Finite-time Formation Tracking Control with Collision Avoidance for Quadrotor UAVs

Journal Pre-proof

Finite-time Formation Tracking Control with Collision Avoidance for

Quadrotor UAVs

Youfang Huang, Wen Liu, Bo Li, Yongsheng Yang, Bing Xiao

PII: S0016-0032(20)30029-6
DOI: https://doi.org/10.1016/j.jfranklin.2020.01.014
Reference: FI 4376

To appear in: Journal of the Franklin Institute

Received date: 20 May 2019

Revised date: 12 October 2019
Accepted date: 20 January 2020

Please cite this article as: Youfang Huang, Wen Liu, Bo Li, Yongsheng Yang, Bing Xiao, Finite-time
Formation Tracking Control with Collision Avoidance for Quadrotor UAVs, Journal of the Franklin Insti-
tute (2020), doi: https://doi.org/10.1016/j.jfranklin.2020.01.014

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2020 Published by Elsevier Ltd on behalf of The Franklin Institute.

Highlights

• The formation tracking control for quadrotor UAVs with complicated

collision avoidance is investigated.

• Two less conservative finite-time disturbance observers are constructed

to compensate external disturbances.

• Based on the artificial potential function (APF) and fast terminal slid-
ing mode surface (FTSM), a novel sliding mode surface-like variable is
proposed as a foundation of the finite-time position controller design.

• The novel distributed global finite-time position formation tracking al-

gorithm and finite-time attitude tracking algorithm are respectively
proposed for the safe position formation tracking, avoiding the moving
obstacles as well as attitude synchronization and tracking.

• The whole control system is guaranteed to be globally finite-time stable.

 Finite-time Formation Tracking Control with Collision
Avoidance for Quadrotor UAVs✩
Youfang Huanga , Wen Liua , Bo Lia,∗, Yongsheng Yanga , Bing Xiaob
a
Institute of Logistics Science and Engineering, Shanghai Maritime University,
Shanghai, 201306, China
b
School of Automation, Northwestern Polytechnical University, Xi’an 710072, China

Abstract

In this paper, the distributed finite-time formation tracking control prob-

lem with collision avoidance is investigated for multiple quadrotor Unmanned
Aerial Vehicles (UAV) subject to external disturbances. Firstly, two finite-
time observers are designed to estimate the external disturbance force and
torque without upper bound, respectively. Then, utilizing the precisely ob-
served information deriving from the observers, a novel sliding mode surface-
like variable based distributed position formation tracking control strategy
and a fast terminal sliding mode based attitude tracking control strategy are
proposed, respectively. It achieves that the multiple quadrotor UAVs can
track the desired trajectory in a specific formation configuration within the
safe distance and avoid dynamic obstacles. Meanwhile, the rapid attitude

✩
This work was supported in part by National Natural Science Foundation of
China (61703272, 61873207), and Young Teachers Program of Shanghai Universi-
ties (ZZHS18004) and Science and Technology Commission of Shanghai Municipality
(17595810300, 18295801100), and Yunnan Science and Technology Program (2018IB022).
∗
Corresponding author
Email addresses: [email protected] (Youfang Huang), [email protected]
(Wen Liu), [email protected] (Bo Li), [email protected] (Yongsheng Yang),
[email protected] (Bing Xiao)

Preprint submitted to Journal of the Franklin Institute February 10, 2020

synchronization and tracking is also achieved. The entire multiple UAVs
distributed formation tracking closed-loop control system are proved to be
globally fast finite-time stable by using Lyapunov-like analysis. Finally, the
effective and fine performance of the proposed schemes are demonstrated by
a numerical simulation example with a group of quadrotor UAVs.
Keywords: Quadrotor UAVs, Formation tracking, Collision avoidance,
Artificial potential function, Sliding mode control

1. Indroduction

With the market shares of quadrotor UAV increased tremendously, the

formation of multiple quadrotor UAVs becomes an even more attractive re-
search focus. It has been successfully applied to target tracking [1, 2], forest
fire monitoring [3], surveillance [4] and light show [5]. As an important clas-
sification of UAV formation, UAV formation tracking means that multiple
UAVs should not only achieve the pre-set formation configuration, but also
track the leader (target) or reference trajectory in real time. It is usually a
significant subtask in UAV applications, such as the moving targets tracking
in military combat and urban counter-terrorism mission, rescuers tracking in
maritime search-and-rescue mission and so on [6]. Especially taking external
disturbances and accidental collision into consideration, the UAVs forma-
tion tracking becomes even more complicated and difficult. Therefore, the
fast formation tracking control problem of UAVs in the presence of external
disturbances and obstacles has attracted considerable attentions.
Recently, many relative control methods for formation tracking have been
proposed. In [7, 8], the distributed consensus method was adopted for for-

3
mation controller design. Then, the adaptive technique was applied for the
parameter estimation of formation control in [9, 10]. Based on backstepping
and leader-follower strategies, Liu [11] investigated the formation tracking
control of quadrotor UAVs in three-dimensional space, under the considera-
tion of the control input with communication block and load factor. He et al.
[12] combined the virtual structure with output-feedback method to realize
the asymptotic stability of tracking error. To reduce energy consumption and
communication frequency, the impulsive control [13], even-triggered control
[14] were effectively applied, respectively. Moreover, in [15], the potential
function and neighboring information in directed network was utilized for
controller design of formation tracking. However, only the asymptotic sta-
bility was achieved in references above. Consider the finite-time convergence
of the formation tracking system, the distributed formation tracking control
was investigated based on fast non-singular terminal sliding mode in [16].
Although the aforementioned literatures could guarantee the stability per-
formance, the control accuracy and security of the UAVs formation tracking
control system in the presence of disturbances or possible collisions in prac-
tice are not well further studied in terms of fast finite-time control.
During the flight mission, external disturbances or uncertainties are in-
evitable, which will make a persistent impact on controller stability. Espe-
cially for some severe disturbances, even directly lead to formation control
system out of control. Thus, the disturbance attenuation scheme needs to
be considered essentially. In [17], the adaptive control and quantum logic
methods were adopted to deal with the external disturbances and uncertain-
ties about actuator failures. In [18, 19], robust neural network estimation

4
scheme was applied to tackled the model uncertainties and external distur-
bances. Then, a fuzzy uncertainty observer was designed for quadrotor UAV
to compensate the unknown nonlinearities in [20]. Nevertheless, the conver-
gence of estimation error was achieved infinitely in these above. To acquire
the accurate disturbances estimation in finite time, extended state observer
in [21, 22], sliding mode based disturbance observer in [23, 24], and their in-
tegration in [25, 26] were designed. However, the above-mentioned methods
handling disturbances have some conservatism, that is with the assumption
of disturbances having constant upper bound as the precondition. Whereas,
the upper bound of uncertainty is generally hard to be determined. Thus,
it will be of great significance to study the perturbation attenuation tech-
nique without upper bound. Meanwhile, Xiao et al. [27] studied this issue in
quadrotor UAV trajectory tracking, and extended this scheme to deal with
the uncertainties for robotic manipulators tracking control [28]. However,
[27, 28] only realized asymptotic observation for uncertainties.
During the maneuver of quadrotor UAVs formation tracking, effective
collision avoidance plays an important part in the whole safe flight mis-
sion. It includes three categories: (1) The collision avoidance between the
follower UAV and the target leader, noted as CAT. (2) The collision avoid-
ance between any two follower UAVs in the formation, noted as CAF. (3)
The collision avoidance between the follower UAV and obstacle, noted as
CAO. Consider the CAO problem, reference [29] integrated dual-mode con-
trol strategy with model predictive control (MPC). The quadrotor UAVs
formation can avoid the static obstacles autonomously and safely in 3D en-
vironment. And distributed MPC was applied in [30]. As for the CAF

5
problem, Xia introduced the artificial potential function(APF) to keep safe
formation by using location estimation formation in [31]. Meanwhile, the
repulsive APF was used in [19]. Moreover, attractive APF and repulsive
APF was combined to solve the problem of CAF, the spacecrafts automatic
rendezvous control is achieved in [32]. At the same time, the authors in [33]
further dealt with this issue with path constraints of chasers by utilizing the
same APF and the adaptive sliding mode method. However, the references
above only covered part of collision avoidance problem about CAF or CAO
. Taking fully account of the problem of CAT, CAF and CAO in formation
tracking, reference [34] integrated the null-space-based method with FNTSM
for spacecraft distributed formation control. In [35], the APF was used to
construct the optimal control scheme, so as to keep safe formation tracking
and avoid the square obstacles accurately. Dong et al. [36, 37] proposed
the comprehensive APF with a new sliding mode method to achieve forma-
tion tracking, but neither is distributed control. It should be noted that the
references mentioned above only achieved theoretically asymptotically sta-
ble. It extremely limits the efficiency of safe formation. Moreover, seldom
literature has studied the full collision avoidance problems in the field of
quadrotor UAVs formation tracking. Thus, it remains a challenging problem
to investigate distributed finite-time formation tracking control for quadrotor
UAVs with collision avoidance. Inspired by the existing deficiencies above,
this paper concentrates on investigating the problem of distributed globally
finite-time formation tracking control with collision avoidance for quadrotor
UAVs. The main contributions are stated as follows:
(1) Two less conservative finite-time disturbance observers are con-

6
structed to compensate external disturbances. Compared with finite-time ob-
servers aforementioned, the proposed disturbance observer can eliminate con-
servative assumption that the disturbance uncertainties have upper bound.
(2) A novel sliding mode surface-like variable is proposed for position
controller design. Different from [32, 33, 36], whose sliding mode surface
combined the gradient of APF with linear sliding mode surface, the pro-
posed new sliding mode surface-like variable integrates the gradient of repul-
sive APF with fast terminal sliding mode surface(FTSM). Then, the mutual
conversion with FTSM can be achieved depending on whether the collisions
actually occur or not. Therefore, it possesses the properties of FTSM in
terms of fast convergence rate, lower steady state tracking error and stronger
robustness.
(3) The distributed globally finite-time position formation tracking algo-
rithm and finite-time attitude tracking algorithm are respectively proposed
for the safe formation tracking control. By this, the system tracking errors
can theoretically converge in global finite time.
Notations: T and −1 respectively represent the sign for trans-
pose and inverse. R denotes the real number. Rm×n represents
the real matrices with dimensions of m × n. For α ∈ R, the vec-
tor x = [x1 , x2 , . . . , xn ]T ∈ Rn×1 , and standard sign function sign(·),
we define sigα (x) = [|x1 |α sign(x1 ), |x2 |α sign(x2 ), . . . , |xn |α sign(xn )]T .
Let diag(x1 , x2 , ..., xn ) ∈ Rn×n be diagonal matrix with
x1 , x2 , . . . , xn as diagonal entries. Define 1n = [1, 1, . . . , 1]T ∈Rn×1 ,
In = diag(1, 1, . . . , 1) ∈Rn×n . Note k·k to be Euclidean norm and ⊗ stands
for the Kronecker product.

7
2. PROBLEM FORMULATION AND PRELIMINARIES

2.1. The Model of Quadrotor UAV

Consider the formation containing N quadrotor UAVs, the reference co-

ordinate system for ith quadrotor UAV is illustrated in Fig. 1.

F4 F1

Z Bi

YBi X Bi
F3 F2
OBi
Body-fixed
frame
ZI
i

YI XI
i i
OI
Earth-fixed frame

Figure 1: Reference coordinate system.

Define FI XI YI ZI as the Earth-fixed inertial frame, FB XBi YBi ZBi to be the

i
Body-fixed frame with OB as the mass center of ith quadrotor UAV. Xi =
[xi , yi , zi ]T represents the position of OB
i
expressed in FI . Θi = [φi , θi , ψi ]T
denotes the UAV’s attitude orientation in FI , where φi ∈ R, θi ∈ R, ψi ∈ R
are respectively represented as the roll angle, the pitch angle and the yaw
angle. The rotation matrix Ri (Θi ) ∈ R3×3 from FI to FBi can be described
as  
r11 r12 r13
 
 
Ri (Θi )=  r21 r22 r23  (1)
 
r31 r32 r33
Define C(·) = cos(·) and S(·) = sin(·). There exist r11 = Cψi Cθi , r12 =
Cψi Sθi Sφi − Sψi Cφi , r13 = Cψi Sθi Cφi + Sψi Sφi , r21 = Sψi Cθi , r22 =

8
Sψi Sθi Sφi + Cψi Cφi , r23 = Sψi Sθi Cφi − Cψi Sφi , r31 = −Sθi , r32 = Cθi Sφi
and r33 = Cθi Cφi .
The position and attitude dynamics for ith quadrotor UAV can be math-
ematically modeled as [27]

mi Ẍi + mi gzI = ui + di (2)

Ji (Θi )Θ̈i + Ci (Θi , Θ̇i )Θ̇i = τi + τdi (3)

where mi ∈ R is the UAV total mass. g ∈ R is the gravitational acceler-

ation constant. zI = [0, 0, 1]T . ui ∈ R3×1 is the total translational con-
trol force, τi ∈ R3×1 is the total rotational control torque, di ∈ R3×1 and
τdi ∈ R3×1 denote the disturbance force and the disturbance torque, respec-
tively. Ji (Θi ) ∈ R3×3 is rotational inertia matrix and represented as
 
J11 0 J13
 
 
Ji (Θi ) =  0 J22 J23  (4)
 
J31 J32 J33

Introduce the positive-definite inertia tensor matrix as I =

diag{[I11 , I22 , I33 ]T }. There exist J11 = I11 , J13 = −I11 Sθi , J22 = I22 C 2 φi +
I33 S 2 φi , J23 = (I22 − I33 )Cφi Sφi Cθi and J33 = I11 S 2 θi + I22 S 2 φi C 2 θi + I33 .
The gyroscope and centrifugal terms Ci (Θi , Θ̇i ) ∈ R3×3 is given as

1 ∂
Ci (Θi , Θ̇i ) = J˙i (Θi ) − (Θ̇T Ji (Θi )) (5)
2 ∂Θi i

Let wi1 , wi2 , wi3 , wi4 be the speeds of four motors for the ith quadrotor
UAV, respectively. The total lift force Ti and control torque τi of quadrotor
UAV can be generated by adjusting the speed of the four motors. Then, Ti

9
and τi in (3) can be described as
    
2
Ti k kif kif kif wi1
   if  
    2 
 τi1   kif li 0 −kif li 0   wi2 
 =   (6)
    2 
 τi2   0 kif li 0 −kif li   wi3 
    
2
τi3 kiτ −kiτ kiτ −kiτ wi4

where li is the distance between any rotor and the mass center. kif ∈ R and
kiτ ∈ R denote the drag force coefficient and the reverse moment coefficient,
which mainly depends on air density, the number and type of blades and
propeller radius.
Let Ui = [0, 0, Ti ]T , the highly coupled relationship between Ui and ui
in (2) can be achieved by
ui = Ri (Θi )Ui (7)

Assumption 1. For the ith (i = 1, 2, . . . , N ) quadrotor UAV, suppose the

disturbance variables di and τdi in (2), (3) are bounded by the constant πi
and πdi , there exist kdi k ≤ πi and kτdi k ≤ πdi .
Remark 1. J (Θi ) should be positive definite all the time, which means
det(J (Θi )) = I11 I22 I33 C(2θ) > 0. Such that

θ(t) 6= M π/2 ∀t ≥ 0 (8)

where M should be odd integer. Then, the initial pitch angle θ needs to
satisfy −π/2 < θ(0) < π/2, and the attitude controller is bound to keep
−π/2 < θ(t) < π/2 all the time. It should be noted that the constraint of
θ(0) is easy to satisfy according to initial placement of the UAV.

10
2.2. Graph Theory

A digraph G = (V , ε, A) can be utilized to simulate the information

interaction in formation with n agents. V = {v1 , v2 , . . . , vn }, ε ⊆ V × V
and A = [aij ] ∈ Rn×n are respectively denoted as vertex set with n nodes,
link edges set and weighted adjacency matrix. If the agent j can receive
the information from the agent i (i 6= j), it means (vj , vi ) ∈ ε, and its
corresponding weighted adjacency satisfies aij > 0, else (vj , vi ) ∈
/ ε nor
aij = 0. Moreover, for a digraph G with N directed sequence nodes, define
the Laplacian matrix of G as L = [lij ]N ×N , where lij = −aij , i 6= j, and
P
lii = Nj=1 aij for i = 1, 2, . . . , N , j = 1, 2, . . . , N .

For a digraph Ḡ with N + 1 agents, suppose there exists directed path

from one node(root node) to all the other nodes, the root node can be called
leader labelled 0, and others are called followers labelled i = 1, 2, . . . , N
in sequence. If the leader has direct access to ith follower, define ai0 = 1,
else ai0 = 0. The information interaction among the leader agent with the
follower agents can be described as B = diag(a10 , a20 , . . . , aN 0 ). Then, we
define Laplacian matrix H = L + B to be the interaction of Ḡ, where L is
the interaction Laplacian matrix among N follower agents above.

2.3. Mathematical Preliminaries

Lemma 1. [23]: For a positive-definite Lyapunov function V (x), if it satisfies

V̇ (x) + c(V (x))d + ςV (x) ≤ 0 with x 6= 0, c > 0, ς > 0 and 0 < d < 1. Then,
V (x) can be fast-finite-time stability, and Tf is the setting time needed to
reach V (x) = 0
1 cV (0)1−d +ς
Tf ≤ 1n (9)
c(1 + d) ς

11
Lemma 2. [40]: For ξi ∈ R, i = 1, 2, . . . , n, if ξ1 , ξ2 , . . . , ξn ≥ 0 and 0 < r ≤
P r P
1, then [ ni=1 ξi ] ≤ ni=1 ξi r .

2.4. Artificial Potential Function

The concept of the artificial potential function (APF) is derived from

physics, its main function is to drive the object under the potential action
always towards the direction of a decreasing potential field (negative gradient
of the potential field), so as to reach the local minimum potential energy.
Consider the collision problem in quadrotor UAVs formation tracking sys-
tem in this paper, the possible collision involves several situations, including
the collision between the follower UAV and the target leader, the collision
between any two UAVs in the formation, as well as the collision between the
follower UAV and obstacles. The aim is to drive UAV from collision field
to safe area in time. In view of the repulsive APF proposed by [36] is an
effective solution, the new comprehensive repulsive APF is designed as
n−1 X
X n n
X n
X
U (Xi , Xt , Xo ) = KF Uij + KT Uit + Ko Uio (10)
i=1 j>i i=1 i=1

2
n
X kXi − Xj k2 − δij2
Uij =
2 ,
kXi − Xj k2 − d2ij
j>i

2
kXi − Xt k2 − δit2
Uit =
2 , (11)
kXi − Xt k2 − d2it
2 2


kXi − Xo k2 − δio
Uio =
2 .
kXi − Xo k2 − d2io
where Uij is the interaction potential field (IPF) between ith and j th follower
UAVs within the formation, Uit is the IPF between ith follower UAV and the

12
leader UAV, Uio is the IPF between ith follower UAV with the obstacle. Xo
and Xt denote the position state of obstacle and target leader expressed in
FI , respectively. δij , δit , δio represent the corresponding potential function
sensing radius, respectively. dij , dit , dio describe the corresponding minimum
danger radius of potential function for collision avoidance, respectively. KF ,
KT , KO are the non-negative weight coefficients, which are selected as follows

 0 kXi − Xj k ≥ δij
KF = ,
 κ kXi − Xj k < δij
ij

 0 kXi − Xt k ≥ δit
KT = , (12)
 κ kXi − Xt k < δit
it

 0 kXi − Xo k ≥ δio
KO = .
 κ kX − X k < δ
io i o io

where κij , κit and κio are positive constants.

For (11), some relative properties can be given as

∇Xi Uin (kXi − Xn k) = −∇Xn Uin (kXi − Xn k) (13)

∇Xin Uin (kXi − Xn k) = (Xi − Xn ) gin (kXi − Xn k) (14)

2 2
4(δin − d2in )(kXi − Xn k − 2
δin )
gin (kXi − Xn k) = 3 (15)
(kXi − Xn k2 − d2in )
where n can represent j, t, o. The mark ∇X U denotes the gradient of U
with respect to X. Thus, the gradient of U (Xi , Xt , Xo ) with respect to
Xi , Xt and Xo can be described as
Xn
∇Xi U = KF (Xi − Xj )gij (kXi − Xj k)
j=1, j6=i
(16)
+ KT (Xi − Xt )git (kXi − Xt k)

+ Ko (Xi − Xo )gio (kXi − Xo k)

13
 n
X
∇Xt U = −KT (Xi − Xt )git (kXi − Xt k) (17)
i=1
n
X
∇Xo U = −Ko (Xi − Xo )gio (kXi − Xo k) (18)
i=1

Using (16), (18) and (17), it yields

n
X n
X n
X
∇Xt U + ∇Xo U = − ∇Xi U + KF (Xi − Xj )gij (kXi − Xj k)
i=1 i=1 j=1, j6=i
(19)
In view of (13) and (14), one can further obtain
n
X n
X
KF (Xi − Xj )gij (kXi − Xj k) = 0 (20)
i=1 j=1, j6=i

Based on (19) and (20), an important property about (10) can be obtained
as
n
X
∇Xt U + ∇Xo U = − ∇Xi U (21)
i=1

Assumption 2. To ensure safety, suppose the initial distance must satisfy

kXi − Xj k > dij , kXi − Xt k > dit , kXi − Xo k > dio .

2.5. Control Objective

Consider the UAVs formation tracking system given by (2), (3) and sat-
isfying Assumption 2, the control objective is to design observer based finite-
time distributed position formation tracking control algorithm u and ob-
server based finite-time attitude tracking control algorithm τi to implement
the following several concrete goals.
(1) lim kXi − Xt − hit k = 0, ∀i ∈ N .
t→T3

(2) lim Ẋi − Ẋt = 0, ∀i ∈ N .
t→T3

14
 (3) lim kψi − ψt k = 0, ∀i ∈ N .
t→T3
(4) ∀t ≥ 0, kXi − Xj k > dij , kXi − Xt k > dit , kXi − Xo k > dio .
(5) t ≥ T2 , kXi − Xj k ≥ δij , kXi − Xt k ≥ δit , kXi − Xo k ≥ δio .
(6) lim U̇ = 0 and lim U = 0.
t→T2 t→T2
T
where [h , 0T
3N ] ∈R 6N
, h = [hT T T T
1t , h2t , . . . , hN t ] and ||hit || > δit . h repre-

sents the whole designed configuration formation. T2 and T3 are the globally
finite time.
Remark 2. (1), (2) and (3) are the basic control objectives for quadrotor
UAVs formation tracking, including forming the formation, position tracking,
velocity tracking, and the synchronization and tracking of ψi . (4), (5) and
(6) are the collision avoidance control objective part. (4) is the most basic
collision avoidance objective. That is to ensure no collision occurred in the
whole process. (5) is the purpose of effective collision avoidance. That is
to guarantee all the UAV stay in safe area after finite time. (6) means that
when the phased collision avoidance is completed, the correlation potential
field will no longer have an effect on the formation.

3. FORMATION TRACKING CONTROLLER DESIGN

The strategy of observer-controller is adopted to accomplish high preci-

sion formation tracking and collision avoidance control for quadrotor UAVs.
We first design two high precision finite-time disturbance observers to re-
spectively estimate di and τdi . Then, combined with the reconstruction from
observers, the full tracking controllers are designed, which are synthesized
with distributed position formation tracking collision avoidance control algo-
rithm and attitude tracking control algorithm. Finally, Lyapunov-like stabil-

15
ity analysis ensures the whole closed-loop control system to be globally fast
finite-time stable.

3.1. Finite-time Disturbance Observer for Disturbance Force

For the purpose to design disturbance observer to precisely estimate di
in (2). The two new variables xi1 = Xi and xi2 = Ẋi are introduced. And
there exists
ẋi1 = xi2 (22)
1 1
ẋi2 = −gzI + ui + di (23)
mi mi
Then, a state measurable auxiliary variable xia is further introduced and
updated as
1 i
ẋia = −gzI + ui + k11 xie (24)
mi
with xia ∈ R3×1 , k11
i i
∈ R, k11 > 0 and xie = xi2 − xia . Let x̂ie and dˆi be
the observed values of xie and di . Their corresponding observation error are
expressed as x̃ie = xie − x̂ie and d˜ie = di − dˆi , respectively.
i
Then, continue to define another variable yi1 = k12 xie , and develop x̂ie
as
1
x̂˙ ie = − k12
i i i i i
k13 x̂ie + i ẏi1 +k13 yi1 +k14 sigα11 (x̃ie ) (25)
k12
Then

ẋie − x̂˙ ie = ẋie − (k12

i i i i
k13 x̃ie +ẋie +k14 sigα11 (x̃ie ))
i
(26)
i i i
= − k12 k13 x̃ie − k14 sigα11 (x̃ie )

Theorem 1. Consider the position dynamics (2) under Assumption 1, a

finite-time disturbance observer is designed as
mi i i
dˆi = i (k11 k12 x̂ie + ẏi1 ) (27)
k12

16
 i i i i
with k11 ∈ R, k12 ∈ R, k13 ∈ R, k14 ∈ R all being positive constants and
i
0 < α11 < 1. Then dˆi can precisely estimate di in finite time. Thus, d˜ie ≡ 0
after finite time.
Proof. Consider a Lyapunov function as follows

1
VXi = x̃2ie (28)
2

Differentiating (28) and substituting (26) yields

1+αi11
i
V̇Xi = − k12 i
k13 kx̃ie k2 − k14
i
(kx̃ie k2 ) 2

(29)
≤ − ςVXi − c VXi d

i i i 1+αi11
Let 0 < ς ≤ k12 k13 , c = k14 and d = 2
. According to Lemma 1,
1−d
VXi ≡ 0 can be reached in fast finite time TXi with TXi ≤ 1
c(1+d)
1n cV (0)ς +ς
.
Such that x̃ie ≡ 0 for t ≥ TXi . Applying (23), (24) and (27), it yields
mi i i
d˜ie = mi (ẋie +k11
i
xie ) − i (k11 k12 x̂ie + ẏi1 )
k12 (30)
i
=mi k11 x̃e

Substituting the conclusion of (29), it attains

d˜ie = mi k11
i
lim x̃e ≡ 0 (31)
x→tXi

3.2. Finite-time Disturbance Observer for Disturbance Torque

For the purpose of designing disturbance observer to precisely estimate

τdi in (3), an auxiliary vector Ja (Θi ) = Ji (Θi )Θ̇i is introduced and updated
as
J˙a (Θi ) = τi + τdi − p(Θi , Θ̇i ) (32)

17
where p(Θi , Θ̇i ) = Ci (Θi , Θ̇i )Θ̇i − J˙i (Θi )Θ̇i . Redefine a new observable
variable as
Z t
i i
θie (t) = k21 (τi − p(Θi , Θ̇i ) − θie (`)d` − k21 Ja (Θi )) (33)
0

i i
with k21 ∈ R and k21 > 0. (33) can be further updated as

i i
θ̇ie = −k21 θie − k21 τdi (34)

Let θ̂ie and τ̂di be the observed values of θie and τdi . The corresponding
observation error are expressed as θ̃ie = θie − θ̂ie , τ̃di = τdi − τ̂di , respectively.
i
Continue to define another variable yi2 = k22 θie , θ̂ie can be further de-
veloped as

˙ i i 1 i i i
θ̂ ie = − k22 k23 θ̂ie + i ẏi2 +k23 yi2 +k24 sigα21 (θ̃ie ) (35)
k22
Theorem 2. Consider the attitude dynamics (3) under Assumption 1, design
a finite-time disturbance observer as
1 i i
τ̂di = − i i
(k21 k22 θ̂ie + ẏi2 ) (36)
k21 k22
i i i i
with k21 ∈ R, k22 ∈ R, k23 ∈ R, k24 ∈ R all being positive constants and
i
0 < α21 < 1. Then τ̂di can precisely estimate τdi in finite time.
The proof of Theorem 2 is similar to Theorem 1, which is omitted here.
Define TΘi as the finite time calculated from the stability analysis, there-
fore, there exists θ̃ie ≡ 0 for t ≥ TΘi . Finally, it yields

τ̃di = lim θ̃ie ≡ 0 (37)

x→TΘi

Eventually, from (31) and (37), the disturbance estimation errors d˜ie and
τ̃di is fast finite-time stable. The external disturbance di and τdi can be

18
precisely estimated by (27) and (36) with zero estimation error after finite
time.
Remark 3: The proposed observer can be used to estimate external dis-
turbances without upper bound (known or even unknown). Compared with
references [27, 28], the proposed observer can precisely estimate the distur-
bances/uncertainties, and the observation error can theoretically converge to
equilibrium point in finite time. It is more suitable for the high frequency
noise disturbance estimation and reconstruction. Compared with the finite
time sliding mode observer [23, 24], whose discontinuous terms (designed by
sign function) will make an unsatisfied chattering impact to system, the pro-
posed observers are continuous with the property of chattering restraining.
Compared with extended state observer (ESO) [21, 22], in which the up-
per bounds of the disturbance and its first derivative need to be known and
bounded, the proposed observers do not require these restricted conditions
in the process of observer design or theoretical proof. Thus, the designed
observer possesses less or conservatism.

4. Global Finite-time Formation Tracking Controller Design

In this section, a novel sliding mode surface-like variable based control

strategy combining APF is adopted for the distributed globally finite-time
position formation tracking controller design. At the same time, the fast
terminal sliding mode (FTSM) control strategy [39] is applied for attitude
synchronization and tracking.
During the process of collision avoidance, dynamic obstacles are usu-
ally encountered. However, the velocity information of obstacles can not be

19
sensed or received by all follower UAVs in timely and accurately. It is more
challenging. Consider the velocity of obstacles should be bounded in the
UAV’s detection range, reasonable assumption is represented as

∀t ≥ 0, ||Ẋo || ≤ ||χ|| (38)

where χ = [χ, χ, χ]T , χ is a positive constant, and greater than the maxi-
mum of obstacle velocity in each position direction.
Based on relative adjacent state information, the overall position forma-
tion tracking error can be redefined.
Firstly, define two auxiliary variables eXi , ewi . eXi represents the overall
position formation tracking error of the ith follower UAV, ewi denotes position
tracking error between the ith follower UAV and leader.
P
eXi = N j=0 aij (Xi − Xj − hij ),
(39)
ewi = Xi − Xt − hit

where hij = hit − hjt represents the desired relative distance between the ith
UAV and the j th UAV.
Suppose h to be a constant vector, the entire position formation tracking
error eX and velocity tracking error ėX can be reformulated as

eX = (H ⊗ I3 )(X − 1N ⊗ Xt − h) = (H ⊗ I3 )ew
(40)
ėX = (H ⊗ I3 )(Ẋ − 1N ⊗ Ẋt )

where eX = [eT T T T T T T T
X1 , eX2 , . . . , eXN ] , ew = [ew1 , ew2 , . . . , ewN ] , X =

[X1T , X2T , . . . , XNT ]T .

Consider the position formation tracking with collision avoidance, a novel
distributed sliding mode surface-like variable is designed as


sX = σ β1 eX + asigk (eX ) + ėX + ζ∇X U + (H ⊗ I3 ) · Q (41)

20
with a > 0, β1 > 0, ζ > 0 and 0 < k < 1. sX = [sT T T T
1X , s2X , . . . , sN X ] ,

∇X U = [∇X1 U, ∇X1 U, . . . , ∇XN U ]T , Q = [q1 , q2 , . . . , qN ]T . σ =

 T

diag [diag(σ1 )T , diag(σ2 )T , . . . , diag(σN )T ] , and satisfies
3N ×3N

 diag(0 ) if ∇ U 6= 0
3×1 Xi
σi =
 diag(1 ) otherwise
3×1
 f or i = 1, 2, · · · N (42)
 _
q i if ∇Xo U 6= 0
qi =
 0 otherwise
!
P
N .
diag KT ∇Xt Uit −KF ∇Xi Uij X t +Ko χdiag(∇Xo Uio )sign(∇Xo Uio )
_ j=1, j6=i
.
with q i = diag(∇Xi U )
+ X t.
Define a new variable as

ki = qi − Ẋt (43)

Let K = [k1 , k2 , . . . , kn ]T . Taking the time derivative of sX and insert-

ing (40) to obtain
 
k−1
ṡX = σ β1 ėX + akdiag(|eX | )ėX + (H ⊗ I3 ) · Ẍ + Ṗ (44)

d
with Ṗ = dt
(ζ∇X U + (H ⊗ I3 ) · K). Let ṡX = 0 and apply the reaching law
in [38], the position formation tracking control algorithm can be designed as
h i
u = − M [H ⊗ I3 ]−1 σ(β1 IN ⊗ I3 +akdiag(|eX |k−1 ))ėX + Ṗ
(45)
− M [H ⊗ I3 ]−1 (η1 sX + η2 sigα1 (sX )) + M g(1N ⊗ zI ) − dˆ

with u = [uT T T T ˆ
1 , u2 , . . . , uN ] , M = diag(m1 , m2 , . . . , mN ) ⊗ I3 , d =

[dˆT ˆT ˆT T
1 , d2 , . . . , dN ] , η1 > 0, η2 > 0 and 0 < α1 < 1.

As for the attitude control part, in view of the quadrotor UAV dynamical
structure analysis in [23], by using a given desired ψi , the corresponding

21
desired φid , θid can be determined as
 
Uia1 Sψid −Uia2 Cψid
φid = arcsin ||Uia ||
,
 
θid = arctan Uia1 CψidU+Uia3
ia2 Sψid
, (46)

Uia = [Uia1 , Uia2 , Uia3 ]T = (Ri (Θi )Ui )/mi .

For the sake of attitude fast synchronization and tracking in attitude

closed-loop system, the FTSM control strategy is applied.
Firstly, define the attitude tracking error for ith UAV as

eiΘ = Θi − Θdi (47)

Then, the FTSM surface is introduced as

siΘ = β2 eiΘ + bsigλ (eiΘ ) + ėiΘ (48)

with b > 0, β2 > 0 and 0 < λ < 1.

Taking the time derivative of siΘ to obtain

ṡiΘ = β2 ėiΘ + bλdiag(|eiΘ |λ−1 )ėiΘ + ëiΘ (49)

Let ṡΘ = 0 and choose the similar reaching law as (45), the attitude
tracking control algorithm can be designed as
 
λ−1 α2
τi = − Ji (Θi ) (β2 I3 + bλdiag(|eiΘ | ))ėiΘ + η4 siΘ + η5 sig (siΘ )
(50)
+ Ji (Θi )Θ̈di + Ci (Θi , Θ̇i )Θ̇i − τ̂di

with η4 > 0, η5 > 0 and 0 < α2 < 1.

Theorem 3. Consider N quadrotor UAVs formation tracking control system
governed by dynamics (2), (3) under the Assumption 1, 2. If the control
algorithms are designed as (45), (50) and satisfy the following conditions

22
 2
λmax ((H⊗I3 )−1 )
(1) η1 ζ ≥ 4λmin ((H⊗I3 )−1 )
;
(2) The dynamic obstacle in each phase can only affect formation in finite
time.
Then, the control objectives in Section 2.5 can be achieved. That is to say,
all the quadrotor UAVs can track the target with specific formation config-
uration and avoid dynamic obstacle safely in globally finite time. Moreover,
the rapid attitude synchronization and tracking is also can be achieved.
Proof. Consider a comprehensive Lyapunov-like function as follows
N
1 1X T
V = sT
X sX + s siΘ +U (51)
2 2 i=1 iΘ

The derivative of V with time yields

N
X
V̇ = sT
X ṡX + sT
iΘ ṡiΘ +U̇ (52)
i=1

n
X
U̇ = (∇Xi U T Ẋi ) + ∇Xt U T Ẋt + ∇Xo U T Ẋo (53)
i=1

Substituting (49), (50) into (52) yields

V̇ ≤ λmax [H ⊗ I3 ]M −1 ||sT ˆ 2 1+α1
X (d − d)|| − η1 sX − η2 ||sX ||
N
X
(54)
+ ||sT
iΘ (τ di − τ̂ di )|| − η 4 s 2
iΘ − η5 ||s iΘ ||1+α2
+U̇
i=1

Combining (27) and (36), it is reasonable to obtain

N
X
V̇ ≤ −η1 s2X − η2 ||sX || 1+α1
− (η4 s2iΘ + η5 ||siΘ ||1+α2 )+U̇ (55)
i=1

Then, the subsequent proofs can be discussed in two cases, depending on

whether ∇Xo U in (42) is 0 or not.

23
Case 1. If ∇Xo U = 0, multiplying both sides of (43) by ∇Xi U T and sum it
up over all i, there exists
N
X N
X
T
∇Xi U ki = − ∇Xi U T Ẋt (56)
i=1 i=1

Multiplying both sides of (21) by Ẋt and integrating (56), one can acquire
N
X
∇Xi U T ki = ∇Xt U T Ẋt + ∇Xo U T Ẋo (57)
i=1

Case 2. If ∇Xo U 6= 0, taking the same operation as case 1 and substituting

(20), it is obvious to obtain

N
X
∇Xi U T ki = ∇Xt U T Ẋt + |∇Xo U T |χ (58)
i=1

Then, whether ∇Xo U in (16) is 0 or not, it always satisfies

N
X
T T
∇Xt U Ẋt + ∇Xo U Ẋo ≤ ∇Xi U T ki (59)
i=1

Substituting (39), (43), (53), (57) and sliding mode surface-like variable

24
in (41) to (55), one can obtain
N
X
V̇ ≤ − η1 s2X − η2 ||sX ||1+α − (η4 s2iΘ + η5 ||siΘ ||1+β )
i=1
N
X
+ ∇Xi U T (Ẋi +qi − Ẋt )
i=1
N
X
1+α
=− η1 s2X − η2 ||sX || − (η4 s2iΘ + η5 ||siΘ ||1+β ) (60)
i=1

+ ∇X U T (H ⊗ I3 )−1 (ėX + (H ⊗ I3 ) · Q)
N
X
1+α
≤− η1 s2X − η2 ||sX || − (η4 s2iΘ + η5 ||siΘ ||1+β )
i=1



+ ∇X U T (H ⊗ I3 )−1 sX − ζ∇X U − σ eX + asigk (eX )

From (42), there exists ∇X U T (H ⊗ I3 )−1 σ = 01×3N , then

V̇ ≤ − η1 s2X + ∇X U T (H ⊗ I3 )−1 (sX − ζ∇X U )

≤ − η1 s2X + λmax (H ⊗ I3 )−1 |∇X U T | · |sX | − λmin (H ⊗ I3 )−1 ζ∇X U 2
 
h i |sX |
≤ − |sT T
X | |∇X U |
Ω 
|∇X U |
(61)
 
λmax ((H⊗I3 )−1 )
η1 −
where Ω =  2 .
λmax ((H⊗I3 )−1 ) −1

− 2
λ min (H ⊗ I 3 ) ζ
To ensure V to be monotone decreasing and converge to 0, according to
2
λmax ((H⊗I3 )−1 )
Schur Lemma [19], there must exist η1 ζ ≥ 4λ (H⊗I )−1 , such that Ω ≥ 0
min ( 3 )
and V̇ ≤ 0 are permanent. Then, the whole control system is asymptotically
stable. In consequence, once the arbitrary UAV moves into the sensing field
of APF, the system will respond quickly to drive it to the safe area. The
results can be reached as follows

25
 (i) ∀t ≥ 0, kXi − Xj k > dij , kXi − Xt k > dit , kXi − Xo k > dio . It
ensures that no collisions occurred all the time.
(ii) ∇Xi U =0, U = 0 and U̇ = 0. As V̇ ≤ 0, then V → 0. Correspondingly,
∇Xi U → 0 and U → 0. It means that the UAV is moving away from the
APF sensing field. Once the UAV leaves the sensing field, it has ∇Xi U = 0,
U = 0 and U̇ = 0, such that the collision avoidance mission is accomplished.
(iii) t → ∞, kXi − Xj k ≥ δij , kXi − Xt k ≥ δit , kXi − Xo k ≥ δio . It
means UAV is bound to be driven away from the sensing field by APF to
avoid collisions.
Once the result (ii) is satisfied, the original problem of formation tracking
with collision is transformed into the formation tracking problem without
collision. Then, (41) is FTSM. (41) and (45) can be expressed as

sX = β1 eX + asigk (eX ) + ėX (62)

 
u = − M [H ⊗ I3 ]−1 β1 IN ⊗ I3 +akdiag(|eX |k−1 ) ėX + M · 1N ⊗ Ẍt

+ M g(1N ⊗ zI ) − dˆ − M [H ⊗ I3 ]−1 (η1 sX + η2 sigα1 (sX ))

(63)
Meanwhile, (55) can be finally expressed as
N
X N
X
V̇ ≤ − (η1 s2X + η4 s2iΘ ) − (η2 |sX |1+α1 + η5 |siΘ |1+α2 )
i=1 i=1
( N
)
1 1 X
≤− 2η1 · sT
X sX +2η4 · sT
iΘ siΘ (64)
2 2 i=1
  1+α 1 N
! 1+α 2

1X T
2
1+α1 1 T 2 1+α2
−2 2 η2 · s sX −2 2 η5 · s siΘ
2 X 2 i=1 iΘ
Let α1 = α2 = α, using the Lemma 2, (55) is further converted to
1+α
V̇ ≤ −η7 V − η8 (V ) 2 (65)

26
 n 1+α 1+α
o
1+α
where 0 < 2
< 1, η7 = min {2η1 , 2η4 } and η8 = min 2 2 η2 , 2 2 η5 .
According to Lemma 1, sX and siΘ can converge to 0 in ts ≤
1− 1+α
In η8 V (0)+η7
1 2
η8 (1+ 1+α ) η7
on the condition of U = 0 and ∇Xi U = 0. Con-
2

sider sX = 0 and siΘ = 0 for t ≥ ts , there exist β1 eX + asigk (eX ) + ėX = 0

in (62) and β2 eiΘ + bsigλ (eiΘ ) + ėiΘ = 0 in (48). Following Lemma 3 in
[39], the formation tracking errors (eX , ėX ) and (eiΘ , ėiΘ ) can reach (0, 0)
in finite-time tX and tΘ , respectively. Meanwhile, the settling time satisfy
1−k 1−λ
tX ≤ 1
a(1+k)
In aV β(0)+β
1
1
and tΘ ≤ 1
b(1+λ)
In bV β(0)+β
2
2
, respectively.
Therefore, under the condition of U = 0 and ∇Xi U = 0, employing (50)
and (63), all UAVs will quickly form formation and track the leader within
finite time. The settling time satisfies

T1 ≤ ts + tX + tΘ (66)

In view of the result in [41, 42], when a control system simultaneously

possesses globally asymptotic stability and locally finite-time stability, it can
be considered as globally finite-time stable . When ∇Xi U 6= 0, (45) is asymp-
totically stable. It means sX , ∇Xi U and U are bound to approach 0, 0 and
0. Once ∇Xi U = 0, (45) is equivalent to (63), the finite time stability of
the position loop control system can be achieved by (45). Thus, (45) and
(50) can implement globally finite-time stability of the full control system.
Suppose the time for dynamic obstacle in each phase acting on formation is
T2 , the time T3 for full control system to reach globally finite time stable can
be determined as
T3 ≤ T2 + ts + tX + tΘ (67)

Remark 4. For (45), there exist circumstances that as eX (t) → 0 and

27
ėX (t) 6= 0, diag(|eX (t)|k−1 )) · ėX (t) → ∞, which means singularity has
occurred. Without loss of generality, the saturation function [43] is adopted
to avoid singularity and designed as

 u diag(|eX |k−1 )) · |ėX | ≤ us
f
sat(uf , us ) = (68)
 u sign(ė ) diag(|eX |k−1 )) · |ėX | > us
s X

where uf = diag(|eX |k−1 )) · ėX , us is a constant matrix. Relevant proof is

detailed in [43]. Then the sliding mode surface-like vector in (41) still can
reach 0 in globally finite time. The control algorithm in (45) can be improved
as follows
h i
u = − M [H ⊗ I3 ]−1 σ(β1 IN ⊗ I3 ėX +aksat(uf , us )) + Ṗ
(69)
− M [H ⊗ I3 ]−1 (η1 sX + η2 sigα1 (sX )) + M g(1N ⊗ zI ) − dˆ

The singularity problem is also appeared in (50) and (63), to which similar
solution can be applied.
Remark 5. Different from utilizing the repulsive APF and attractive APF
to realize the collision avoidance and formation tracking in the literatures
[32, 33, 36], this paper only uses the comprehensive repulsion APF for col-
lision avoidance. As for the part of formation tracking, the sliding mode
surface-like based distributed output feedback control algorithm is further
adopted. It should be noted that, due to the application of the potential
function formation method, only the asymptotic stability can be achieved by
using the output feedback controller in the references above. The proposed
sliding mode surface-like is skillfully integrated by the gradient of APF and
FTSM, then, the designed output feedback controller can realize the globally
finite-time position formation tracking control with collision avoidance. Once

28
all the relative distances (UAV and UAV, UAV and target, UAV and obsta-
cle) meet the safe distance, the position closed-loop system can achieve the
finite time stability. In addition, the distributed position formation tracking
controller still possesses the properties of FTSM in terms of fast convergence
rate, lower steady state tracking error and stronger robustness.

5. SIMULATION AND ANALYSIS

This section validates the effectiveness of the proposed formation track-

ing control algorithm of observer-controller structure by numerical simula-
tion. Assuming there exists 4 followers and 1 virtual leader, their networked
formation communication topology digraph is described in Fig.2.
1 2

4 3

Figure 2: Formation communication topology digraph.

The corresponding Laplace matrix

4 of the above diagraph
3 is represented
as
 
1 2
2 0 0 −1
 
 0 
 −1 1 0 0 
H=

.

 −1 −1 2 0 
 
0 −1 −1 2
The relevant model parameters and external disturbances sustained
for ith quadrotor UAV are described as mi = 1.2kg, I =
diag[0.015, 0.015, 0.026](N.m), τdi = [0.1sint, 0.1cost, 0.2sin(2t)]T (N.m).

29
  
sin t + 0.5 cos(0.5t) + 0.5 sin(0.7t)
 
 
di =  0.5 sin(0.2t) + 0.5 cos(0.4t)  (N).
 
cos t + 0.5 sin(0.5t)
Set desire yaw angle as ψid = 0.1 rad. The desired relative formation
configuration h is hT T T T T T
1 = [−3, 3, 0] , h2 = [3, 3, 0] , h3 = [3, −3, 0] ,

hT T T
4 = [−3, −3, 0] . The trajectory of the target Xt = [xt , yt , zt ] is de-

signed as

 8 − 8cos(0.025πt) t ≤ 40s
xt = ,
 16 − 4cos(0.025π(t − 40)) t > 40s
yt = 0,
zt = 2t.

Suppose the moving obstacle is a sphere with radius being 2.2m. it’s
moving trajectory Xo = [xo , yo , zo ]T is described as

 16 t ≤ 57s
xo = ,
 3 − 0.4(t − 57) t > 57s
yo = 0 t ≥ 0s ,


 78 t ≤ 32s


zo = 78 + 0.25(t−32) 32s < t ≤ 48s .



 120 t > 48s

The initial position and attitude state values for each follower in dynamics
are set as
X1 (0) = [4, 0, 0]T , X2 (0) = [−2, 0, 0]T , X3 (0) = [2, 0, 0]T , X4 (0) =
[8, 0, 0]T , Ẋi (0) = [0, 0, 0]T , Θi (0) = [0, 0, 0]T , Θ̇i (0) = [0, 0, 0]T .
The parameters for APF are dij = 1, δij = 2.5, dit = 1, δit = 2.5, dio = 1,
δio = 2.5 and κij = 6.5, κit = 6.5, κio = 5.5.

30
 Moreover, the relevant control gains for proposed observer-controller
i
schemes in (27), (36), (45) and (50) are respectively selected as k11 = 0.025,
i i i i i i i
k12 = 8, k13 = 70, k14 = 10, α11 = 0.9, k21 = 0.025, k22 = 1, k23 = 100,
i i
k24 = 1, α21 = 0.9, χ = 0.4, β1 = 10.5, a = 4.1, k = 3/5, ζ = 50, η1 = 1,
η2 = 0.1, α1 = 0.7, β2 = 4, b = 1, λ = 3/5, η4 = 10, η5 = 0.1, α2 = 0.7.

Figure 3: 3D trajectories motion of UAVs, leader and obstacle.

31
 Figure 4: Profile view in X, Z axis. Figure 5: Profile view in Y, Z axis.

i=1 i=2 i=3 i=4

3
5
2

1
0
xie ((m))

yie ((m))

6 -1
-5 2
0 0
-2
-6 -2
-10 -12 -3 -4
0 5 10 0 2 4 5
-4
0 20 40 60 80 90 0 20 40 60 80 90
Time (sec) Time (sec)

7 0.1

0.5
0 0.05
-0.5
4
-1
ψ ie (rad))
z ie ((m))

-1.5
0 5 10 0
2
×10-4
1
0 -0.05 0

-1
-2 -0.1 2 4 6 8 10
0 20 40 60 80 90 0 1 2 3 4
Time (sec) Time (sec)

Figure 6: Fomation tracking errors in xi , yi , zi , ψi .

As is shown in Fig. 3, Fig. 4 and Fig. 5, under the proposed algorithms

in (45) and (50), all the follower UAVs can quickly track the leader with
the desired formation configuration referring to the communication topology
diagraph in Fig. 2. At the same time, it still has the strong ability of au-

32
 i=1 i=2 i=3 i=4
50

u i1 (N)
0

-50
0 10 20 30 40 50 60 70 80

50
u i2 (N)

-50
0 10 20 30 40 50 60 70 80

50
u i3 (N)

-50
0 10 20 30 40 50 60 70 80
Time (sec)

Figure 7: Control forces of UAVs.

i=1 i=2 i=3 i=4
1.5
1
0.5
τ i1 (N.m)

0
-0.5
-1
-1.5
0 10 20 30 40 50 60 70 80

1.5
1
0.5
τ i2 (N.m)

0
-0.5
-1
-1.5
0 10 20 30 40 50 60 70 80

0.5
τ i3 (N.m)

-0.5
0 10 20 30 40 50 60 70 80
Time (sec)

Figure 8: Control torques of UAVs.

tonomous collisions avoidance and fast formation tracking restoration. From

Fig. 6, the time for completing the formation tracking mission from initial
position is less than 15 sec. From Fig. 7, Fig. 8 and Fig. 9, there exists cir-
cumstances that UAV stays or moves into the APF sensing field, then quickly
leaves the APF action area, and finally, all the relative distances about UAV

33
 10
||X 1 - X 2||
8 ||X 1 - X 3||

||Xi - Xj || (m)
||X 1 - X 4||
6
3 ||X 2 - X 3||
4 2.5 ||X 2 - X 4||
2 ||X 3 - X 4||
2
1.5
0 0.5 1 1.5 Sensing radius
0 Danger radius
0 10 20 30 40 50 60 70 80 90 100

10
3
2.5 ||X 1 - X t ||
8
||Xi - Xt || (m)

2 ||X 2 - X t ||
6 1.5 ||X 3 - X t ||
0 0.5 1 1.5
4 ||X 4 - X t ||
Sensing radius
2 Danger radius
0
0 10 20 30 40 50 60 70 80 90 100

10
||X 1 - X o||
8 2.6
||Xi - Xo || (m)

||X 2 - X o||
6 2.4 ||X 3 - X o||
4 ||X 4 - X o||
40 45 50 55 60 65
Sensing radius
2 Danger radius
0
0 10 20 30 40 50 60 70 80 90 100
Time (sec)

Figure 9: Relative distances about UAV to UAV, leader, or obstacle.

to UAV, leader or obstacle are greater than the APF sensing range in spite of
the limitation of UAV’s control force and control torque output. This illus-
trates that the potential function plays an effective role in collision avoidance.
And it further verifies that the control algorithms can effectively achieve col-
lision avoidance function. In summary, the above simulation results can
effectively verify that the proposed control schemes have good disturbance
compensation performance for unknown external disturbances. Meanwhile,
safe and effective formation tracking collision avoidance performance is also
guaranteed even in the environment with dynamic obstacles.

6. CONCLUSION

This paper focuses on the problem of globally finite-time safe formation

tracking control with collision avoidance for quadrotor UAVs subject to ex-
ternal disturbances. Two less conservative finite-time disturbance observers
are firstly proposed to construct the disturbance forces and torques without

34
upper bound, respectively. Then, utilizing the construction from observers,
the sliding mode surface-like variable based distributed position control strat-
egy is proposed for safe position formation tracking and automatical collision
avoidance. Meanwhile, the FTSM based attitude control strategy is applied
for fast attitude synchronization and tracking. The whole closed-loop system
is theoretically proven to be globally finite-time stable through Lyapunov-
like analysis. Considering that actuator failure of UAV will make a serious
difference in formation tracking control system, the future work will further
concentrate on fault tolerant control of UAVs formation tracking.

7. References

References

[1] Zhao. Wu, Karimi H. R., and P. Shi, “Practical trajectory tracking of
random Lagrange systems,” Automatica, vol. 105, pp. 314-322, Jul. 2019.
[2] J. Gu , S. Tao, and Q. Wang, “Multiple Moving Targets Surveillance
Based on a Cooperative Network for Multi-UAV,” IEEE Communica-
tions Magazine, vol. 56, no. 4, pp. 82–89, 2018.
[3] S. Farid, M. Mostafa, and Y. Zhang, “Cooperative Multi-Vehicle Search
and Coverage Problem in an Uncertain Environment,” Unmanned Sys-
tems, vol. 3, no. 1, pp. 35–47, 2015.
[4] H. Kim, L. Mokdad, and J. Ben-Othman, “Designing UAV Surveillance
Frameworks for Smart City and Extensive Ocean with Differential Per-
spectives,” IEEE Communications Magazine, vol. 56, no. 4, pp. 98–104,
Sep. 2018.

35
 [5] K. Z. Y. Ang, X. Dong, and W. Liu, “High-Precision Multi-UAV Team-
ing for the First Outdoor Night Show in Singapore,” Unmanned Systems,
vol. 6, no. 1, pp. 1–27, 2018.
[6] R. Olfati-Saber, and R.M. Murray, “Distributed structural stabilization
and tracking for formations of dynamic multi-agents,” Proceedings of
the 41st IEEE Conference on Decision and Control, vol. 1, pp. 209–215,
2002.
[7] H. Liu, Karimi H. R., and S. Du, “Leader-following consensus of discrete-
time multiagent systems with time-varying delay based on large delay
theory,” Information Sciences, vol. 417, pp. 236–246, 2017.
[8] J. Mao, Karimi H. R., and Z. Xiang, “Observer-Based Adaptive Consen-
sus for a Class of Nonlinear Multiagent Systems,” IEEE Transactions
on Systems, Man, and Cybernetics: Systems, no. 99, pp. 1–8, 2017.
[9] Y. Sun, L. Chen, G. Ma, and C. Li, “Adaptive neural network tracking
control for multiple uncertain Euler-Lagrange systems with communi-
cation delays,” Journal of the Franklin Institute, vol. 354, no. 7, pp.
2677-2698, May. 2017.
[10] Y. Chen, Y. Zhang, and Z. Wang, “An adaptive backstepping design for
formation tracking motion in an unknown Eulerian specification flow-
field,” Journal of the Franklin Institute, vol. 354, no. 14, pp. 6217–6233,
Sep. 2017.
[11] H. Liu, X. Wang, and H. Zhu, “A novel backstepping method for the
three-dimensional multi-UAVs formation control,” 2015 IEEE Interna-
tional Conference on Mechatronics and Automation (ICMA), pp. 923–
928, Aug. 2015.

36
[12] H. Lei, X. Sun, and L. Yan, “Distributed output-feedback formation
tracking control for unmanned aerial vehicles,” International Journal of
Systems Science, vol. 47, no. 16, pp. 3919–3928, 2016.
[13] L. Chen, C. Li, Y. Sun, and G. Ma, “Cooperative impulsive formation
control for networked uncertain Euler-Lagrange systems with communi-
cation delays,” Chinese Physics B, vol. 26, no. 6, pp. 068703, 2017.
[14] L. Chen, C. Li, B. Xiao, and Y. Guo, “Formation-containment control
of networked Euler-Lagrange systems: An event-triggered framework,”
ISA Transactions, vol. 86, pp. 87-97, 2019.
[15] S. J. Yoo, and T. H. Kim, “Distributed formation tracking of networked
mobile robots under unknown slippage effects,” Automatica, vol. 54, no.
C, pp. 100–106, 2015.
[16] D. Wang, Q. Zong, and B. Tian, “Neural network disturbance observer-
based distributed finite-time formation tracking control for multiple un-
manned helicopters,” ISA Transactions, vol. 73, pp. 208–226, 2018.
[17] J. Wang, J. Du, and F. Chen, “Direct adaptive control scheme for a
quadrotor helicopter with actuator failures via quantum logic,” Journal
of Aerospace Engineering, vol. 29, no. 5, pp. 789–796, Sep. 2016.
[18] L. Chen, C. Li, J. Mei, and G. Ma, “Adaptive cooperative formation-
containment control for networked Euler-Lagrange systems without us-
ing relative velocity information,” IET Control Theory and Applications,
vol. 11, no. 9, pp. 1450-1458, 2017.
[19] C. Li, L. Chen, and Y. Guo, “Cooperative surrounding control with
collision avoidance for networked Lagrangian system,” Journal of the
Franklin Institute, vol. 355, no. 12, pp. 5182–5202, Aug. 2018.

37
[20] N. Wang, and Y. Wang, “Fuzzy Uncertainty Observer Based Adaptive
Dynamic Surface Control for Trajectory Tracking of a Quadrotor,” Acta
Automatica Sinica, vol. 44, no. 4, pp. 685–695, 2018.
[21] B. Li, K. Qin, B. Xiao, and Y. Yang, “Finite-time Extended State
Observer based fault tolerant output feedback control for attitude
stabilization,” ISA Transactions, vol. 91, pp. 11–20, 2019, Doi:
10.1016/j.isatra.2019.01.039.
[22] B. Li, Q. Hu, and Y. Yang, “Continuous Finite-time Extended State
Observer based Fault Tolerant Control for Attitude Stabilization,”
Aerospace Science and Technology, vol. 89, pp. 204–213, Jan. 2019.
[23] B. Xiao, X. Yang, and X. Huo, “A Novel Disturbance Estimation Scheme
for Formation Control of Ocean Surface Vessels,” IEEE Transactions on
Industrial Electronics, vol. 64, no. 6, pp. 4994–5003, Jun. 2017.
[24] L. Dou, T. Chen, and Q. Mao, “Disturbance Observer-Based Distributed
Formation Tracking Control for Multiple Quadrotor,” Journal of Tianjin
University(Science and Technology), vol. 51, no. 8, pp. 677–682, Aug.
2018.
[25] B. Li, Q. Hu, Y. Yang, and O. A. Postolache, “Finite-time Disturbance
Observer Based Integral Sliding Mode Control for Attitude Stabilization
under Actuator Failure,” IET Control Theory and Applications, vol. 13,
no. 1, pp. 50–58, Jan. 2019.
[26] B. Li, Q. Hu, G. Ma, and Y. Yang, “Fault Tolerant Attitude Stabi-
lization Incorporating Closed-loop Control Allocation,” IEEE Transac-
tions on Aerospace and Electronic Systems, Published online, 2018, doi:
10.1109/TAES.2018.2880035.

38
[27] B. Xiao, and S. Yin, “A New Disturbance Attenuation Control Scheme
for Quadrotor Unmanned Aerial Vehicles,” IEEE Transactions on In-
dustrial Informatics, vol. 13, no. 6, pp. 2922–2932, Dec. 2017.
[28] B. Xiao, and S. Yin, “Exponential Tracking Control of Robotic Manip-
ulators with Uncertain Kinematics and Dynamics,” IEEE Transactions
on Industrial Informatics, vol. 15, no. 2, pp. 689–698, Feb. 2019.
[29] X. Wang, V. Yadav, and S. N. Balakrishnan, “Cooperative UAV For-
mation Flying With Obstacle/Collision Avoidance,” IEEE Transactions
on Control Systems Technology, vol. 15, no. 4, pp. 672–679, 2007.
[30] L. Dai, Q. Cao, Y. Xia, and Y. Gao, “Distributed MPC for formation
of multi-agent systems with collision avoidance and obstacle avoidance
,” Journal of the Franklin Institute, vol. 354, no. 4, pp. 2068-2085, Mar.
2017.
[31] Y. Xia, X. Na, and Z. Sun, “Formation control and collision avoidance
for multi-agent systems based on position estimation,” ISA Transac-
tions, vol. 61, pp. 287–296, Mar. 2016.
[32] Q. Li, J. Yuan, and H. Wang, “Sliding mode control for autonomous
spacecraft rendezvous with collision avoidance,” Acta Astronautica, vol.
151, pp. 743–751, Oct. 2018.
[33] Q. Li, B. Zhang, and J. Yuan, “Potential function based robust safety
control for spacecraft rendezvous and proximity operations under path
constraint,” Advances in Space Research, vol. 62, no. 9, pp. 2586–2598,
Nov. 2018.
[34] N. Zhou, and Y. Xia, “Coordination control design for formation recon-
figuration of multiple spacecraft,” Aerospace Science and Technology,

39
 vol. 42, no. 15, pp. 2222–2231, 2015.
[35] Z. Shi, J. He, T. Wang, C. Zhou, and J. Guo, “Optimal formation control
and collision avoidance in environment with multiple rectangle obstacles
,” Journal of the Franklin Institute, vol. 355, no. 15, pp. 7626-7642, Oct.
2018.
[36] Q. Hu, H. Dong, and Y. Zhang, “Tracking control of spacecraft forma-
tion flying with collision avoidance,” Aerospace Science and Technology,
vol. 42, pp. 353–364, Apr. 2015.
[37] G. Ma, H. Dong, and Q. Hu, “Fault-tolerant translational control for
spacecraft formation flying with collision avoidance requirement,” Acta
Aeronautica Et Astronautica Sinica, vol. 38, no. 10, pp. 321129, 2017.
[38] S. Yu, X. Yu, and B. Shirinzadeh, “Continuous finite-time control for
robotic manipulators with terminal sliding mode,” Automatica, vol. 41,
no. 11, pp. 1957–1964, 2005.
[39] X. Yu, and Z. Man, “Fast terminal sliding-mode control design for non-
linear dynamical systems,” IEEE Transactions on Circuits and Systems
I: Fundamental Theory and Applications, vol. 49, no. 2, pp. 261–264,
2009.
[40] Q. Zong, and S. Shao, “Decentralized finite-time attitude synchroniza-
tion for multiple rigid spacecraft via a novel disturbance observer,” ISA
Transactions, vol. 65, pp. 150–163, Apr. 2016.
[41] Y. Hong, Y. Xu, and J. Huang, “Finite-time control for robot manipu-
lators,” Systems and Control Letters, vol. 46, no. 4, pp. 243–253, Apr.
2002.
[42] H. Gui, L. Jin, and S. Xu, “Simple finite-time attitude stabilization

40
 laws for rigid spacecraft with bounded inputs,” Aerospace Science and
Technology, vol. 42, pp. 176–186, Jan. 2015.
[43] Y. Feng, X. Yu, and F. Han, “On nonsingular terminal sliding-mode
control of nonlinear systems,” Automatica, vol. 49, no. 6, pp. 1715–1722,
2013.
[44] Q. Hu, J. Zhang, and Y. Zhang, “Velocity-free attitude coordinated
tracking control for spacecraft formation flying,” ISA Transactions, vol.
73, pp. 54-65, 2018.

41