hard-coded in input signal, whereas the distance between the maxima and
the PID software routine. The routine outputs the direction of minima of the input signal is denoted by ‘A’ as shown in the
the motor and it also calculates the duty cycle of the PWM Fig. 12. The value of A is variable and like TU, it has to be
signal that is to be provided to the motors (Thomas Bräunl, found by the system. Finding the value of A and TU is quite
2008). simple. When the oscillatory input signal is read by the
microcontroller, its peaks are identified. In each sampling
3.2 Automatically Tuned PID Control time frame window, the maximum value is found by reading
and comparing the new value of the input signal with the
Tuning the coefficients KP, KI and KD manually is always a previous largest value, and keeping the larger one as MAX
nuisance. Therefore the industrial applications employ the (Arduino PID Autotune Library, 2012). Similarly the same
auto-tuning feature when using the PID controller. This new value is also compared with the previous smallest value
feature aids in adjusting the three parameters automatically. and the smaller one amongst them is stored as MIN. The
In the proposed robot, the relay method is employed to auto- difference between the MAX and MIN is equal to the value
tune the parameters. Theoretically, it is quite similar to of A, as shown in (5).
CONTROL ENGINEERING AND APPLIED INFORMATICS 103

A MAX MIN (5) overshoot is calculated by using (8). The rise time (TR),
settling time (TS), percentage overshoot (%OS) and steady-
Table 1. PID parameters for the Auto-tuner (Arduino PID
state error (eSS) of the two responses are summarized in Table
Auto-tune Library, 2012)
2.
Control KP KI KD
P 0.5KU - -
PI 0.4KU 0.48KU/TU -
PID 0.6KU 1.2KU/TU 0.075KU/TU
The period TU is found by detecting two zero-crossings and
computing the time between them. For this purpose, the
number of samples between the two zero-crossings is found
and using (6), TU is calculated.
T n T (6)
where,
TS = sampling time Fig. 13. Experimental Setup.
n = number of samples
%OS 100 (8)
Once these KU and TU are found, we next compute the KU via
(7). The PID coefficients are found using the mathematical where,
relations given in Table 1. θMax = Highest peak value of the response in the graph
θSS = Steady-state value of response in the graph
K (7)
π

where,
D = Amplitude of relay output step response
A = peak-to-peak signal value of the input
4. RESULTS
The microcontroller communicates the robot body’s tilt
(pitch) angle with LABVIEW over serial link. The setup used
for testing and recording the simulations results is shown in
Fig. 13. The sampling time used in the application is 10
msec.
Fig. 14. Response with simple PID controller.
Two tests are commenced on the robot to control its attitude
and stabilize it in the upright position. The first test is done
by using a manually-tuned PID controller. The KP, KD and KI
co-efficient are found experimentally to be equal to12.05,
1.075and 0.355respectively. They are hard-coded in the PID
software routine. The step-response of the system is shown in
Fig. 14. The graph is plotted with robot body’s tilt angle
(degrees) and the time (seconds) along the x-axis.
The second test is done by using the automatically tuned PID
controller. The KP, KD and KI co-efficient are found using the
relay method. These coefficients are adjusted automatically.
The step-response of the system is shown in Fig. 15. The
graph is plotted with robot body’s tilt angle (degrees) and the Fig. 15. Response with auto-tuned PID controller.
time (seconds) along the x-axis. It can be clearly seen from Table 2. Summary of robot’s body angle response
the response(s) in Fig. 14 and Fig. 15, that the system
energizes the motors of the robot to oscillate it back and Control Type TR (sec) TS(sec) %OS eSS
forth. Once the robot has gained sufficient energy, the PID 0.30 3.10 13.89% ±6o
balance control scheme(s) implemented via the PID Auto-PID 0.25 1.45 25.10% ±1o
controller and its auto-tuning variant, tend to keep it erect.
5. CONCLUSION
The rise time is experimentally found by looking at the time
required by the response to reach from 10% to 90% of its step The auto-tuning feature saves the trouble of going through
height in Fig. 14 and Fig. 15. Similarly, the settling time is the problematic and redundant task of manually tuning all the
found by observing the time taken by the response to fall PID coefficients, but a comparative analysis of the responses
within ±2% of the steady state value. Finally the percentage in Fig. 14 and Fig. 15, reveals that the overshoot in the
104 CONTROL ENGINEERING AND APPLIED INFORMATICS

system with auto-tuned PID is quite large. However, unlike Dew, ‘Discrete-time PID Controller Implementation,’ 4 May
the manually-tuned PID controllers, the auto-tuned system 2014. [Online]. Available:
shows a smaller steady state error and smaller rise time and http://controlsystemslab.com/category/articles/control-
settling time. The steady state error is reduced by 5o, whereas engineering/pid/. [Accessed May 2014].
the rise time (TR) and settling time (TS) are reduced by 0.05 http://brettbeauregard.com/, ‘Arduino PID Autotune Library,’
sec and 1.65 sec respectively. 28 January 2012. [Online]. Available:
http://brettbeauregard.com/blog/2012/01/arduino-pid-
It is shown in this paper, that application of auto-tuned PID
autotune-library/. [Accessed May 2014].
controllers tend to balance the two-wheeled self-balancing
MicroChip, ‘PIC18FXX2 DataSheet,’ 2006. [Online].
robots in a much more effective manner that the manually-
Available:
tuned PID controllers. Scientifically speaking; with the
http://ww1.microchip.com/downloads/en/DeviceDoc/395
application of the auto-tuned algorithm, the steady state error,
64c.pdf. [Accessed May 2014].
rise time and the settling time of the dynamic system has
Microchip, n.d. ‘Software PID Control of an Inverted
improved. The comparison and hence the improvement in the
Pendulum Using the PIC16F684’. [Online] Available at:
transient as well as the steady state analysis of the system
http://ww1.microchip.com/downloads/en/AppNotes/009
with manually tuned and the auto-tuned controller, also
64A.pdf [Accessed May 2014].
validates the proposed technique. This also manifests that the
Nawawi, S.W., Ahmad, M.N., Osman, J.H.S.
auto-tuned PID controllers can adapt to changes in the
(2007),‘Development of a Two-
physical properties of the robot. That is to say, if the robot’s
Wheeled Inverted Pendulum Mobile Robot,’ IEEE
battery drains over time, the robot will update its PID
Conference Publications, 5th Student Conference
coefficients and would try to cope with the effect of battery
on Research and Development (SCOReD), pp. 1 – 5.
loss to some extent. However in the case of ordinary PID
Nguyen Gia Minh Thao, Duong HoaiNghi, Nguyen HuuPhuc
controller, since the coefficients are hard-coded, the robot
(2010),‘A PID Back-stepping controller for two-wheeled
would get unstable and collapse under such circumstances.
self-balancing robot,’IEEE Conference Publications,
Similarly if the mass of the robot is changed during
International Forum on Strategic Technology (IFOST),
operation, the PID coefficients of system would adjust
pp. 76 – 81.
themselves automatically in order to continue stabilizing. The
Qian Hao, Liping Chen, WeiweiQiao, Peng Li, Songling
relay method for auto-tuning provides us with the ease of
Yang, Qifang Liu (2011),‘Controlling Simulation Study
implementation and great flexibility in usage. It gives us
On Two-Wheeled Self-Balancing Electrical Motorcycle
fairly good values of KP, KD and KI to balance and control the
Based on ADAMS And MATLAB,’IEEE Conference
attitude of our robotic system. Thus, we are able to balance a
Publications, Cross Strait Quad-Regional Radio
two-wheeled platform in a very effective and an innovative
Science and Wireless Technology Conference
way. However, in exchange, some performance sacrifices are
(CSQRWC), pp. 1704 – 1707.
made such as excessive overshoot.
Seul Jung, H. T. Cho, T. C. Hsia (2007), ‘Neural network
There is still a lot of room for further research and control for position tracking of a two-axis inverted
enhancements that can improve the performance of this pendulum system: Experimental studies,’IEEE
platform. Instead of using the manually-tuned PID and the Transaction on Neural Networks, vol. 18, no.4, pp. 1042-
auto-tuned PID controllers, adaptive fuzzy PID controller can 1048.
be used to tune the KP, KD and KI co-efficient in real time. Solis, J., Takanishi, A. (2010),‘Development of a Wheeled
Although the adaptive fuzzy PID control technique would be Inverted Pendulum Robot and a Pilot Experiment with
computationally expensive and slow. But being more Master Students,’IEEE Conference Publications, 7th
sensitive to the changes it would yield a much better dynamic International Symposium on Mechatronics and its
performance for the stability and balance control of the two- Applications (ISMA), pp. 1 – 6.
wheeled self-balancing robot. STMicroElectronics, ‘LPR550AL,’ [Online]. Available:
http://www.starlino.com/wp-
REFERENCES content/uploads/data/acc_gyro/LPR550AL.pdf.
Analog Devices (a), ‘AN-1057 Application Note,’ [Online]. [Accessed June 2014]
Available: www.analog.com/static/imported- Thomas Bräunl, Embedded Robotics: Mobile Robot Design
files/application_notes/AN-1057.pdf. [Accessed May and Applications with Embedded Systems, Springer, 3rd
2014]. ed., 2008
Analog Devices (b), ‘Small, Low Power, 3-Axis ±3 g Wu Junfeng, Zhang Wanying (2011),‘Research on Control
Accelerometer ADXL335,’ [Online]. Available: Method of Two-wheeled Self-
www.analog.com/static/imported- balancing Robot,’IEEEConference Publications,
files/data_sheets/ADXL335.pdf. [Accessed May 2014]. Intelligent Computation Technology And Automation
Araghi, M.H., Kermani, M.R. (2011), ‘Computer-Aided (ICICTA), pp. 476 – 479.
System Design for Educational Purposes: An
Autonomous Self-Balancing Two-Wheeled Inverted
Pendulum Robot,’ IEEE Conference Publications, 24th
Canadian Conference on Electrical and Computer
Engineering (CCECE), pp. 001357- 001360.

View publication stats