HANOI UNIVERSITY OF SCIENCE AND TECHNOLOGY

School of Electrical & Electronic Engineering

PROJECT I
DESIGN FORMATION CONTROL SYSTEM FOR UAVs
NGUYEN MANH CUONG
cuong.nm210143@sis.hust.edu.vn
DAO DUC CONG
cong.dd222817@sis.hust.edu.vn
Major: Control Engineering and Automation

Instructor : Assoc. Prof.Nguyễn Hoài Nam

Department : Department of Automation Engineering
School : School of Electrical & Electronic Engineering

HA NOI, 05/2024
 ABSTRACT

This project investigates the leader - follower formation control and collision avoid-
ance problem for a class of UAVs. The approach involves several mild assumptions on
the interaction graph among the leader and the follower robots, utilizing observers to in-
directly gain the information of leader agent. Firstly, observers are designed to estimate
the leader’s states, which remove the requirement that each follower robot knows the full
state of the leader only by using local interactions among the robots. Secondly, based on
the observers, a distributed formation controller is proposed to achieve the desired setup
of a multi-robot system. Thirdly, Lyapunov function technique is employed to analyze
the efficiency of observers and convergence of formation tracking errors. Finally, sim-
ulations on a group of quadrotor are presented to demonstrate the effectiveness of the
theoretical results. This project is organized as follows:
Chapter 1. Introduction
Chapter 2. Preliminaries
Chapter 3. Control design
Chapter 4. Simulation
 TABLE OF CONTENTS

LIST OF FIGURES i

LIST OF TABLE ii

CHAPTER 1. INTRODUCTION 1
1.1 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Conclusions for Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . 1

CHAPTER 2. PRELIMINARIES 2
2.1 Graph Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Quadrotor Mathematical Model . . . . . . . . . . . . . . . . . . . . . 2
2.2.1 Coordinate Systems and Transformations . . . . . . . . . . . . 2
2.2.2 Forces and Moments . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.3 Quadrotor Dynamics Model . . . . . . . . . . . . . . . . . . . 3
2.3 Formation Control Problem . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Conclusions for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . 5

CHAPTER 3. CONTROL DESIGN 6

3.1 Distributed Observer Design . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Distributed Formation Controller Design . . . . . . . . . . . . . . . . . 7
3.2.1 Position Control . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2.2 Attitude Control . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.3 Conclusions for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . 11

CHAPTER 4. SIMULATION 12
4.1 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.3 Conclusions for Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . 17
CONCLUSION 18

REFERENCES 19
 LIST OF FIGURES

Figures 2.1 Schematic of a quadrotor from top view[1] . . . . . . . . . . . 4

Figures 3.1 Model system . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Figures 4.1 Topology graph . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Figures 4.2 Trajectories of all follower quadrotors . . . . . . . . . . . . . . 14
Figures 4.3 Control signal of all follower quadrotors . . . . . . . . . . . . . 14
Figures 4.4 Position output of all quadrotors . . . . . . . . . . . . . . . . . 15
Figures 4.5 Formation position errors of all follower quadrotors . . . . . . . 15
Figures 4.6 Attitude output of all follower quadrotors . . . . . . . . . . . . 16
Figures 4.7 Formation attitude errors of all follower quadrotors . . . . . . . 16
 LIST OF TABLE

Table 4.1 Quadrotor Simulation Parameters . . . . . . . . . . . . . . . . . 12

Table 4.2 Initial Position and Linear Velocity of Each Quadrotor . . . . . . 13
Table 4.3 Initial Angle and Angular Velocity of Each Quadrotor . . . . . . 13
Table 4.4 Initial States of Each Observer . . . . . . . . . . . . . . . . . . . 13
 CHAPTER 1. INTRODUCTION

1.1 Literature Review

In recent years, there has been remarkable progress in controlling the distributed
cooperation of networked multi-agent systems. As a typical application of the consen-
sus/synchronization of complex networks, the problem of controlling the formation of
multiple quadrotor aircraft has garnered considerable attention from various research
communities. Due to the numerous distinctive advantages of quadrotor aircraft (e.g.,
small size, simple structure, fast maneuverability, and vertical takeoff and landing ca-
pability), they are often employed to execute complex tasks in hazardous environments
such as disaster monitoring, wilderness exploration, search and rescue, among many
other activities.
This project will investigate the formation control problem for a group of quadrotor
aircraft using the consensus theory of multi-agent networks. Compared to single aircraft,
the advantages of controlling the formation of multiple quadrotor aircraft lie in several
aspects: Firstly, the overall payload capacity of the aircraft can be enhanced, meaning
more equipment or sensors can be equipped to perform more challenging tasks.
The usual assumption is that everyone monitoring knows the leader’s information,
but in reality, only a subset of the group may directly acquire data from the leader due to
hardware configuration constraints or communication range. Utilizing communication
links between agents, distributed observation sets for each agent are introduced to esti-
mate the leader’s state. Therefore, the stability and reliability of the system can also be
improved through information exchange among multiple aircraft.
Because both the control problem of a single quadrotor aircraft and the cooperative
control problem among multiple quadrotor aircraft need to be addressed, designing a for-
mation control system and analyzing the stability of the formation achieved for multiple
quadrotor aircraft will be much more challenging than controlling a single quadrotor
aircraft.
1.2 Research Objectives
Research, propose, and develop a control system for a set of UAVs (with quadrotors
as the main focus) into a specific dynamic model.
1.3 Conclusions for Chapter 1
Chapter 1 has presented some design methods, set out the scope, and outlined
the objectives to be achieved. This serves the construction of the algorithmic model in
Chapter 2 and the control system design in Chapter 3.

1
 CHAPTER 2. PRELIMINARIES

2.1 Graph Theory

This project studies a formation control problem of networked quadrotors, in which
the leader quadrotor labeled 0 decides the reference trajectory and others labeled i, i =
1, . . . , N , are followers. The network among these quadrotors is constituted by onboard
sensors and data transmitting/receiving devices, of which topology is described as fol-
lows.
The graph is denoted as G = {ν, ε} with a set of nodes ν = {0, 1, . . . , N} repre-
senting N+1 quadrotors and a set of edges ε = {( j, i)|i, j ∈ ν; j ̸= i} . A directed edge
( j, i), i, j ∈ ν means that the i th quadrotor can have access to the information of the j
th quadrotor, and the node j is also called a neighbor of node i . Let Ni denotes the set
which contains all neighbors of node i . The adjacency matrix of the graph G is denoted
by A = [ai j ] ∈ ℜ(N+1)×(N+1) with ai j = 0 and ai j = 1 for ( j, i) ∈ ε . The Laplacian matrix
of the graph G is denoted by L = [li j ] ∈ ℜ(N+1)×(N+1) with lii = ∑Nj=0 ai j and li j = −ai j
for ( j, i) ∈ ε. Let the Laplacian matrix L of the graph G be partitioned as follows.
!
N
∑i=1 a0i [a01 , . . . , a0N ]
L= (1)
−A0 1N H

where A0 = diag(a10 , a20 , ..., aN0 ) and each entry of 1N is 1.

Lemma 1 [2] All nonzero eigenvalues of H are with positive real parts. Furthermore,
H is nonsingular if and only if the graph G contains a directed spanning tree with the
node 0 as the root.
2.2 Quadrotor Mathematical Model
Consider a group of N + 1 three-dimensional quadrotors, of which mathematical
models are derived based on the assumption that each quadrotor is regarded as a rigid
body and the origin of the body-fixed coordinate system is at its center of mass.
2.2.1 Coordinate Systems and Transformations
To describe the states of quadrotor motion, appropriate coordinate systems are es-
tablished as follows:

• Inertial coordinate system Oxyz : coordinate origin O is at a certain point of the

earth surface; Ox axis is in the ground plane and points to the East; the nega-
tive direction of Oz axis is perpendicular to the ground plane and points to the
geocentre, Oy axis is in the ground plane as well and makes up the right-handed
coordinate system with Ox axis and Oz axis.

2
 • Body-fixed coordinate system oi xb yb zb : coordinate origin oi is at the centroid of
the i th quadrotor; oi xb axis coincides with the headward direction of the quadro-
tor; oi zb axis is in the quadrotor symmetric plane containing oi xb axis and points
upward; oi yb axis is perpendicular to the symmetric plane and makes up the right-
handed coordinate system with oi xb axis and oi zb axis.

The transformation matrix from the inertial coordinate system Oxyz to the body coordi-
nate system oi xb yb zb for the i th quadrotor is given by
 
cθi cψi cθi sψi −sθi
RI→B =  sφi sθi cψi − cφi sψi sφi sθi sψi + cφi cψi sφi cθi  (2)
 
i
cφi sθi cψi + sφi sψi cφi sθi sψi − sφi cψi cφi cθi

where (RI→B
i )T = (RI→Bi )−1 , s and c denote the trigonometric functions sin and cos
respectively, and θi , ψi , and φi are the pitch angle, yaw angle and roll angle of the i th
quadrotor, respectively.
2.2.2 Forces and Moments
The forces acting on the i th quadrotor are thrust Ti , gravity Gi and drag Di . Thrust
Ti equals to the resultant thrust from four rotors; gravity Gi is obtained as

Gi = mi g (3)

where g is the gravitational acceleration and mi the mass of the i th quadrotor and drag
Di is calculated by
Di = −ki,χ χ̇i (4)
where ki,χ = diag{ki,x , ki,y , ki,z } is the aerodynamic damping coefficient matrix, and χi =
[xi , yi , zi ]T is the position vector of the i th quadrotor under the inertial coordinate system.
To facilitate analysis of rotational motion for quadrotors, resultant moment can be
decomposed into pitch moment Mi,θ , roll moment Mi,φ and yaw moment Mi,ψ , respec-
tively. Pitch momentMi,θ and roll moment Mi,φ are generated by thrust difference, and
yaw moment Mi,ψ is generated by the reaction moment caused by the drag acting on four
rotors. The drag makes quadrotors have yaw tendency whose direction is opposite to the
corresponding rotor rotation.
2.2.3 Quadrotor Dynamics Model
It follows from [1] that the 6-DOF dynamics model of the i th quadrotor is de-
scribed by        
ẍi  0 0 −k i,x ẋ i 
  1 
 
 B→I   
ÿ = 0 +R 0 + −k ẏ (5)

 i  i    i,y i 
mi 


z̈i −mi g Ti −ki,z żi
 

3
      
Ji,θ θ̈i −ki,θ li,θ θ̇i Mi,θ
 Ji,ψ ψ̈i  =  −ki,ψ ψ̇i  +  Mi,ψ  (6)
     

Ji,φ φ̈i −ki,φ li,φ φ̇i Mi,φ

where Ji,θ , Ji,ψ , Ji,φ are moments of inertia, ki,θ , ki,ψ , ki,φ are aerodynamic damping co-
efficients, and li,θ , li,φ represent the distance from the epicenter of rotors to the corre-
sponding longitudinal plane of symmetry of the fuselage as shown in figure 2.1.

Figure 2.1. Schematic of a quadrotor from top view[1]

2.3 Formation Control Problem

Throughout this project, the communication topology among quadrotors satisfies
the following assumptions.

Assumption 1 [3] There is no directed path from followers to the leader; conversely,
there exists at least one path from the leader to any one of the followers. The data
communication between two followers is mutua

Assumption 2 [3] The graph G contains a directed spanning tree with the node 0 being
the root.

The objective of this paper is to synthesize a suitable control law such that the for-
mation control problem of multiple quadrotors defined as follows is addressed. It should
be noted that the formation trajectory is decided by the leader quadrotor and the forma-
tion structure is determined by the desired relative position di j = [di j,x , di j,y , di j,z ]T , i, j ∈
ν, from the i th quadrotor to its neighbor, the j th quadrotor.

Definition 1 Consider N follower quadrotors and a leader quadrotor as described in

4
(5)–(6), and define the formation tracking errors as

ei,x = x0 − xi + di0,x ,
ei,y = y0 − yi + di0,y , (7)
ei,z = z0 − zi + di0,z ,

Given the graph G and the formation trajectory (x0 , y0 , z0 ) decided by the uncontrolled
leader quadrotor, find a dynamic control law such that

lim ei,x (t) = 0

t→∝
lim ei,y (t) = 0 (8)
t→∝
lim ei,z (t) = 0
t→∝

2.4 Conclusions for Chapter 2

This chapter introduced the fundamental basis of graph theory, interested robot’s
model. In addition, the research scope of this project is limited in terms of robot’s config-
uration, environment’s status, and initial condition for the multi-robot system. Finally,
the problem statements are presented in mathematical expressions to form the control
objectives.

5
 CHAPTER 3. CONTROL DESIGN

In what follows, the main results of this paper will be presented in two steps. First,
a distributed observer is designed to provide the estimated information on the leader
for each follower. Second, a controller under a cascade structure is designed to achieve
leader-follower formation.(figure 3.1)

Figure 3.1. Model system

3.1 Distributed Observer Design

It is known from Assumption 1 that not all follower quadrotors have knowledge of
information on the leader quadrotor. Moreover, it is also known from the communication
network setup among quadrotors that each quadrotor only has access to local informa-
tion. Therefore, the formation tracking errors in (7) cannot be directly used in formation
controller design. To circumvent the problem, a distributed observer [4]is designed to
estimate the leader information for each follower, which takes the form of
!
1 η1
x̂˙i = ∑ ai j x̂˙ j − ∑ ai j (x̂i − x̂ j ) , (9)
∑ ai j j∈Ni ∑ ai j j∈Ni
j∈Ni j∈Ni
!
1 η2
ŷ˙i = ∑ ai j ŷ˙ j − ∑ ai j (ŷi − ŷ j ) , (10)
∑ ai j j∈Ni ∑ ai j j∈Ni
j∈Ni j∈Ni
!
1 η3
ẑ˙i = ∑ ai j ẑ˙ j − ∑ ai j (ẑi − ẑ j ) , (11)
∑ ai j j∈N i
∑ ai j j∈N i
j∈Ni j∈Ni

i = 1, . . . , N

where [x̂i , ŷi , ẑi , x̂˙i , ŷ˙i , ẑ˙i ]T are the states of the i th observer, which are the estimates of po-
sitions and velocities of the leader quadrotor for the i th follower quadrotor, η1 , η2 , η3 >

6
0. The distributed observer (9)–(11) has the following property.

Lemma 2 [4] Consider the graph G and the distributed observer (9)–(11) Under As-
sumptions 1 and 2, the estimates x̂i , ŷi , ẑi , i = 1, . . . , N, converge to x0 (t), y0 (t), z0 (t)).

Proof: Take the distributed observer design on the x axis as an example. It follows from
(9) that
!
∑ ai j (x̂˙i − x̂˙ j ) = −η1 ∑ ai j (x̂i − x̂ j ) (12)
j∈Ni j∈Ni

Define ∆i = ∑ ai j (x̂i − x̂ j ), ta có:

j∈Ni

∆˙ i = −η1 ∆i (13)

Consider Lyapunov functions Vi = 12 ∆2i , i = 1, . . . , N. The time derivative of Vi along the

trajectory of (13) satisfies:

V̇i =∆i ∆˙ i
= − η1 ∆2i ≤ 0 (14)

One has that:

[∆1 (t), . . . , ∆N (t)]T = H[x̂1 − x0 , . . . , x̂N − x0 ] = 0

Applying Assumption 2 and Lemma 1, the proof is thus completed.

3.2 Distributed Formation Controller Design
The proposed controller is composed of an inner-loop attitude controller and an
outer-loop position controller in terms of cascade structures since it is known from the
analysis of dynamic characteristics for aircrafts that quadrotors have slower position dy-
namics and faster attitude dynamics. The distributed position controller receives quadro-
tor states to make a group of quadrotors achieve a desired formation pattern, while the
attitude controller receives desired angles generated by the position controller to regulate
quadrotor attitudes.
3.2.1 Position Control
For the convenience of description, the translational dynamics of the i th quadrotor
in (5) is expressed as 
ki,x

 ẍi = − ẋi + ui,x ,
m

i


 ki,y
ÿi = − ẏi + ui,y , (15)
 mi
k

z̈i = − i,z żi − g + ui,z ,



mi

7
Where: 
Ti

ui,x = (cos ψi sin θi cos φi + sin ψi sin φi ),
mi



 Ti
ui,y = (sin ψi sin θi cos φi − cos ψi sin φi ), (16)
 mi
Ti


ui,z = (cos θi cos φi ).


mi
To address the formation control problem as formulated in Definition 1, the distributed
position controller for each quadrotor is designed as follows:

ui,x = ∑ ai j ki,1 (x̂i −x̂ j −xi +x j +di j,x )
j∈Ni

+ ki,2 (v̂i,x − v̂ j,x − vi,x + v j,x )
ki,x
+ v̂˙i,x + ẋi (17)
mi

ui,y = ∑ ai j ki,1 (ŷi −ŷ j −yi +y j +di j,y )
j∈Ni

+ ki,2 (v̂i,y − v̂ j,y − vi,y + v j,y )
ki,y
+ v̂˙i,y + ẏi (18)
mi

ui,z = ∑ ai j ki,1 (ẑi − ẑ j − zi + z j + di j,z )
j∈Ni

+ ki,2 (v̂i,z − v̂ j,z − vi,z + v j,z )
ki,z
+ v̂˙i,z − g + żi (19)
mi
With ki,1 , ki,2 > 0.

Theorem 1 [5] Consider the graph G and the closed-loop system consisting of the sys-
tem (14) and the distributed dynamic control law (9)–(11) and (17)–(19). Under As-
sumptions 1–2, then the leader-follower formation control problem is solved via the pro-
posed control law.

Proof:When designing the position controller, we only take the proof on the x axis as an
example.
Define:

êi,x =x̂i − xi + di0,x ,

êi,vx =v̂i,x − vi,x , (20)

8
where x̂i and v̂i,x are the estimates of the leader information on position and velocity for
the i th follower.
From (15), (17) and (20) one has that:

ê˙i,x =êi,xv ,
ki,x
ê˙i,vx =v̂˙i,x + ẋi − ui,x
mi
= − ∑ ai j ki,1 (êi,x − ê j,x )
j∈Ni

− ∑ ai j ki,2 (êi,vx − ê j,vx ). (21)

j∈Ni

Consider Lyapunov functions

" #
N
1
Vp = ∑ ∑ a′i j ki,1 (êi,x −ê j,x )2 + ê2i,vx (22)
i=1 j∈Ni 2

Where a′i j = ai j , i, j ∈ V and a′i0 = 2ai0 , i ∈ V . With ai j = a ji , The time derivative:

" #
N
V̇p = ∑ ∑ a′i j ki,1 (êi,x − ê j,x )(êi,vx − ê j,vx )
i=1 j∈Ni
N 
− 2 ∑ êi,vx ∑ ai j ki,1 (êi,x − ê j,x )
i=1 j∈Ni

+ ∑ ai j ki,2 (êi,vx − ê j,vx )
j∈Ni

After some algebraic transformation, one has:

N
V̇p = − ∑ ∑ a′i j ki,2 (êi,vx − ê j,vx )2
i=1 j∈Ni

≤0. (23)

Let ∆e,i = ∑ a′i j (êi,x − ê j,x ):

j∈Ni

[∆e,1 , . . . , ∆e,N]T = H ′ [ê1,vx −e0,vx , . . . , êN,vx −e0,vx ]T (24)

H ′ = H + A0

Since H ′ is nonsingular ê0,x = 0,Upon using Lemma 1, ∆e,i = 0 as êi,x = 0 with (23), the
proof is thus completed.

9
3.2.2 Attitude Control
The desired attitude of the quadrotors calculated by intermediate control inputs
ui = [ui,x , ui,y , ui,z ]T , i = 1, . . . , N, is given by
 q
d =m 2 2 2
T i ui,x + ui,y + ui,z ,



 i
 
mi (ui,x sin ψid −ui,y cos ψid )


d
φi = arcsin ,
Tid (25)
  
u cos ψid +ui,y sin ψid


θid = arctan i,x ,


ui,z

To facilitate the attitude controller design, rewriting the rotational dynamics of the
i th quadrotor in (6) gives that

k l

 θ̈i = − i,θJ i,θ θ̇i + τi,θ ,

 i,θ
ki,ψ
ψ̈i = − Ji,ψ ψ̇i + τi,ψ , (26)

φ̈i = − ki,φ li,φ φ̇i + τi,φ ,


Ji,φ

M
i,θ Mi,ψ Mi,φ
Where τi,θ = Ji,ψ , τi,ψ = Ji,ψ , τi,φ = Ji,φ The attitude controller for each quadrotor is
designed as follows:

τi,θ =ki,3 (θid − θi )

ki,θ li
+ ki,4 (θ̇id − θ̇i ) + θ̈id + θ̇i , (27)
Ji,θ
τi,ψ =ki,3 (ψid − ψi )
ki,ψ
+ ki,4 (ψ̇id − ψ̇i ) + ψ̈id + ψ̇i , (28)
Ji,ψ
τi,φ =ki,3 (φid − φi )
ki,φ li
+ ki,4 (φ̇id − φ̇i ) + φ̈id + φ̇i , (29)
Ji,φ

With ki,3 , ki,4 > 0.

Theorem 2 [6]The closed-loop system consisting of the system (26) and the control law
(27)–(29). Then with the proposed control law, the desired attitudes in (25) can be
tracked.

Proof:We take pitch angle θ as an example.

Define:
ei,θ = θid − θi (30)

10
From (26), (27) and (30) one has that:

ėi,θ =θ̇id − θ̇i ,

ëi,θ = − ki,3 ei,θ − ki,4 ėi,θ (31)

Consider Lyapunov functions

N
1 1
Va = ∑ [ki,3 e2i,θ + ė2i,θ ], (32)
i=1 2 2

The time derivative:

N
V̇a = ∑ [ki,3 ei,θ ėi,θ − ėi,θ (ki,3 ei,θ + ki,4 ėi,θ )]
i=1
N
= − ∑ ki,4 ė2i,θ ≤ 0 (33)
i=1

The proof is thus completed.

3.3 Conclusions for Chapter 3
In this chapter, the procedure of designing controller for the multi-robot system is
presented with the utilization of observers, position control and attitude control. The
Lyapunov technique is deployed to demonstrate the convergence of formation errors.

11
 CHAPTER 4. SIMULATION

4.1 Numerical Simulations

Consider four quadrotors with dynamics (5), (6) to demonstrate the performance
of the proposed control algorithm. Simulations with three quadrotors following a leader
quadrotor while maintaining a triangle formation have been conducted. The parameters
of quadrotor in table 4.1 and communication graph in gigure 4.1.

Table 4.1. Quadrotor Simulation Parameters

Parameters Value Parameters Value

mi 1(kg) g 9.8(m/s2 )
Ji,θ , Ji,φ 1.466x10−2 (kg.m2 ) Ji,ψ 2.848x10−2 (kg.m2 )
ki,x , ki,y 6.579x10−2 (kg/s) ki,z 0.054(kg/s)
ki,θ , ki,φ 0.01(kg.m/s) ki,ψ 0.02(kg.m/s)
li,θ , li,φ 0.025(m) i 0,1,2,3

Figure 4.1. Topology graph

The desired formation trajectory is [x0 (t), y0 (t), z0 (t)]T = [5sin(0.2t), 5cos(0.2t),t]T ,
, the desired formation structure is a triangle, and the desired relative positions from each
follower to the leader are given by dd10 = [0, 0, 10]T , d20 = [10cos( 7π 7π T
6 ), 0, 10sin( 6 )] , d30 =
[10cos( −π −π T
6 ), 0, 10sin( 6 )] .

In the simulation, the initial states of each follower quadrotor are listed in Tables
4.2–4.3 and the initial states of each finite-time observer are given in Table 4.4.

12
 Table 4.2. Initial Position and Linear Velocity of Each Quadrotor

Label [xi (0), yi (0), zi (0)]T , m [ẋi (0), ẏi (0), żi (0)]T , m/s
1 [0, 0, 0]T [0, 0, 0]T
2 [0, 2.5, 0]T [0, 0, 0]T
3 [−2.5, 0.8, 0]T [0, 0, 0]T

Table 4.3. Initial Angle and Angular Velocity of Each Quadrotor

Label [θi (0), φi (0), ψi (0)]T , rad [θ̇i (0), φ̇i (0), ψ̇i (0)]T , rad/s
1 [0, 0, π/6]T [0, 0, 0]T
2 [0, 0, 5π/6]T [0, 0, 0]T
3 [0, 0, −π/6]T [0, 0, 0]T

Table 4.4. Initial States of Each Observer

Label x̂i (0), m ŷi (0), m ẑi (0), m

1 0 0 0
2 0.15 -0.21 -0.18
3 0.1 0.2 -0.13

Other parameters:

η1 = η2 = η3 = 10,
ki,1 = 10, ki,2 = 10,
ki,3 = 10, ki,4 = 10,
i = 1, 2, 3

4.2 Results
The simulation results of the project will be present3d. It includes the motion
trajectory in figure 4.2, control signals in figure 4.3, output and errors about position in
figure 4.4-4.5, attitide in figure 4.6-4.7 of all follower quadrotors

13
 Figure 4.2. Trajectories of all follower quadrotors

Figure 4.3. Control signal of all follower quadrotors

14
 Figure 4.4. Position output of all quadrotors

Figure 4.5. Formation position errors of all follower quadrotors

15
 Figure 4.6. Attitude output of all follower quadrotors

Figure 4.7. Formation attitude errors of all follower quadrotors

Remark 1 From the simulation results, we see that the position output of the following
quadrotors follows the virtual leading quadrotor with position distances after about 4s.
The position and state errors approach 0. From there, we can see that the method of
combining observers to support formation formation achieves the desired results through
simulation.

16
4.3 Conclusions for Chapter 4
In this chapter, a scenarios in simulation are introduced to verify the effectiveness
of the proposed control algorithm. All results are analyzed in data and figures with
different criteria to perceive the entire terms of the proposed control methodology

17
 CONCLUSION

In this thesis, the problem of formation control for a class of quadrotor systems has
been investigated. To reduce reliance on the requirement for accurate state information of
the leading mobile robot, a state estimator is employed to facilitate coordinated support
tasks through theoretical analysis. Distributed formation control laws are constructed
based on data from each agent’s observation set. Lyapunov stability theory is applied to
demonstrate the convergence of the tracking error during the formation process.

18
 REFERENCES

[1] B. Zhao, B. Xian, Y. Zhang, and X. Zhang, “Nonlinear robust adaptive tracking con-
trol of a quadrotor uav via immersion and invariance methodology,” IEEE Transac-
tions on Industrial Electronics, vol. 62, no. 5, pp. 2891–2902, 2014.

[2] Y. Su and J. Huang, “Cooperative output regulation of linear multi-agent systems by

output feedback,” Systems & Control Letters, vol. 61, no. 12, pp. 1248–1253, 2012.

[3] H. Du, W. Zhu, G. Wen, Z. Duan, and J. Lü, “Distributed formation control of
multiple quadrotor aircraft based on nonsmooth consensus algorithms,” IEEE trans-
actions on cybernetics, vol. 49, no. 1, pp. 342–353, 2017.

[4] X. Yu and L. Liu, “Distributed formation control of nonholonomic vehicles subject

to velocity constraints,” IEEE Transactions on Industrial Electronics, vol. 63, no. 2,
pp. 1289–1298, 2015.

[5] L. Dou, X. Ma, C. Yang, M. Du, and X. Wang, “Distributed finite-time formation
control for quadrotors based on adaptive state observer,” in 2019 Chinese Control
Conference (CCC), pp. 8085–8090, IEEE, 2019.

[6] L. Dou, C. Yang, D. Wang, and T. Chen, “Distributed finite-time formation con-
trol for multiple quadrotors with leader state unmeasurable,” in 2018 37th Chinese
Control Conference (CCC), pp. 9843–9848, IEEE, 2018.

19