University of South Carolina

Scholar Commons

Theses and Dissertations

5-2017

Dynamic Model and Control of Quadrotor in the Presence of

Uncertainties
Courage Agho
University of South Carolina

Follow this and additional works at: https://scholarcommons.sc.edu/etd

Part of the Electrical and Computer Engineering Commons

Recommended Citation
Agho, C.(2017). Dynamic Model and Control of Quadrotor in the Presence of Uncertainties. (Master's
thesis). Retrieved from https://scholarcommons.sc.edu/etd/4069

This Open Access Thesis is brought to you by Scholar Commons. It has been accepted for inclusion in Theses and
Dissertations by an authorized administrator of Scholar Commons. For more information, please contact
[email protected].
DYNAMIC MODEL AND CONTROL OF QUADROTOR IN THE
PRESENCE OF UNCERTAINTIES

by

Courage Agho

Bachelor of Science
University of South Carolina, 2015

_________________________________________________

Submitted in Partial Fulfillment of the Requirements

For the Degree of Master of Science in

Electrical Engineering

College of Engineering and Computing

University of South Carolina

2017

Accepted by:

Xiaofeng Wang, Director of Thesis

Bin Zhang, Reader

Cheryl L. Addy, Vice Provost and Dean of the Graduate School

© Copyright by Courage Agho, 2017
All Rights Reserved.

ii
 ACKNOWLEDGMENTS

I would like to thank Professor Xiaofeng Wang who is my adviser for giving me

this opportunity to work on this thesis. I thank him for his support and time invested in my

education. I also want to thank the department, my friends and family for their constant

support.

iii
 ABSTRACT

This thesis considers the control of quadrotor using a linear PID control and ℒ1

adaptive control. In a justifiable concept, PID controller can be used to control a quadrotor,

but in the presence of uncertainties or disturbance, the quadrotor can’t be automatically

adjusted to control the changing dynamics of the quadrotor. To solve the problem

associated with uncertainties, various control methodology can be used for controlling the

changing dynamic of quadrotor, but in this thesis, ℒ1 adaptive control is used because it

allows for fast and robust adaptation for desired transient performance in the presence of

matched and unmatched uncertainties.

In this thesis, we would derive the quadrotor model which gives us an access on

how we can track positions, then design a controller to track these desired positions using

PID control. Same concept used for PID control would be used for ℒ1 adaptive control in

chapter 5 except this control methodology is used for the cancellation of uncertainty at a

faster and robust manner. ℒ1 Adaptive control would use state feedback for its position

control and this position control is then used to design the final controller to cancel

uncertainties.

iv
 TABLE OF CONTENTS

ACKNOWLEDGEMENTS................................................................................................iii

ABSTRACT.......................................................................................................................iv

LIST OF FIGURES...........................................................................................................vii

CHAPTER 1: INTRODUCTION…………........................................................................1

CHAPTER 2: QUADROTOR STRUCTURE ....................................................................3

2.1 QUADCOPTER COORDINATE SYSTEM ………………......…………3

2.2 STATES OF THE QUADCOPTER ………………..…………….....…....4

2.3 HOW QUADROTOR WORKS ……...………..…….……………...........4

CHAPTER 3: DYNAMIC MODEL…………....................................................................6

3.1 MOMENT ACTION ON THE QUADROTOR …………...…………......6

3.2 INERTIAL MATRIX ……………...……...…………...............................7

3.3 CONTROL INPUT VECTOR U …………………….………...................8

3.4 ROTATIONAL SUBSYSTEM ……………...….……..............................8

3.5 TRANSLATION SUBSYSTEM …………………...…..…………...……9

CHAPTER 4: CONTROL OF THE QUADROTOR…....................................................11

4.1 TRANSLATION CONTROL ALONG Z-AXIS …………………...…...12

4.2 TRANSLATION CONTROL ALONG X-AXIS ………….……............13

4.3. TRANSLATION CONTROL ALONG Y-AXIS ………....……...….….14

4.4. ROTATION CONTROL ABOUT Z-AXIS ……….………………..…..15

v
CHAPTER 5: ℒ1 ADAPTIVE CONTROL ……………………………………….…….17

5.1 DIRECT & INDIRECT ADAPTIVE CONTROLLER ……….…...........18

5.2 ℒ1 ADAPTIVE CONTROLLER ARCHITECTURE …….......................19

5.3 CONTROLLER DESIGN WITH ℒ1 ADAPTIVE CONTROLLER.…..21

5.4 TRANSLATION CONTROL ALONG Z-AXIS …………....…………..21

5.5 ROTATION CONTROL ABOUT THE Z-AXIS …………………...…..27

5.6 TRANSLATION CONTROL ALONG THE X-AXIS ………..……......29

5.7 TRANSLATION CONTROL ALONG THE Y AXIS …..……..…….....38

CHAPTER 6: CONTROLLER SIMULATION WITH

NON-LINEAR MODEL …………………...…………........................................41

CHAPTER 7: SUMMARY AND CONCLUSION…………………………………...…45

REFERENCES …………………………………………………………….....................47

vi
 LIST OF FIGURES

Figure 2.1: Body and Earth Frame with their Corresponding Angle …………..…..……..3

Figure 2.2: Movement around Angles on the Coordinates

System because of Force …………………………………..………………….5

Figure 3.1: Moment acting on the Quadcopter …………………...……………..……......7

Figure 4.1: Altitude Position Tracking Using PID Controller [Z-Axis] ……..……….....13

Figure 4.2: Block Diagram for Translational Control ……………..…………………....14

Figure 4.3: Roll Position Tracking Using PID Controller [X-Axis] ………..………..….14

Figure 4.4: Pitch Position Tracking Using PID Controller [Y-Axis] ………..…….........14

Figure 4.5: Yaw Angle Tracking Using PID Controller [Z-Axis] …………..…………..15

Figure 4.6: Controller Gain Constant ………………………………….…………..…….15

Figure 4.7: Altitude Response to Step Input Disturbance ……………..……..……….....16

Figure 5.1: Closed Loop Architecture for Direct MRAC ………………..……………...18

Figure 5.2: Closed Loop Architecture for Indirect MRAC ……………..…………....….19

Figure 5.3: Closed Loop Architecture for ℒ1 Adaptive controller …………..……….....21

Figure 5.5: Altitude Position Using State Feedback Control [Z-Axis] ………….....……23

Figure 5.6: State Feedback for Altitude Control ………………………..…….….……...24

Figure 5.7: Altitude Position Tracking with Response to Varying Uncertainties

and Extra Mass [Z-Axis]………………………..…………………………....26

Figure 5.8: Structural Design for Altitude Control …………………..…………..……...26

vii
Figure 5.9: Estimate Uncertainty vs Real Uncertainty for Altitude Control …...….........26

Figure 5.10: Yaw Angle Tracking Using State Feedback Control.………………….......27

Figure 5.11: Tracking of Desired Reference Yaw Angle with

Varying Input Uncertainties …………………………………...…...….......29

Figure 5.12: Estimated Uncertainty vs Real Uncertainty for Yaw Control …………......29

Figure 5.13: State Feedback Design in Simulink ……………………………………..…31

Figure 5.14: Roll Position Tracking without Varying Input Uncertainties ……………..31

Figure 5.15: Pitch Angle at a Desired Position of 1.4m …………………………...........32

Figure 5:16 Pitch Angle at a Desired Position of 25m …………………………….........32

Figure 5.17: Pitch Angle with Desired Position of 25m ………………………..……….33

Figure 5.18: Pitch Angle with Desired Position of 1.4m ……………………………..…33

Figure 5.19: Tracking of Desired Position of 1.4m with Pitch Angle Limitation ............34

Figure 5.20: Tracking of Desired Position of 25m with Pitch Angle Limitation .............34

Figure 5.21: Roll Position Tracking with Varying Matched and Unmatched
Uncertainties without Compensation for Unmatched Uncertainties ...........36

Figure 5.22: Roll Position Tracking with Varying Matched Uncertainties ....…..…..…..37

Figure 5.23: Structural Design for Roll Control ………………………………...............37

Figure 5.24: Estimate Uncertainty vs Real Uncertainty for Roll Control .........................37

Figure 5.25: Estimate Uncertainty vs Real Uncertainty for Unmatched

Uncertainties Roll Control ...........................................................................38

Figure 5.26: Roll Position Tracking with Varying Matched and

Unmatched Uncertainties …………………...…………………..………....38

Figure 5.27: Feedback Response without Disturbance ……..…......................................39

Figure 5.28: Pitch Position Tracking with Varying Matched

and Unmatched [Y-Axis] ………...……………………………….…....…..39

Figure 5.29: Estimate Uncertainties vs Real Uncertainties for Matched

viii
 Uncertainties Pitch Control ………………..……………………...…….....39

Figure 5.30: Estimate Uncertainty vs Real Uncertainty for Unmatched

Uncertainties Pitch Control ……………………...……………..…...……..40

Figure 6.1: Disturbance Limit …………………………………………………………...41

Figure 6.2: Altitude Position Tracking Using ℒ1 Adaptive Controller [Z-Axis] .............42

Figure 6.3: Yaw Position Tracking Using ℒ1 Adaptive Controller [Z-Axis] ………..….42

Figure 6.4: Roll Position Tracking Using ℒ1 Adaptive Controller [X-Axis] …...............42

Figure 6.5: Pitch Position Tracking Using ℒ1 Adaptive Controller [Y-Axis] …..............43

Figure 6.6: Simulink Design with Uncertainties Input ……………...……..………..…..43

Figure 6.7: Expanded Simulink Design ………...……………………………..…….......44

ix
 CHAPTER 1

INTRODUCTION

In recent year, research on quadrotors has become popular, but the first quadrotor

was built in 1907 by Louis Charles Bréguet. Helicopter development was desired over

quadrotor, but with recent progress in technological development, more funds and time has

been dedicated to the research of quadrotors. Quadrotor became popular because helicopter

uses tail rotors to counterbalance the torque or rotating forces generated by the single main

rotor. Because of the counterbalancing tail rotor, it inefficient in term of control, power

consumption and cost production. Due to the work of Charles Richet and Dr George de

Bothezat in 1956, propellers where used to control the quadcopter roll, pitch and yaw angle.

In addition to those improvements, technological advancement in batteries weight, life and

density has greatly helped the research of quadrotors. Recent development in processors

and low cost efficient sensors has greatly impacted quadrotor development as well.

Common application of quadrotors includes surveillance, inspections, military

operation and transportation. It is very applicable because of its size; it is easier for it take

off, to land and occupies less space. Because of its advantages, the demand for accurate

control is needed for stabilized flight when hovering. The demand for accurate stabilized

flight has led for the use of sensors and camera to achieve accurate result. Although PID

controller is commonly used for control of quadrotors, however due to uncertainties and

disturbance that occur during flight, PID controller is not a good mechanism for control.

1
 This thesis focuses on creating an accurate control mechanism to achieve stabilized

flight control using ℒ1 adaptive control because it mitigates the issues of uncertainties and

disturbance.

A quadrotor is a nonlinear system device and control for nonlinear system is

complex because we desire a system that has fast adaptation and response in real time and

robust enough to mitigate the issue of disturbance and uncertainties. Adaptive control was

developed as a technique for automatic adjustment in real time. To achieve and maintain

desired system performance, the aerospace industries and institution started research on

adaptive controller and today it is used widely by different industries and for different

purposed. Model reference adaptive control (MRAC) is a type of adaptive control and ℒ1

adaptive control is a further use of model reference adaptive control (MRAC). Apart from

canceling of uncertainties and disturbance, ℒ1 adaptive control theory has architecture in

place for faster adaptation that is decoupled from robustness. ℒ1 Adaptive control doesn’t

losses robustness, because it is resolved by conventional classical control and ℒ1 adaptive

control system guarantees robustness in the presence of fast adaption with high gain

feedback. This thesis includes the derivation of quadrotor model, and the model is then

controlled by a PID controller and an ℒ1 adaptive controller in the presence of uncertainty.

2
 CHAPTER 2

QUADROTOR STRUCTURE

2.1 QUADCOPTER COORDINATE SYSTEM

A quad employs different control mechanism such as roll, pitch, and yaw which in

most cases are represented by angle of rotation around the center of the quad craft. These

angles make up for the control of the altitude of the quadcopter and to track the altitude of

the quadcopter, a two-coordinate system is required. There is the body frame system which

is attached to the quad at its center of gravity and the earth frame system which is fixed to

the earth and it is sometimes refer to as an inertial coordinate system. The angular different

between the two coordinate helps define the behavior of the quad altitude in space. The

attitude system can be derived by rotating the body frame around the z axis of the earth

frame by the yaw angle 𝜑, which is tend followed by rotating around the y-axis by the pitch

angle 𝜙 and finally by rotating around the x-axis by the roll angle 𝜃. This is shown in figure

2.1 as well as its rotation matrix that has the body and earth frame parallel to each other

and their sequence of rotation is known as the Z-Y-X rotation and its rotation matrix is

shown in equation 2.1.

Figure 2.1: Body and Earth Frame with their Corresponding Angle

3
 cos(𝜃) sin(ψ) sin(ϕ) sin(𝜃) cos(ψ) cos(ϕ) sin(𝜃) cos(ψ) + sin(ϕ) sin(ψ)
𝑅 = [cos(𝜃) sin(ψ) sin(ϕ) sin(𝜃) sin(ψ) + cos(𝜃) cos(ψ) cos(ϕ) sin(𝜃) sin(ψ) − sin(𝜃) cos(ψ) ] (2.1)
−sin(𝜃) sin(ϕ) cos(𝜃) cos(ϕ) cos(𝜃)

2.2 STATES OF THE QUADCOPTER

From the section of coordinate system, the angle of roll, pitch, and yaw are

represented as 𝜙, 𝜃, 𝜓 in addition to their angle, angular velocity is also required and can

be represented as 𝜙̇, 𝜃̇, 𝜓̇ these are the first six state of the quadcopter that shows a

relationship between the quadcopter and the earth coordinate system. The next six states

show a physical relationship of the physical location within the earth fixed system and it is

denoted as 𝑋, 𝑌, 𝑍. In addition to their physical position is their quad velocity along these

axes and it is denoted as 𝑋,̇ 𝑌,̇ 𝑍̇. Together they make up the 12 states of the quadcopter and

as shown below.

2.3 HOW QUADROTOR WORKS

A quad as we know it has four motors and it is important to know that the thrust of

each motor force a change around the pitch, roll, and yaw angle. To understand that, it is

very important to know that a quadcopter is under actuated which means that the six degree

of freedom (𝜙, 𝜃, 𝜓, 𝑋, 𝑌, 𝑍) is only controlled by the four inputs. Two of the six DOF are

couple and they are the x-axis and y-axis on the translational size of the quadcopter. The

translational part of the quadcopter is dependent on the attitude of the craft with respect to

the other four degrees of freedom. More would be discussed on how to account and control

the under actuated part of the quad when designing a control method for the quad. It is also

important to know that a quadcopter has four motor where two of them spins in the

clockwise direction while the order two spins in the counter clockwise direction and if the

thrust generated by the motors is equal, the ability for the quadcopter to roll, hover pitch,

4
and yaw is possible. The diagram in figure 2.2 shows how the force generated by each

motor affects a change the direction of the quadcopter along the angles on the coordinate

system

Figure 2.2: Movement around Angles on the Coordinates System because of Force

5
 CHAPTER 3

DYNAMIC MODEL

The dynamic model of a quadcopter is a subsystem that is divided into two

subsystems known as the rotational subsystem that represent (roll angle, pitch angle and

yaw angle) and translational subsystem that represent (Z position, X position, Y position).

The rotational side of the Quadcopter is completely actuated while the translational side of

the subsystem is under actuated

3.1 MOMENT ACTION ON THE QUADROTOR

An effect of rotation is the force generated called aerodynamic force and moment

generated called aerodynamic moment. The aerodynamic moment is the combination of

aerodynamic force multiplied by its distance. It is dependent on the geometry of the

propeller and by identifying the moment and force generated by the propeller, we can

understand the moment acting on the quadcopter. From figure 3.1, when F2 is multiplied

by the moment arm, a negative moment is generated about the y-axis. Using the same

concept, F4 generates a positive moment and the total moment about the x-axis can be

expresses as

𝑀𝑋 = (𝐹4 − 𝐹2)𝑙 (3.1)

Same concept is applied for My Where

𝑀𝑦 = (𝐹1 − 𝐹3)𝑙 (3.2)

6
For the moment about the Z axis, the thrust of the rotor does not cause a moment, but rather

the rotor rotation in relation with the rotor speed causes a moment and is represented as

𝑀𝑧 = (𝐹1 + 𝐹3 − 𝐹2 − 𝐹4)𝑙𝑐 (3.3)

Where c gives the relationship between the rotor speed and its effect on the quadrotor

rotation about the body frame and the combination of each body frame axis gives us a

matrix as shown below.

(𝐹4 − 𝐹2)𝑙
𝑀𝑏 = [ (𝐹1 − 𝐹3)𝑙 ] (3.4)
(𝐹1 + 𝐹3 − 𝐹2 − 𝐹4)𝑙𝑐

Figure 3.1: Moment acting on the Quadcopter

3.2 INERTIAL MATRIX

The inertial matrix for a quadcopter is a diagonal matrix as shown below. The

structure of the matrix is because quadcopters are built symmetrically with respect to the

coordinate systems that were explained in section 2.1.

𝐽𝑥 0 0
𝐽 = [0 𝐽𝑦 0] (3.5)
0 0 𝐽𝑧

From the matrix above 𝐽𝑥 , 𝐽𝑦 and 𝐽𝑧 are the area moments of inertia about the principle axes

on the body frame.

3.3 CONTROL INPUT VECTOR U

7
The control input is from the controller and the input to a quadcopter is from the force

generation by the motor, and for simplicity, Control Input Vector (U) can represent Force

(F) where

𝑈𝑡𝑧 = (𝐹1 + 𝐹2 + 𝐹3 + 𝐹4) (3.6)

𝑈𝑡𝑥 = (𝐹4 − 𝐹2) (3.7)

𝑈𝑡𝑦 = (𝐹1 − 𝐹3) (3.8)

𝑈𝑟𝑧 = (𝐹1 + 𝐹3 − 𝐹2 − 𝐹4) (3.9)

From equation 3.4, the moment acting on the quadcopter in its body frame can be

represented as

𝑈𝑡𝑥 𝑙
[ 𝑈𝑡𝑦 𝑙 ] (3.10)
𝑈𝑟𝑧 𝑙𝑐

3.4 ROTATIONAL SUBSYSTEM

The rotational part of the quadcopter is derived from the concept of rotational

equation of motion and by using Newton-Euler method derived from the body frame of the

quadcopter with a generalized formula as shown below.

𝑀𝑏 = (𝐽𝑤̇ + 𝑤×𝑗𝑤 + 𝑀𝑔 ) (3.11)

Where: 𝐽 represents quadrotor inertia Matrix, 𝑤 represents angular velocity, 𝑀𝑔 represents

the gyroscopic moment generated due to its rotor inertial and 𝑀𝑏 represents moments

acting on the quadcopter in its body frame. For simplicity, the gyroscopic moment would

not be considered because the inertial generated by the quadcopter is much larger that the

inertial generated by the rotor. So, our final rotor equation would be

𝑀𝑏 = (𝐽𝑤̇ + 𝑤×𝑗𝑤) (3.12)

From equation 3.12, we would have the expression below

8
 𝑈𝑡𝑥 𝑙 𝐽𝑥 0 0 𝜙̈ 𝜙̇ 𝐽𝑥 0 0 𝜙̇
[ 𝑈𝑡𝑦 𝑙 ] = [ 0 𝐽𝑦 0 ] [ 𝜃̈ ] + [ 𝜃̇ ] × [ 0 𝐽𝑦 0 ] [ 𝜃̇ ]
𝑈𝑟𝑧 𝑙𝑐 0 0 𝐽𝑧 𝜓̈ 𝜓̇ 0 0 𝐽𝑧 𝜓̇

When the matrix is rewritten to have its angular acceleration, we would have

𝑙 𝐽𝑦 𝐽𝑧
𝜙̈ = 𝑈𝑡𝑥 + 𝜓̇𝜃̇ − 𝜃̇𝜓̇ (3.13)
𝐽𝑥 𝐽𝑥 𝐽𝑥

𝑙 𝐽𝑧 𝐽𝑥
𝜃̈ = 𝑈𝑡𝑦 + 𝜓̇𝜙̇ − 𝜙̇𝜓̇ (3.14)
𝐽𝑦 𝐽𝑦 𝐽𝑦

𝑙𝑐 𝐽𝑥 𝐽𝑦
𝜓̈ = 𝑈𝑟𝑧 + 𝜃̇𝜙̇ − 𝜙̇𝜃̇ (3.15)
𝐽𝑧 𝐽𝑧 𝐽𝑧

3.5 TRANSLATION SUBSYSTEM

The translation subsystem is based on translational equation of motion and it is based on

newton second law which is derived from the earth inertial frame and it is presented in the

format below.

0
̈
𝑚𝑑 = [ 0 ] + 𝑅×𝐹𝑛𝑔 (3.16)
𝑚𝑔

Where: m is the mass, g is gravitational acceleration, 𝐹𝑛𝑔 is non-gravitational force which

is the physical location within the earth fixed system and R is the rotational matrix. 𝐹𝑛𝑔 is

shown in equation 3.17. It is the addition of all the thrust force produced by the four

propellers and the negative sign is because the thrust force generated is acting upward while

the z-axis of the body frame is point down. R is for the rotational matrix that is generated

to transform the forces generated from the body frame to the earth frame. By substitution

and simplification, we would derive the acceleration along the X, Y, and Z axis.

0
𝐹𝑛𝑔 = [ 0 ] (3.17)
−𝑈𝑡𝑧

9
By substitution of these equations into equation 3.16, we would get

𝑋̈ 0 cos(𝜃) sin(ψ) sin(ϕ) sin(𝜃) cos 𝜓 − cos(ϕ) sin(𝜃) cos(ψ) + sin(ϕ) sin(ψ) 0
𝑚 [𝑌̈ ] = [ 0 ] + [cos(𝜃) sin(ψ) sin(ϕ) sin(𝜃) sin(φ) + cos(𝜃) cos(ψ) cos(ϕ) sin(𝜃) sin(ψ) − sin(𝜃) cos(ψ) ] × [ 0 ]
𝑍̈ 𝑚𝑔 −sin(𝜃) sin(ϕ) cos(𝜃) cos(ϕ) cos(𝜃) −𝑈𝑡𝑧

When the matrix is rewritten to have its acceleration, we would have

−𝑈𝑡𝑧
𝑋̈ = (− cos(𝜙) sin(𝜃) cos(𝜓) + sin(𝜙) sin(𝜓)) (3.18)
𝑚
−𝑈𝑡𝑧
𝑌̈ = (cos(𝜙) sin(𝜃)sin(𝜓) + sin(𝜃) cos(𝜓)) (3.19)
𝑚
−𝑈𝑡𝑧
𝑍̈ = (cos(𝜙) cos(𝜃)) − 𝑔 (3.20)
𝑚

The quadrotor parameter are defined below and these parameters would be use for

simulation
𝑁
𝑚 = 0.8𝑘𝑔; 𝑙 = 0.25𝑚; 𝑔 = 9.81 𝑘𝑔; 𝑐 = 0.02; 𝐽𝑥 = 𝐽𝑦 = 0.015𝑘𝑔𝑚2 ; 𝐽𝑧 = 0.02𝑘𝑔𝑚2

10
 CHAPTER 4

CONTROL OF THE QUADROTOR

Building a control system to control how the quadcopter operate is very important.

The dynamic model of the quadcopter is an open loop and closed loop control is a preferred

method for any design because the system records the output instead of the input and

modifies its output per a preferred condition. For successful maneuvering of the

quadcopter, we need 4 controllers that would be designed to act as in input to the model of

the quadcopter. These 4 controllers represent the 4-input coming from the transmitter

assuming this was a physical quadcopter. They are throttle, roll, pitch, and yaw input. The

throttle coming from the transmitter can be represented and called the altitude controller or

translational controller in the z axis. The roll, pitch, and yaw input from the transmitter can

be called rotational controller that is dependent on the angle ϕ, θ, 𝜑. The translational

subsystem of the model is partially dependent on the rotational subsystem because it is

under-actuated.

The dependent axes on the translational controller are the X, Y axis. So, a controller

must be built that takes in desired X, Y values and produce an angular desired output that

would be sent to rotational controller. For the control of the quadrotor using PID controller,

the dynamic model of the system would be linearized using small angle approximation.

The equation below represents the linear dynamic model of the quadrotor and an in-depth

concept on how the model is linearized is explained in chapter 5.

11
 1
𝑍̈ = (−𝑈𝑡𝑧 ) (4.1)
𝑚

1
𝑋̈ = − 𝜃(𝑈𝑡𝑧 ) (4.2)
𝑚
1
𝑌̈ = 𝜙(𝑈𝑡𝑧 ) (4.3)
𝑚

𝑙
𝜓̈ = 𝑈 (4.4)
𝐽𝑧 𝑟𝑧

𝑙
𝜃̈ = 𝑈 (4.5)
𝐽𝑦 𝑡𝑥

𝑙
𝜙̈ = 𝑈 (4.6)
𝐽𝑋 𝑡𝑦

4.1 TRANSLATION CONTROL ALONG THE Z AXIS

The translational controller takes an error signal as an input which is the difference

between the actual altitude and desired altitude and produces a control signal 𝑈𝑡𝑧 . The

control signal 𝑈𝑡𝑧 is responsible for the altitude of the quadcopter and mathematical

equation is shown below. The equation below uses the concept of a PID controller and for

simulation of result, we would multiply the input data by (-1) to compensate for the

negative sign in equation 4.1 and the controller response to the closed loop simulation is

shown in figure 4.1.

𝑑𝑧
𝑈𝑡𝑧 = 𝐾𝑝(𝑧 − 𝑧𝑑) + 𝐾𝑖 ∫(𝑧 − 𝑧𝑑) + 𝐾𝑑 (𝑧 − 𝑧𝑑) (4.7)
𝑑𝑡

12
 Figure 4.1: Altitude Position Tracking Using PID Controller [Z-Axis]

4.2 TRANSLATION CONTROL ALONG THE X AXIS

The translational controller takes an error signal as an input which is the difference

between the actual roll angle and the desired roll angle to produce a control signal 𝑈𝑡𝑥 . The

control signal 𝑈𝑡𝑥 is responsible for moving the quadcopter left and right about the X axis.

The mathematical equation is shown below. The equation below uses the concept of a PID

controller and because translation along the X axis is dependent on 𝜃, control of its desired

position is not under a direct control.

𝑑𝑧
𝑈𝑡𝑥 = 𝐾𝑝(𝑥 − 𝑥𝑑) + 𝐾𝑖 ∫(𝑥 − 𝑥𝑑) + 𝐾𝑑 (𝑥 − 𝑥𝑑) (4.8)
𝑑𝑡

This is because the X positions are under the translational subsystem which is under-

actuated. To compensate for the under actuation of the system, the pitch angle can be used

to control desired X position.

We would assume 𝑈𝑡𝑧 = 𝑚𝑔 which would cause loss in precision, but it also simplifies the

model for easier control. Equation 4.2 is then simplified as

𝑋̈ = −𝜃𝑔 (4.9)

The rotational control provides us 𝜃 which can then be used for translational control. For

simulation of result, we would multiply the input data by (-1) to compensate for the

13
negative sign in equation 4.2 and the controller response to the closed loop simulation is

shown in figure 4.5.

Figure 4.2: Block Diagram for Translational Control

Figure 4.3: Roll Position Tracking Using PID Controller [X-Axis]

4.3 TRANSLATION CONTROL ALONG THE Y AXIS

𝑌̈ = 𝜙𝑔 (4.10)

The design for controller movement along the Y axis is design in similar method for

translation along the X axis. The only difference is the Y axis is dependent on the roll angle.

Figure 4.4: Pitch Position Tracking Using PID Controller [Y-Axis]

4.4 ROTATION CONTROL ABOUT Z-AXIS

14
 The yaw controller takes an error signal as an input which is the difference between

the actual yaw angle and the desired yaw angle to produce a control signal 𝑈𝑟𝑦 and the

mathematical equation is shown below. The equation below uses the concept of a PID

controller and the block diagram represent how the close loop method of the yaw controller

would look like.

𝑑𝑧
𝑈𝑟𝑦 = 𝐾𝑝(𝜓 − 𝜓𝑑) + 𝐾𝑖 ∫(𝜓 − 𝜓𝑑) + 𝐾𝑑 (𝜓 − 𝜓𝑑) (4.11)
𝑑𝑡

Figure 4.5: Yaw Angle Tracking Using PID Controller [Z-Axis]

From the design concept above, 6 PID controllers were designed with Kp, Ki, and Kd gain

derived. Where the controller design for translational control on the X and Y axis would

have two PID controller and their gain constants are shown in figure 4.6.

Kp (gain) Ki (gain) Kd (gain)

Translation Control along the Z Axis 0.4001 0.01019 3.491

Translation Control along the Y Axis 0.0366 0.0008775 0.3392

Translation Control along the X Axis 0.0366 0.0008775 0.3392

Rotation Control along the Z Axis 0.21 0.01225 0.7997

Rotation Control along the Y Axis 1.553 0.8375 0.5677

Rotation Control along the X Axis 1.553 0.8375 0.5677

Figure 4.6: Controller Gain Constant

15
Figure 4.7: Altitude Response to Step Input Disturbance

16
 CHAPTER 5

ℒ1 ADAPTIVE CONTROL

As you can see from section 4, using a conventional linear controller work fine

when there is no disturbance or uncertainty, but when disturbance is added to the system

as shown in figure 4.10, the controller is not able to adapt fast enough to compensate for

the additional error. The process of creating an adaptation mechanism leads to further

insight to the concept of adaptive controller.

Adaptive controller is a control mechanisms used by a controller to control a system

with varying parameters or uncertainties. It is a technique used for the automatic

adjustment of the controller in real time. Various controllers can be used to cancel

uncertainties; the most common one is the robust controller. The concept of adaptive

controller to robust controller is the control law for adaptive controller change while the

control law for robust controller doesn’t change. Adaptive controllers are the combination

of online parameter estimators and automatic control design which has two architecture

designs. There is direct method which only estimates the controller parameters and there is

indirect method which only estimates the state parameter of system. The structure of this

architecture is shown below. From the structures, the estimate is calculated using adaptive

law where Γ is the adaptive gain.

17
5.1 DIRECT & INDIRECT ADAPTIVE CONTROLLER

The architecture is shown in figure 5.1 and the differential equation of the real plant is

𝑥̇ (𝑡) = 𝐴𝑚 𝑥(𝑡) + 𝑏(𝑢(𝑡) + 𝑘𝑥𝑇 𝑥(𝑡), 𝑥(0) = 𝑥0 (5.1)

𝑦(𝑡) = 𝑐 𝑇 𝑥(𝑡) (5.2)

Where 𝐴𝑚 defines matrix of the closed loop system, x(t) is the state of the system, b and

c are known constant; 𝑘𝑥 is the vector of the unknown constant.

𝑢𝑛𝑜𝑟𝑚 (𝑡) = −𝑘𝑥𝑇 𝑥(𝑡) + 𝑘𝑔 𝑟(𝑡) (5.3)

1
𝑘𝑔 ≜ (5.4)
𝑐 𝑇 𝐴𝑚
−1 𝑏

Ideal system

𝑥̇ 𝑚 (𝑡) = 𝐴𝑚 𝑥(𝑡) + 𝑏𝑘𝑔 𝑟(𝑡), 𝑥𝑚 (0) = 𝑥0 (5.5)

𝑦𝑚 (𝑡) = 𝑐 𝑇 𝑥(𝑡) (5.6)

𝑒(𝑡) ≜ 𝑥𝑚 (𝑡) − 𝑥 (5.7)

Adaptation Law becomes

k̂̇ x = −Γ𝑥(𝑡)𝑒̃ 𝑇 (𝑡)𝑃𝑏, 𝑘̂𝑥 (0) = 𝑘𝑥0 (5.8)

And P solves Lyapunov equation

𝐴𝑇𝑚 𝑃 + 𝑃𝐴𝑚 = −𝑄 (5.9)

Figure 5.1: Closed Loop Architecture for Direct MRAC

18
For stability test and analysis, Lyapunov function is used to test Lyapunov stability. By

taking the first derivative of Lyapunov function, the signal stays bounded and the second

derivative proves that the error converges when 𝑒(𝑡) ⟶ 0 𝑎𝑠 𝑡 ⟶ ∞ which in turns

proves the application of Barbalat’s lemma equation on stability of time varying system.

The theoretical approach used for direct MRAC is used as well for Indirect MRAC. The

main difference is that the indirect method estimates the system parameters and the

derivation of the adaptive law is independent of the control signal with its system

architecture as shown in figure 5.2. The asymptotical convergence of its tracking error is

also concluded using Barbalat’s lemma equation from Lyapunov stability and its adaptive

law becomes

𝑘̂𝑥 = Γ𝑥(𝑡)𝑥̃ 𝑇 (𝑡)𝑃𝑏, 𝑘̂𝑥 (0) = 𝑘𝑥0 (5.10)

Figure 5.2: Closed Loop Architecture for Indirect MRAC

5.2 ℒ1 ADAPTIVE CONTROLLER ARCHITECTURE

The architecture of both control method works, but with an increase in Γ, estimate

is reduced as t ≥ 0, but high adaptation gain leads to a high frequency oscillation. So,

increasing Γ, does allow a faster adaptation, but hurt robustness and stability because of the

frequency oscillation. To have a faster adaptation and maintain robustness, a controller

19
known as the ℒ1 adaptive controller is used. With ℒ1 adaptive controller, adaptation can be

separated from robustness. The architecture in figure 5.3 shows how ℒ1 adaptive controller

is designed. The design approach is a combination of the MRAC state predictor and a low

pass filter from the control input to the estimated model and the real plant. The controller

compared to indirect MRAC is giving as

𝑢(𝑠) = 𝐶(𝑠)𝑛(𝑠) (5.11)

C(s) transfer function is bounded input-bounded output stable which is subjected to C(s)=1

with zero initialization. This is achievable with a first order low pass filter where
ωc
𝐶(𝑠) = 𝑠+𝜔𝑐 (5.12)

and 𝜔𝑐 is the filter bandwidth, the filter should be chosen to maintain equation 5.13.

1
‖𝐺(𝑠)‖ℒ1 < (5.13)
𝐿

Where

𝑏
𝐺(𝑠) = 𝐻(𝑠)[1 − 𝐶(𝑠)], 𝐻(𝑠) = [1 − 𝐶(𝑠)] (5.14)
𝑠 + 𝑎𝑚

𝐿 ≜ 𝑚𝑎𝑥‖𝜃‖1 , 𝜃 ∈ Θ ⊂ ℝ𝑛 (5.15)

With low pass filter added to the system, robustness can be maintained with increased

adaptation gain. The adaptation law for ℒ1 adaptive controller is giving by

ΓProj(𝜃̂(𝑡), −𝑥̃ 𝑇 (𝑡)𝑃𝐵𝑥(𝑡)) where the Proj operator ensures that the unknown parameter

𝜃̂ stays bounded. 𝜃 would be replaced by 𝜎 for the quadrotor controller and the adaptive

for ℒ1 adaptive controller is giving by ΓProj(𝜎̂(𝑡), −𝑥̃ 𝑇 (𝑡)𝑃𝐵𝑥(𝑡)).

For projection operator with a given compact set{𝜎 ∈ ℝ | ‖𝜎‖ ≤ 𝑐},

𝜎̂ ′ (1 − ℎ(𝜎̂)) 𝑖𝑓 (ℎ(𝜎̂) > 0 & 𝜎̂ ′ ℎ(𝜎̂) > 0

Proj(𝜎̂, 𝜎̂ ′ ) = { (5.16)
𝜎̂ ′ 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

20
Where

𝜎̂ 2 − 𝜎̂𝑚𝑎𝑥
2
ℎ(𝜎̂) = 2
(5.17)
𝜀𝑎 𝜎̂𝑚𝑎𝑥
𝑐
𝜎̂𝑚𝑎𝑥 = , 𝜀𝑎 ∈ (0,1), 0≤𝑐≤1 (5.18)
√1+𝜀𝑎

Figure 5.3: Closed Loop Architecture for ℒ1 Adaptive controller

5.3 CONTROLLER DESIGN WITH ℒ1 ADAPTIVE CONTROLLER

Just like is session 4, we would design four controllers to control the movement of

the quadrotor using a state feedback approach. The information derived from the feedback

approach would then be used to design an ℒ1 adaptive controller to cancel out uncertainties.

5.4 TRANSLATION CONTROL ALONG THE Z-AXIS

From translation about the z axis,

1
𝑧̈ = (cos(𝜙) cos(𝜃))(−𝑈𝑡𝑧 ) + 𝑔 (5.19)
𝑚

1
𝑧̈ − 𝑔 = (cos(𝜙) cos(𝜃))(−𝑈𝑡𝑧 ) (5.20)
𝑚

𝑈𝑡𝑧 is the addition of the position control output and the control output for the cancelation

of gravity?

𝑈𝑡𝑧 = 𝑈𝑧𝑝 + 𝑈𝑧𝑔 (5.21)

21
Since the quadrotor is moving about the z axis, 𝜃 and 𝜙 would be equal to zero. So

cos(𝜙) cos(𝜃) would be equal to 1. Therefore

1
𝑧̈ − 𝑔 = (1) (−(𝑈𝑧𝑝 + 𝑈𝑧𝑔 )) (5.22)
𝑚

From equation 5.13, gravity would be ignored and later compensated where 𝑈𝑧𝑔 is equal

to mg. when it is ignored,

1
𝑧̈ = (𝑈 ) (5.23)
𝑚 𝑧𝑝

State space equation

𝑥̇ = 𝐴𝑧𝑝 𝑥 + 𝐵𝑧𝑝 𝑈𝑧𝑝

𝑇
𝑦 = 𝐶𝑧𝑝 𝑥

𝑧̇ 0 1 𝑧 0 𝑧
[ ]=[ ][ ] + [ ] 𝑢; 𝑦 = [1 0 ] [ ] (5.24)
𝑧̈ 0 0 𝑧̇ 1/𝑚 𝑧̇

With state feedback controller

𝑥̇ = 𝐴𝑧𝑝 𝑥 + 𝐵𝑧𝑝 𝑈𝑧𝑝

𝑇
𝑥̇ = 𝐴𝑧𝑝 𝑥 + 𝐵𝑧𝑝 (𝑘𝑔𝑟 − 𝑘𝑚𝑧𝑝 𝑥) (5.25)

where kg r is the reference gain and kmTz = [k1tz k2tz ]T which is the feedback gain.

Therefore
𝑇
𝑥̇ = (𝐴𝑧𝑝 − 𝐵𝑧𝑝 𝑘𝑚𝑧𝑝 )𝑥 + 𝐵𝑝 𝑘𝑔𝑟 (5.26)

𝑇
Where 𝐴𝑚𝑧𝑝 = (𝐴𝑃 − 𝐵𝑝 𝑘𝑚𝑧𝑝 ) So, the state spaces become

𝑥̇ = 𝐴𝑚, 𝑧𝑝 𝑥 + 𝐵𝑝 𝑘𝑔𝑟

This changes the state space model to

0 1 0
𝑧̇
[ ]=[ 1 1 ] [𝑧] + [ 1 ] 𝑘𝑔𝑟 (5.27)
𝑧̈ −𝑘1𝑧 ( ) −𝑘2𝑧 ( ) 𝑧̇
𝑚 𝑚 𝑚

22
Transfer function of the state space model becomes

1
𝐺𝑝 (𝑠) = 𝑚 (5.28)
1 1
𝑠2 + 𝑘2𝑧 (𝑚) 𝑠 + 𝑘1𝑧 (𝑚)

Compare to a reference transfer function

𝜔𝑛 2
𝐺𝑟𝑒𝑓 (𝑠) = (5.29)
𝑠 2 + 2𝜁𝜔𝑛 𝑠 + 𝜔𝑛 2

From equation 5.23, 𝜔𝑛 2 = 1/𝑚 and for critical damping, 𝜁 = 1. 𝑘1𝑡𝑧 = 1 and 𝑘2𝑡𝑧 = 2𝜁𝜔𝑛

𝑘2𝑡𝑧 = 2√𝑚 (5.30)

Therefore 𝑘2𝑡𝑧 = 1.79, where 𝑚=0.8. 𝑈𝑡𝑧 the controller output would be determined as

𝑈𝑡𝑧 = (𝑘𝑔𝑟 + 𝑘1𝑧 𝑧 + 𝑘2𝑧 𝑧̇ + 𝑚𝑔) (5.31)

Figure 5.5 is the simulation result tracking desired reference position and figure 5.6 is the

Simulink design for control using state feedback.

Figure 5.5: Altitude Position Using State Feedback Control [Z-Axis]

23
 Figure 5.6: State feedback for Altitude Control

With the feedback controller designed to ensure stable flight, we would then design a

controller to cancel out uncertainties which is dependent on the adaptation law for

uncertainty prediction. From equation 5.27, the state space model of the closed loop system

is defined below.

0 1
𝐴𝑚,𝑧𝑝 = [ 1 1 ]= [ 0 1
] (5.32)
−𝑘1𝑧 ( ) −𝑘2𝑧 ( ) −1.25 −2.24
𝑚 𝑚

0
𝐵𝑝 = [ 1 ] (5.33)
𝑚

For the adaptive law, we first need to determine the P matrix that will satisfy Lyapunov

equation 𝐴𝑇𝑚 𝑃 + 𝑃𝐴𝑚 = −𝑄. In order to achieve that, we would set Q as an identity matrix

as shown below.

1 0
𝑄= [ ] (5.34)
0 1

1.30 0.50
𝑃= [ ] (5.35)
0.50 0.50

Adaptation control law

θ̂̇ = ΓProj(𝜎̂(𝑡), −𝑥̃ 𝑇 (𝑡)𝑃𝐵𝑥(𝑡)) (5.36)

𝑧̃ 1.30 0.50 01
θ̂̇ = ΓProj(𝜎̂(𝑡), − [ ̇ ] [ ] [ ]) (5.37)
𝑧̃ 0.50 0.50 𝑚

θ̂̇ = ΓProj(𝜎̂(𝑡), −(0.63𝑧̃ + 0.63𝑧̃̇ )) (5.38)

−1
𝐺ℒ1,𝑡𝑧 = [[𝑠𝐼 − 𝐴𝑚,𝑝 ] 𝐵𝑝 ] [1 − 𝐶𝑡,𝑧 (𝑠)] (5.39)

Using matlab, the transfer function is where 𝜔𝑐=100

24
 1.25𝑠
𝑠3 + 102.24𝑠 2 + 225.25𝑠 + 125
𝐺ℒ1,𝑡𝑧 = (5.40)
1.25𝑠 2
[𝑠 3 + 102.24𝑠 2 + 225.25𝑠 + 125]

‖𝐺ℒ1,𝑡𝑧 ‖ℒ1 = 0.003 which give us an uncertainty limit of 333.33N. with that, the

adaptation gain can be big enough for faster adaptation. Also from equation 5.38, 𝑧̃ = 𝑧̂ −

𝑧 and 𝑧̃̇ = 𝑧̂̇ − 𝑧̇ . For simulation, a varying uncertainty of -10N to 10N would be used with

an adaptation gain of 10000. Figure 5.7 is the simulation result tracking desired reference

position using ℒ1 adaptive controller. Figure 5.8 shows the structural design of the altitude

controller in Simulink and figure 5.9 shows the simulation of the estimated uncertainty vs

the real uncertainty.

Another external force that could affect quadrotor movement in the Z axis is the

input of extra mass. Theoretically, there is no limit on the amount of extra mass that a

quadrotor can lift, but with a physical system, the amount of thrust and rotor speeds are

limited. Most physical quadrotor are designed so that the sum of all four motors can lift at

least twice its original mass. Assuming our simulation is designed for that purpose,
1
𝑚𝑒𝑥𝑡𝑟𝑎 ≤ 0.8. Therefore equation (5.23) can be rewritten as 𝑧̈ = 𝑚+𝑚 (𝑈𝑧𝑝 ). Same
𝑒𝑥𝑡𝑟𝑎

position controller input can be used to control the new rewritten model because 𝑚𝑒𝑥𝑡𝑟𝑎 is

constant and the controller output is well bounded. Extra mass can also affect movement

in the X and Y axis and same concept would be applied to compensate with the input of

extra mass.

25
 Figure 5.7: Altitude position [Z-Axis] with response to Varying Uncertainties and Extra

Mass

5.8: Structural Design of the Altitude Controller

5.9: Estimate Uncertainty vs Real Uncertainty for Altitude Control

5.5 ROTATION CONTROL ABOUT THE Z-AXIS

This is the yaw controller. The aim of the controller is to design a stable yaw rotation

controller to cancel uncertainties. From equation 5.4, using small angle approximation

26
 𝑙
𝜓̈ = 𝑈𝑟𝑧 (5.41)
𝐼𝑧𝑧

State space feedback is

0 1
𝜓̇ 𝑙𝑐
𝜓
𝑙𝑐 ] [ ] + [ 0 ] 𝑟𝑟𝑧
[ ]=[ (5.42)
𝜓̈ −𝑘1𝑟𝑧 ( ) −𝑘2𝑟𝑧 ( ) 𝜓̇ 𝑙𝑐/𝐽𝑧
𝐽 𝑧 𝐽 𝑧

𝜓
𝑦 = [1 0] [ ̇ ] (5.43)
𝜓

Transfer function is

𝑙𝑐
𝐽𝑧
𝐺𝑟,𝑧 (𝑠) = (5.44)
𝑙𝑐 𝑙𝑐
𝑠 2 + 𝑘2𝑟,𝑧 ( 𝐽 ) 𝑠 + 𝑘1𝑟,𝑧 (𝐽 )
𝑧 𝑧

𝑙𝑐
From equation 5.23, 𝜔𝑛 2 =𝐽 and for critical damping, 𝜁 = 1. 𝑘1𝑟,𝑧 = 1 and 𝑘2𝑟,𝑧 = 2𝜁𝜔𝑛
𝑧

𝑙𝑐
𝑘2𝑡,𝑥 = 2√ (5.45)
𝐽𝑧

𝑘2𝑟,𝑧 = 4 𝑤ℎ𝑒𝑟𝑒 𝑙 = 0.25𝑚, 𝑐 = 0.02, 𝐽𝑧 = 0.02𝑘𝑔𝑚2

𝑈𝑟𝑧 = 𝑟𝑟𝑧 − 𝑘1𝑟,𝑧 𝜓 − 𝑘2𝑟,𝑧 𝜓̇ (5.46)

Figure 5.10: Yaw Angle Tracking Using State Feedback Control

Where

0 1
𝐴𝑚 = [ 1 1 ]= [ 0 1
] (5.47)
−𝑘1𝑧 ( ) −𝑘2𝑧 ( ) −0.25 −1
𝑚 𝑚

27
 0
𝑙𝑐
𝐵𝑚 = [ ] (5.48)
𝐽𝑧

For the adaptive law, we first need to determine the P matrix that will satisfy Lyapunov

equation 𝐴𝑇𝑚 𝑃 + 𝑃𝐴𝑚 = −𝑄.

The P matrix to satisfy Lyapunov equation is

4.50 0.50
𝑃= [ ] (5.49)
0.50 0.625

Adaptation control law

θ̂̇ = ΓProj(𝜎̂(𝑡), −𝑥̃ 𝑇 (𝑡)𝑃𝐵𝑥(𝑡)) (5.50)

𝜓̃ 4.50 0.50 0
θ̂̇ = ΓProj(𝜎̂(𝑡), − [ ̇ ] [ ] [𝑙𝑐 ]) (5.51)
𝜓̃ 0.50 0.625 𝐽𝑧

θ̂̇ = ΓProj(𝜎̂(𝑡), − (0.125𝜓̃ + 0.156𝜓̃̇)) (5.52)

𝐺ℒ1,𝑟𝑧 = [[𝑠𝐼 − 𝐴𝑚 ]−1 𝐵𝑚 ][1 − 𝐶𝑟,𝑧 (𝑠)] (5.53)

Using matlab, the transfer function is where wc =100 for the low pass filter

0.25𝑠
𝑠3 +101𝑠 2 + 100.25𝑠 + 25
𝐺ℒ1,𝑟𝑧 = (5.54)
0.25𝑠 2
[𝑠 3 + 101𝑠 2 + 100.25𝑠 + 25]

‖𝐺ℒ1,𝑡𝑧 ‖ℒ1 = 0.001 which give us an uncertainty limit of 1000N. With that, the adaptation

gain can be big enough for faster adaptation. Also from equation 5.52, 𝜓̃ = 𝜓̂ − 𝜓 and

𝜓̃̇ = 𝜓̂̇ − 𝜓̇. For simulation, a varying uncertainty of -10N to 10N would be used with an

adaptation gain of 10000. Figure 5.11 provides result tracking desired reference angle using

ℒ1 adaptive controller with varying uncertainties while figure 5.12 provides simulation of

the estimated uncertainty vs the real uncertainty.

28
 Figure 5.11: Tracking of Desired Reference Yaw Angle with Varying Input

Uncertainties

Figure 5.12: Estimated Uncertainty vs Real Uncertainty for Yaw Control

5.6 TRANSLATION CONTROL ALONG THE X-AXIS

For movement along the X or Y axis, the quadrotor needs to rotate because it is

dependent on the rotation angle. From equation 5.56, you can notice that the translation

movement is dependent on 𝜙, 𝜃, 𝜑 and by using small angle approximation, we can move

in the X direction without the quadrotor getting unstable. For stability to be maintained,

the pitch and roll angle would be limited to maintain small angle approximation. For

movement along the X axis, the quadrotor rotates about the Y axis which provides us the

pitch angle. The yaw angle can then be set equal to zero because the yaw angle is relatively

29
small and does not have a huge effect on movement in the X and Y direction. Also, equation

5.55 can be linearized around hovering mode were 𝜙̇, 𝜃̇, 𝜑̇ is equal to zero.

𝑙 ̈
𝜃̈ = 𝑈 (5.55)
𝐽𝑦 𝑡𝑥

1
𝑋̈ = (cos(𝜙) sin(𝜃) sin(𝜑) + sin(𝜃) 𝑠𝑖𝑛(𝜑)(−𝑈𝑡𝑧 ) (5.56)
𝑚

Like the translational controller using small angle approximation,

1
𝑋̈ = − 𝑚 𝜃(𝑈𝑡𝑧 ) (5.57)

It is also important to note that when small angle approximation is used,

sin 𝜃 ≈ 𝜃

cos 𝜃 ≈ 1

tan 𝜃 ≈ 𝜃

We would assume that the control input is positive and we would compensate for it on the

controller output for movement along the x axis. We would also set 𝑈𝑡𝑧 = 𝑚𝑔 so that our

model can be simplified for easy calculation of the gain constant.

For our state feedback, we would four states and our state space feedback would be

modelled as

0 1 0 0 0
𝑥1̇ 0 0 𝑔 0 𝑥1
0
𝑥̇ 𝑥̇ 0 𝑈𝑡𝑥
[ 2] = 0 0 0 1 [ 1] + (5.58)
𝜃̇ 𝑙 𝑙 𝑙 𝑙 𝜃 𝑙
𝜃̈ − ( ) 𝑘1𝑡𝑥 − ( ) 𝑘2𝑡𝑥 −( ) 𝑘3𝑡𝑥 − ( ) 𝑘4𝑡𝑥 𝜃̇
[ 𝐽𝑦 𝐽𝑦 𝐽𝑦 𝐽𝑦 ] [𝐽𝑦 ]

The Simulink model is shown below and the transfer function becomes

𝑙
𝐽𝑦 𝑔
𝐺𝑡𝑥 (𝑠) = (5.59)
𝑙 𝑙 𝑙 𝑙
𝑠 4 + 𝑘4𝑟,𝑦 (𝐽 ) 𝑔𝑠 3 + 𝑘3𝑟,𝑦 (𝐽 ) 𝑔𝑠 2 + 𝑘2𝑟,𝑦 (𝐽 ) 𝑔𝑠 + 𝑘1𝑟,𝑦 (𝐽 ) 𝑔
𝑦 𝑦 𝑦 𝑦

30
 Figure 5.13: State Feedback Design in Simulink
4
𝑙 𝑙 𝑙 𝑙 4 𝑙
𝑠 4 + 𝑘4𝑟,𝑦 ( ) 𝑔𝑠 3 + 𝑘3𝑟,𝑦 ( ) 𝑔𝑠 2 + 𝑘2𝑟,𝑦 ( ) 𝑔𝑠 + 𝑘1𝑟,𝑦 ( ) 𝑔 = (𝑠 + √ 𝑔)
𝐽𝑦 𝐽𝑦 𝐽𝑦 𝐽𝑦 𝐽𝑦

The result using pole placement for the feedback gain constants becomes

𝐽𝑦3
4 2 𝐽
𝑦 4 𝐽𝑦
(𝑘1𝑡𝑥 = 1; ) (𝑘2𝑡𝑥 √
= 4 3 𝑔) ; (𝑘3𝑡𝑥 = 6 √ 𝑔) ; (𝑘4𝑡𝑥 = 4 √ )
𝑙 𝑙 𝑙𝑔

k1tx = 1; k2tx = 1.12; k3tx = 4.60; k4tx = .86;

Where 𝐽𝑦 = 0.015𝑘𝑔𝑚2 ; 𝑙 = 0.25𝑚; 𝑎𝑛𝑑 𝑔 = 9.81𝑁/𝑘𝑔

Figure 5.14: Roll Position Tracking without Varying Input Uncertainties

Because we are designing a controller for movement around the x axis, we would focus on

the pitch angle 𝜃. It is also important to know that figure 5.14 did not consider the limitation

31
of the pitch angle, but to maintain small angle approximation, 𝜃 should not exceed 14𝑜 or

0.244rad. However, when position is greater than 1.4m as shown in figure 5.16, the pitch

angle is greater than 0.244rad and if our desired position is as high as 25m, 𝜃 would be

higher than 360𝑜 or 3.14rad which makes no sense for the quadrotor to completely rotate

and possibly leads to more difficulty for the control of the quad.

Figure 5.15: Pitch Angle at a Desired Position of 1.4m

Figure 5:16 Pitch angle at a desired position of 25m

The problem with limitation is that it could affect state feedback control law and to avoid

that, we would have to first limit tracking error. Where

𝑒𝑡𝑥 = 𝑥𝑡𝑥 + 𝑒𝑡𝑥 𝑙𝑖𝑚𝑖𝑡

𝑒𝑡𝑥 𝑙𝑖𝑚𝑖𝑡 = −3𝑚 < 𝑒𝑡𝑥 < 3𝑚

𝑒𝑡𝑥 = 𝑟𝑡𝑥 − 𝑥𝑡𝑥 (5.60)

32
From equation 5.60, if the error between the reference input and the actual position is

greater than 3m, then 𝑒𝑡𝑥 𝑙𝑖𝑚𝑖𝑡 is equal to 3m and if less than -3m then 𝑒𝑡𝑥 𝑙𝑖𝑚𝑖𝑡 is equal to -

3m. By using this technique, the tracking error cannot exceed 3m even if the actual

reference is greater than 3m. With the limitation in place, it takes a longer time for the

quadrotor to reach desired position. Also 3m was the best limitation that would result in a

faster response and still maintain small angle approximation at a degree higher than

14𝑜 . Figure 5.17 and figure 5.18 shows the pitch angle remains same even if the distance

is greater than 1.4m. Figure 5.19 and 5.20 shows the response of the quadrotor at 1.4m and

at 25m.

Figure 5.17: Pitch Angle with Desired Position of 25m

Figure 5.18: Pitch Angle with Desired Position of 1.4m

33
 Figure 5.19: Tracking of Desired Position of 1.4m with Pitch Angle Limitation

Figure 5.20: Tracking of Desired Position of 25m with Pitch Angle Limitation

Unlike the altitude controller we cannot directly affect translation without affecting

rotation. Therefore, we have matched and unmatched uncertainty. Figure 5.21 provides a

simulation where unmatched uncertainties were not compensated for and in other to

compensate for these uncertainties, the model of the system is redefined as

𝑥̇ (𝑡) = 𝐴𝑚 𝑥(𝑡) + 𝐵𝑚 ((𝑢(𝑡) + 𝜎𝑚 (𝑡)) + 𝐵𝑢𝑚 𝜎𝑢𝑚 (5.61)

Where

𝐵𝑢𝑚 ∈ ℝ𝑛×𝑛(−𝑚) Is a matrix such that 𝐵𝑢𝑚

𝑇
= 0 and rank([𝐵𝑚 𝐵𝑢𝑚 ]) = 𝑛

There for

0
0 0
𝑔
𝐵𝑚 = 0 𝑈𝑡𝑥 𝑎𝑛𝑑 𝐵𝑢𝑚 = [ ] 𝜃 (5.62)
𝑙 0
0
[𝐽𝑦 ]

34
The control law for the controller to compensate for unmatched uncertainty would be defined as

̂ 𝑚 − 𝐶2 𝐻−1
𝑈 = −𝐶1 (𝑆)𝜎 𝑚 𝐻𝑢𝑚 𝜎
̂ 𝑢𝑚 + 𝐾𝑔𝑟 (5.63)

Where

𝐶1 & 𝐶2 are the low pass filter; 𝐻𝑚 = 𝐶[𝑠𝐼 − 𝐴𝑚 ]−1 𝐵𝑚 and 𝐻𝑢𝑚 = 𝐶[𝑠𝐼 − 𝐴𝑚 ]−1 𝐵𝑢𝑚 .

−1
𝐶2 𝐻𝑚 𝐻𝑢𝑚 𝜎̂𝑢𝑚 is needed to cancel the effect of unmatched uncertainty. In this paper, 𝐶2

filter would be a third order low pass filter. It is needed because it makes the transfer
−1
function of 𝐶2 𝐻𝑚 𝐻𝑢𝑚 𝜎̂𝑢𝑚 a proper transfer function. At a filter bandwidth of 100,

−1
60s2 + 860s + 4600
𝐶2 𝐻𝑚 Hum = (5.64)
s3 + 24.14s 2 + 241.4s + 1000

P matrix to satisfy Lyapunov equation

1.36 0.50 0.21 0.05

0.50 2.08 0.05 2.20
P= [0.21 0.05 2.22 0.50] (5.65)
0.05 2.20 0.50 5.51

Adaptation control law

θ̂̇m = ΓProj(σ
̂(t), − (0.834x̃ + 36.674x̃̇ + 8.335θ̃ + 91.852θ̃̇)) (5.66)

θ̂̇um = ΓProj(σ
̂(t), − (4.905x̃ + 20.405x̃̇ + 0.491θ̃ + 21.582θ̃̇)) (5.67)

The adaptation law for matched uncertainties does not include does not include movement

on the translational axis and therefore adaptation law for unmatched uncertainties does not

include movement on the rotational axis and can be shown in the equation below.

θ̂̇𝑚 = ΓProj(𝜎̂(𝑡), − (0𝑥̃ + 0𝑥̃̇ + 8.335𝜃̃ + 91.852𝜃̃̇)) (5.68)

θ̂̇𝑢𝑚 = ΓProj(𝜎̂(𝑡), − (4.905𝑥̃ + 20.405𝑥̃̇ + 0𝜃̃ + 0𝜃̃̇)) (5.69)

𝐺ℒ1,𝑡𝑥(𝑚𝑎𝑡𝑐ℎ𝑒𝑑) = [[𝑠𝐼 − 𝐴𝑚 ]−1 𝐵𝑚 ][1 − 𝐶𝑟,𝑧 (𝑠)] (5.70)

35
 1.635𝑒06𝑠 3 + 3.925𝑒08𝑠 2 + 3.925𝑒10𝑠
10000𝑠 7 + 2.543𝑒06𝑠 6 + 2.752𝑒08𝑠 5 + 1.363𝑒10𝑠 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2 + 1.871𝑒12𝑠 + 1.635𝑒12
1.635𝑒06𝑠 4 + 3.925𝑒08𝑠 3 + 3.925𝑒10𝑠 2
7 6 5 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2
= 10000𝑠 + 2.543𝑒06𝑠 + 2.752𝑒08𝑠 + 1.363𝑒10𝑠 + 1.871𝑒12𝑠 + 1.635𝑒12 (5.71)
1.635𝑒06𝑠 5 + 3.925𝑒08𝑠 4 + 3.925𝑒10𝑠 3
10000𝑠 7 + 2.543𝑒06𝑠 6 + 2.752𝑒08𝑠 5 + 1.363𝑒10𝑠 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2 + 1.871𝑒12𝑠 + 1.635𝑒12
1.635𝑒06𝑠 6 + 3.925𝑒08𝑠 5 + 3.925𝑒10𝑠 4
[10000𝑠 7 + 2.543𝑒06𝑠 6 + 2.752𝑒08𝑠 5 + 1.363𝑒10𝑠 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2 + 1.871𝑒12𝑠 + 1.635𝑒12]

𝐺ℒ1,𝑡𝑥(𝑢𝑛𝑚𝑎𝑡𝑐ℎ𝑒𝑑) = [[𝑠𝐼 − 𝐴𝑚 ]−1 𝐵𝑢𝑚 ][1 − 𝐶𝑟,𝑧 (𝑠)] (5.72)

1.635𝑒06 𝑠^5 + 3.939𝑒08 𝑠^4 + 3.959𝑒10 𝑠^3 + 3.554𝑒10 𝑠^2 + 1.805𝑒11 𝑠

10000𝑠 7 + 2.543𝑒06𝑠 6 + 2.752𝑒08𝑠 5 + 1.363𝑒10𝑠 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2 + 1.871𝑒12𝑠 + 1.635𝑒12
1.635𝑒06 𝑠^6 + 3.939𝑒08 𝑠^5 + 3.959𝑒10 𝑠^4 + 3.554𝑒10 𝑠^3 + 1.805𝑒11 𝑠^2
7 6 5 4 3 2
= 10000𝑠 + 2.543𝑒06𝑠 + 2.752𝑒08𝑠 + 1.363𝑒10𝑠 + 1.621𝑒11𝑠 + 8.11𝑒11𝑠 + 1.871𝑒12𝑠 + 1.635𝑒12 (5.73)
−1.832𝑒06 𝑠^4 − 4.412𝑒08 𝑠^3 − 4.435𝑒10 𝑠^2 − 3.925𝑒10 𝑠
10000𝑠 7 + 2.543𝑒06𝑠 6 + 2.752𝑒08𝑠 5 + 1.363𝑒10𝑠 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2 + 1.871𝑒12𝑠 + 1.635𝑒12
−1.832𝑒06 𝑠^5 − 4.412𝑒08 𝑠^4 − 4.435𝑒10 𝑠^3 − 3.925𝑒10 𝑠^2
[10000𝑠 7 + 2.543𝑒06𝑠 6 + 2.752𝑒08𝑠 5 + 1.363𝑒10𝑠 4 + 1.621𝑒11𝑠 3 + 8.11𝑒11𝑠 2 + 1.871𝑒12𝑠 + 1.635𝑒12]

‖𝐺ℒ1,𝑡𝑥(𝑚𝑎𝑡𝑐ℎ𝑒𝑑) ‖ = 0.008 which give us an uncertainty limit of 125N. With that, the
ℒ1

adaptation gain can be big enough for faster adaptation. From equation 5.68, 𝜃̃ = 𝜃̂ − 𝜃

and 𝜃̃̇ = 𝜃̂̇ − 𝜃̇. For simulation, a varying uncertainty of -10N to 10N would be used with

an adaptation gain of 10000.

‖𝐺ℒ1,𝑡𝑥(𝑢𝑛𝑚𝑎𝑡𝑐ℎ𝑒𝑑) ‖ = 0.056 which give us an uncertainty limit of 17.54N. From

ℒ1

equation 5.69, 𝑥̃ = 𝑥̂ − 𝑥 and 𝑥̃̇ = 𝑥̇ − 𝑥. For simulation, a varying uncertainty of -10N

to 10N would be used with an adaptation gain of 10000.

Figure 5.21: Tracking of Desired Reference Position with Varying Matched and

Unmatched Uncertainties without compensation for the Unmatched Uncertainties

36
Figure 5.22: Roll Position Tracking with Varying Matched Uncertainties

Figure 5.23: Structural Design for Roll Controller

Figure 5.24: Estimate Uncertainty vs Real Uncertainty for Roll Controller

37
Figure 5.25: Estimate Uncertainty vs Real Uncertainty for Unmatched Uncertainties Roll

Control

Figure 5.26: Roll Position Tracking with Varying Matched and

Unmatched Uncertainties

5.7 TRANSLATION CONTROL ALONG THE Y AXIS

This controller design is similar for the controller design for translation about the x

axis, so controller output for movement is

𝑢𝑡𝑦 = (𝑒𝑡𝑦 − 𝑘1𝑡𝑥 x − 𝑘2𝑡𝑥 𝑥̇ − 𝑘3𝑡𝑥 (𝜙) − 𝑘4𝑡𝑥 (𝜙̇))

You notice there is no multiplication of (-1) at the dynamic model which is the only

difference between both design. Therefore

k1ty = 1; k2ty = 1.12; k3ty = 4.60; k4ty = .86;

38
 Figure 5.27: Feedback Response without Disturbance

Figure 5.28: Pitch Position Tracking with Varying Matched

and Unmatched [Y-Axis]

Figure 5.29: Estimate Uncertainty vs Real Uncertainty for Matched Uncertainties Pitch

Control

39
Figure 5.30: Estimate Uncertainty vs Real Uncertainty for Unmatched Uncertainties Pitch

Control

40
 CHAPTER 6

CONTROLLER SIMULATION WITH NON-LINEAR MODEL

In Chapter 5, controllers were designed based on a linearized model of the

quadrotor, but in this chapter 6, the nonlinear model is used. The non-linear model is based

on what was discussed in chapter 3 and in addition to the model we would calculate the

thrust generated by each motor as shown below. These thrusts are mathematical relations

that were discussed in chapter 3 from equation 3.6-3.9.

𝐹1 1 1 1 1 −1 𝑈𝑡𝑧
𝐹2 0 −1 0 1 𝑈𝑡𝑥
[ ] = [1 0 −1 0 ] [𝑈𝑡𝑦 ]
𝐹3
𝐹4 1 −1 1 −1 𝑈𝑟𝑧

For the simulation, the table below shows the disturbances that were used on the quadrotor

model.

Tz Disturbance -10< 𝑇𝑧 < 10

Ty Disturbance -0.4< 𝑇𝑦 < 0.4

Tx Disturbance -0.4< 𝑇𝑥 < 0.4

Rz disturbance -10< 𝑇𝑧 < 10

Ry disturbance -12< 𝑅𝑦 < 12

Rx disturbance -12< 𝑅𝑥 < 12

Tz Extra mass -5< 𝑇𝑧 < 5

Figure 6.1: Disturbance Limit

41
Figure 6.1-6.4 shows the quadrotor response to desired position or location. It is exactly

same as most of the result shown in chapter 5. To avoid repetition of results such as tracking

of uncertainties, pitch and roll limitation would be avoided. Figure 6.5 and 6.6 provides the

overall Simulink design and each detail in each subsystem have been discussed from

chapter 3 to chapter 5.

Figure 6.2: Altitude Position Tracking Using ℒ1 Adaptive Controller [Z-Axis]

Figure 6.3: Yaw Position Yaw Position Tracking Using ℒ1 Adaptive Controller [Z-Axis]

Figure 6.4: Roll Position Tracking Using ℒ1 Adaptive Controller [X-Axis]

42
Figure 6.5: Pitch Position Tracking Using ℒ1 Adaptive Controller [Y-Axis]

Figure 6.6: Simulink Design with Uncertainties Input

43
Figure 6.7: Expanded Simulink Design

44
 CHAPTER 7

SUMMARY AND CONCLUSION

The main goal of this thesis is to design a controller that can control a quadrotor in

the presence of disturbance or uncertainties. Although they are different control method

used such as PID controller, this paper focused on the application of ℒ1 adaptive controller

in simulation. This paper includes the linear and non-linear dynamic model of the

quadrotor, application using PID controller, ℒ1 adaptive controller, and resistance to

uncertainties. As figure 4.10 shows, when a step disturbance is added to the input of the

altitude controller, it took about 160 seconds for the quadrotor to adapt so the main

advantage of ℒ1 adaptive controller over PID controller is its fast adaptation to

uncertainties or disturbance because of its high adaptation gain.

In the process of designing ℒ1 adaptive controller for the quadrotor, various

limitations and problems are taking into consideration such as the under actuation of the

quadrotor, presentation of matched and unmatched uncertainties, addition of extra mass

and failure of rotating motor. All these limitations were taking care of in this paper, except

for failure of the rotating motor. Theoretically when there is a motor failure, it is difficult

to have a stable flight. To avoid these problems, a different technique would be used for

future work. Future work would also focus on the implementation of ℒ1 adaptive controller

on a physical quadrotor. Various limitation from the micro controller, body

45
frame battery, IMU and cost of design are expected to affect performance, but with ongoing

research in the field of quadrotors, most of these limitations can be subdued.

46
 REFERENCES

Bouabdallah, Samir and Siegwart, Roland. 2007. Full Control of Quadrotor. Autonomous

Systems Lab: Swiss Federal Institute of Technology, ETHZ Zurich, Switzerland.

Brudigam, Tim. 2014. Fault-Tolerant Control of Quadrotor. Thesis, Technical University

of Munich.

ElKholy, Heba talla Mohamed Nabil. 2014. Dynamic Modeling and Control of a

Quadrotor Using Linear and Nonlinear Approaches. Thesis, America University

in Cairo.

Hovakimyan, Naira. 2013. ℒ1 Adaptive Control and Its Transition to Practice. Department

of Mechanical Science and Engineering University of Illinois.

Hovakimyan, Naira. 2013. ℒ1 Adaptive Control. Department of Mechanical Science and

Engineering University of Illinois.

Hovakimyan, Naira and Cao, Chengyu. 2010. ℒ1 Adaptive Control Theory: Guaranteed

Robustness with Fast Adaptation. Philadelphia: Society for Industrial & Applied

Mathematics, U.S.

Murray, M. Richard and Aström, J. Karl. 2009. Feedback Systems: An introduction for

Scientist and Engineering. Princeton: Princeton University Press.

Premerlani, William and Bizard, Paul. 2009. Direction Cosine Matrix IMU: Theory.

Retrieved from Web.

Raza, A. Syed and Gueaieb, Wail. 2010. Intelligent Flight Control of an Autonomous

Quadrotor. Science, Technology and Medicine. Doi: 10.5772/6968.

47
Schmidt, M. David. 2011. Simulation and Control of a Quadrotor Unmanned Aerial

Vehicle. University of Kentucky Library.

Seiler, Peter, Dorobantu, Andrei and Balas, Gary. 2010. Robustness Analysis of an

ℒ1 Adaptive Controller. AIAA Guidance, Navigation, and Control Conference.

Wang, Xiaofeng and Hovakimyan, Naira. 2012. ℒ1 Adaptive Controller for Nonlinear

Time-Varying Reference Systems. Systems & Control Letters, 61(4): 455-463.

48