AUT Journal of Modeling and Simulation

AUT J. Model. Simul., 50(2) (2018) 147-156

DOI: 10.22060/miscj.2018.14252.5100

Cooperative Control of Multiple quadrotors for Transporting a Common Payload

R. Babaie, A. F. Ehyaei*
Department of Electrical Engineering, Imam Khomeini International University, Qazvin, Iran

ABSTRACT: This paper investigates the problem of controlling a team of quadrotors that cooperatively Review History:
transport a common payload. The main contribution of this study is to propose a cooperative control Received: 28 March 2018
algorithm based on a decentralized algorithm. This strategy is comprised of two main steps: the first Revised: 19 October 2018
one is calculating the basic control vectors for each quadrotor using Moore–Penrose theory aiming at Accepted: 22 October 2018
cooperative transport of an object and the second one is combining these vectors with individual control Available Online: 25 October 2018
vectors, which are obtained from a closed-loop non-linear robust controller. In this regard, a nonlinear
robust controller is designed based on Second Order Sliding Mode (SOSMC) approach using Extended
Kalman-Bucy Filter (EKBF) to estimate the unmeasured states which is capable of facing external Keywords:
disturbances. The distinctive features of this approach include robustness against model uncertainties quadrotor
along with high flexibility in designing the control parameters to have an optimal solution for the Second Order Sliding Mode Control
nonlinear dynamics of the system. Design of the controller is based on Lyapunov method which can Extended Kalman-Bucy Filter
provide the stability of the end-effecter during the tracking of the desired trajectory. Finally, simulation Cooperative Decentralized Control
results are given to illustrate the effectiveness of the proposed method for the cooperative quadrotors to External Disturbances
transport a common payload in various maneuvers.

1- Introduction advantages such as decreased number of sensors and fast

The cooperative control of multiple vehicle systems despite
its wide range of practical applications requires tackling
important theoretical and practical challenges which have
attracted many researchers in recent years. Formation control
problems for multiple vehicle systems can be categorized
with applications to unmanned aerial vehicles (UAVs),
autonomous underwater vehicles, cooperative transport,
mobile robots, cooperative role assignment and cooperative
search. In this paper, we seek to drive a control algorithm
for cooperation between quadrotors that allow the robots
to control their position and angles to grasp and transport
a common payload in various maneuvers. The controller is
designed to move the object by two or more quadrotors.
To make a framework for interplay between a group of
quadrotors and payload, many control schemes have been
extensively used to solve the problems of creating formation
control for UAVs. Some of them have focused on centralized
and leader-follower approach to access interaction between
cooperative quadrotors and payload [1-7]. Although these
methods have acceptable results on small robotic systems
but suffer from several disadvantages including high
computational complexity of centralized methods in large-
scale systems and possibility of disappearing the group
formation in leader-follower strategy due to not receiving the
position of leader by the followers.
In this regard, there is a plethora of research in cooperative
multi-robot controller design based on decentralized control
methods which can solve a significant number of problems
in cooperative control strategies and benefit also from the
Corresponding author, E-mai: f.ehyaei@eng.ikiu.ac.ir
performance. In [8], the problem of cooperation by a team
of ground robots is addressed under quasi-static assumptions
based on decentralize controllers considering a unique solution
to robot and object motion. Also, transporting a payload
by aerial manipulation using cables based on decentralized
control is studied in [9,10]. In other research, cooperative
aerial towing-based decentralized mechanism is studied
[11]. The authors of [12] employ bilinear matrix inequalities
to present optimal solutions for decentralized nonlinear
multi-agent systems. Despite the numerous advantages of
decentralized methods, there are still some critical issues
related to the performance of the control system in presence
of disturbances and uncertainties.
On the other hand, some of the most popular control schemes
for uncertain nonlinear systems are based on discontinuous
schemes; in particular, sliding mode controllers (SMC)
have shown robustness against uncertainties and matched
disturbances [13-18]. Moreover, the finite-time convergence
of SMC as well as their simplicity have made them suitable
for a large variety of applications such as designing a robust
formation control scheme. A high order sliding mode concept
can effectively reduce the chattering while keeping the
invariant characteristics in the sliding mode. Some successful
implementations of high order (second order) SMC schemes
in UAVs have been reported in [19,20]. Also, in [21] a robust
second order sliding mode controller was proposed for the
attitude stabilization of a four-rotor helicopter. This controller
was able to overcome the chattering phenomena in classical
sliding mode control while preserving the invariance
property of sliding mode. In [22], using an equivalent

147
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100

approach, the performance of control system was improved stabilizes the object along three dimensional trajectories
by using a second-order sliding mode control (SOSMC) for based on a SOSMC incorporated with optimal Moore-
second-order uncertain plants. An adaptive SOSMC with a Penrose theory. We review and analyze experimental study
nonlinear sliding surface was introduced in [23] and a linear with teams of quadrotors cooperatively grasping, stabilizing,
switching surface was proposed for similar underactuated and transporting payloads in Section 5. It includes point-to-
systems in [24]. Wind as a disturbance source has also point path planning and trajectory tracking. Finally, Section
been considered in the flight process of the quadrotor to 6 concludes the paper with some future improvement
demonstrate the robustness of the control algorithm [25,26]. suggestions.
So, SOSMC is a suitable method for controlling uncertain 2- Mathematical Modeling and Coordinate Reference
(and possibly disturbed) nonlinear systems, specially UAVs, Frame
where the necessity of robust tracking schemes, as well as 2- 1- Dynamic Model of single quadrotor
fast convergent algorithms is crucial. Fig.1 shows the configuration of our quadrotor system that can
Hence, in this paper, a decentralized robust optimal controller fly in all directions and has no limit on maneuver. Two main
is designed for a group of UAVs in a formation which enables reference frames are considered: model frame, E = {Ex , E y , Ez },
manipulation of a common payload in three dimensions. Our and the body-fixed reference frame, B = {Bx , By , Bz } . In this
nomenclature, ξ = [ x, y, z ] and η [ϕ ,θ , Ψ ] are the position
T T
cooperative control algorithm is comprised of two parts. In the =
first part, basic control vectors, which guarantee the optimal and orientation vectors in reference E . Fi ( i = 1, 2,3, 4 ) is
performance of the coupled system, are determined using the thrust force produced by four rotors. The full quadrotor
Moore-Penrose theory. In the second part, individual control dynamical model with the x, y, z motions as outcome of a
vectors are designed for each quadrotor which can guarantee pitch, roll and yaw rotation is [27]:
robustness of the coupled system against uncertainties
 ¨   Iy − Iz  jr  l 
= ϕ θψ  Ix  − Ix θΩ + I U 2 + ωϕ  =( ϕ θ ψ + ϕ ψ )
and disturbances. Moreover, a combination of second
order sliding mode controller with an optimal cooperative    x

algorithm is employed to control all the quadrotors such  ¨  Iz − Ix  jr l  ¨
that a soft touching occurs between the object being carried =  θ ϕψ     − ϕΩ + U 3=
 + ωθ  ( ϕ θ ψ − ϕ ψ)
  Iy  Ix y 
and the cooperative quadrotors. A nonlinear filter equivalent  ¨ 
Ix − Iy  l
to well-known Extended Kalman-Bucy filter (EKBF) in a  ψ ϕθ
=     + U 4 + ωψ  =− + ( ϕ θ ) +ω
continuous-time stochastic system has also been developed in   Iz  Iz 
our work to improve the estimation accuracy and robustness (1)
 − 
 implementation
 and Ω + and+ ω  x¨ = cosϕ sinθ cosψ + sinϕ sinψ l U + ω
=
of the system. The simple ϕ
design
 θψ    − θ
 under noisy condition
ϕ  ( ) 1 x

accurate estimation of state variables  mi
make EKBF suitable for robotic  ¨  − 
manipulators and other  ¨ l
industrial applications.=  θ ϕψ   − ϕΩ + + ωθ  y ( cosϕ sinθ sinψ − sinϕ sinψ ) U1 + ω y
3=
    m i
Finally, the proposed controller  successfully rejects the ¨
 −  l
external disturbances (e.g.= ψ ϕθ
wind 
 effect)

for
+ + ωψ
cooperative  z =− g + ( cosϕ cosθ ) U1 + ω z
quadrotors. The other advantages 
of using  approach are
this  mi
from simple implementation and flexible parameter learning where
to design criteria customization. The simulation results show
the effectiveness and superior performance of the proposed    
   b b b b  2
control strategy for transporting a common payload in various
U1  lb 0 Ω1
−lb 0   2 
maneuvers by cooperative quadrotors. = U 2   Ω2  (2)
The subsequent parts of this paper are as it follows. First of    0 −lb 0 lb   2 
all, developing a model for a single quadrotor and modeling U
     Ω 
−k k −k k   32 
3
U 4  
of a team of UAV’s rigidly attached to a payload is discussed Ω4  ,
in Section 2. In Section 3, the Extended Kalman Bucy in which mi is considered as mass of one quadrotor; also k
Filter is described briefly. Section 4 proposes cooperative and b represent the thrust and drag coefficient, respectively.
control algorithms defined with respect to the payload that The input U1 is related to the total thrust; while the inputs
U 2 , U 3 , U 4 are related to the rotations of the quadrotor and
ω x , ω y ,ω z ,ωϕ ,ωθ ,ωΨ are external disturbances, jr denotes
the inertia of the z-axis and Ω = −Ω1 + Ω 2 − Ω3 + Ω 4 is the
overall residual propeller angular speed. The model in Eq. (1)
can be rewritten in a state-space form, X = f ( X,U ) , where
T
 ÿ  
=
X [ x1 ,…….., x12=
]T ϕ , ϕ , θ , θ ,ψ ,ψ , x, x , y, y , z , z 


is the state vector of the system.

Fig.1 Configuration frame of a quadrotor [27]
Assumption 1 .It is assumed that the roll ,pitch and
yaw angle satisfy the conditions ϕ ( t ) < π / 2 , θ ( t ) < π / 2
and , ψ < π for t ≥ 0
2- 2- Dynamic Model of Coupled System
According to Fig.2 , we consider a group of 4 quadrotors
jointed through a fixed interaction to the cross object

148
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100

process and measurement noise. EKBF is designed by

determining the filtering and prediction equations. Consider
a nonlinear stochastic system with uncorrelated Gaussian
process and measurement noise as it follows [29-31]:

dx ( t ) f ( x ( t ) , t ) dt + G ( t ) dw ( t ) , ∀t ∈ 
=
(7)
=Zk h ( xk , t k ) + v k
where x ( t ) is the state vector and Z k is the measurement
vector. w ( t ) and v k denote the state and measurement noise,
a noise that apply to the system state variables in kalman filter
process. In filtering step, the estimation of error covariance
matrix is determined by:

P=
k ( ) (
Pk´ − Pk´ H T xˆ ´k , t k K xˆ ´k , t k H xˆ ´k , t k Pk´ ) ( ) (8)
Fig.2. The coordinate Cooperative quadrotor system [27] The estimated parameters by kalman filter have indicated by
coordinate frame B . The axes of the coupled system are notation prime. and Kalman filter gain is computed as:
described as body frame axes and each UAV has an individual
( ) ( ) ( )
−1
body frame Qi where=i {1, 2,…, n} and n is the number= of K xˆ ´k , t k  H xˆ ´k , t k Pk´ H T xˆ ´k , t k + R k  (9)
UAVs in the group. For extracting the dynamic model of
 
the coupled system we assumed the quadrotors are rigidly where R k represents the covariance matrix of the sensor
connected to the object and also, the center of mass of the noise. In prediction step, we have:
system is intended as the origin of the rigid-body coordinate
dxˆ ´ ( t )
system. Therefore, the motion formulation for each quadrotor
can be written as: dt
= f xˆ ´ ( t ) , t ( ) (10)
¨
mi z Qi =
−mi g + FBi , dP′ ( t )
( ) ( )
(3)
=F xˆ ´ ( t ) , t P′ ( t ) + P′ ( t ) F T xˆ ´ ( t ) , t + Q ( t ) (11)
where FBi is lift force generated by each quadrotor. By dt
applying Newton’s second law to the payload, we have: ( ) (
where F xˆ ´ ( t ) , t and H xˆ ´k , t k are the Jacobean matrixes )
of f ( x ( t ) , t ) and h ( x k , t k ) evaluated at xˆ ´ ( t ) respectively.
Moreover, Q ( t ) represents the power of measurement noise.
¨
mi Z QB = −mL g , (4)
After applying EKBF on system, the estimated is shown as
where mL is the mass of the object. Since, all the forces are
ϕˆ , ϕˆ , θˆ, θˆ,ψˆ ,ψˆ , xˆ , xˆ , yˆ , yˆ , zˆ, zˆ  ,
T

used for lifting and moving toward the positive Z direction, =

Xˆ [ xˆ 1 ,…….., xˆ 12=
]T  
the equation of motion for the coupled system can be easily where X ˆ ∈  4×4n is a constant vector containing the states of
written as: translational and rotational subsystems.
mz =
−mg + FB , m =
m1 + m2 + m3 + m4 + mL Remark 1. Usually, in a real application, the “true” states
are not available and the states are estimated from the
m = m1 + m2 + m3 + m4 + (5) measurement system. To design the controller based on the
“estimated” states using a Kalman filter, although the choice
which FB is the total lift force imposed by all quadrotors
of the assumed covariance matrices Q and R can have a
and m is sum of the quadrotors and object’s mass. Hence,
significant effect on the estimation performance, it is also
the rotor forces can be rewritten as a total force from each
necessary to decrease the sampling time sufficiently. In this
quadrotor Fq ,i . By considering each UAV’s generated
regard, one can observe that both parameters Q and R have a
forces and moments in its own frame, we need to develop
direct relation with sampling time which means that lowering
a relationship between the behavior of the system and the
the sampling time leads to a decrease in the estimation error
agents. Finally, the equation of motion for the coupled system
and directs the estimated values toward the “true” states.
is obtained as it follows [28]:
Therefore, in this paper, in order to ensure that the “true”
 FB 
M   1 0 0 0   Fq ,i  states converge to the desired values using the estimated
 x   y cosψ − sinψ 0   M  states, the following conditions will be considered based on
M  = ∑ i  i   xq ,i 
B
i i
(6) simulations:
 y   − x sin ψ cosψ 0   M y q ,i 
  N ≥100000; ≤
i i i

B
  number of time steps ,
 0 0 0 1   M z q ,i 
 M z 
B
≥ , dt ≤ 0.0001; Sampling time ( s ),
where M x B , M y B , M z B are the body moments in the Z Qi
direction as shown in Fig. 2. Also, Fq ,i is the total force from t = dt * (1: N ) ; time vector ( s )
each quadrotor and M xq ,i, M y q ,i , M z q ,i are moments around in which, to attain more accurate results in simulations, the
each quadrotor’s body frame axis. sampling time is defined considering a constraint on the
3- Nonlinear Observer Based on EKBF upper limit of input frequency of the desired value (i.e. 0.5Hz
The Extended Kalman Bucy Filter is considered as an optimal in this paper). This computational method is based on trial
recursive estimation algorithm for calculating the states of and error strategy to minimize an ITAE cost function of
a nonlinear stochastic system with uncorrelated Gaussian estimation error.

149
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100

4- Cooperative Control Strategy Therefore, four basic control vectors for n cooperative
In this section, the perfect state cooperative guidance quadrotors obtained according (16) and in the in the next T
controller of UAV presented in [28] will be developed to section the individual control vectors,  FBdes , M xBdes , M yBdes , M zBdes  ,
qn qn qn qn

achieve a decentralized control strategy with superior results will be established. Finally, by combination of two set of
regarding state estimation. Therefore, the problem considered mentioned vectors, the cooperative control vectors will be
in this part is separated in two steps: the first step focuses defined as it follows:
on designing the basic control vectors for n quadrotors T
U coop Qn = U Fqn , U Mzqn , U Mxqn , U Myqn   FBdes des
, M xB des
, M yB des 
, M zB (17)
using the Moore-Penrose theory based on estimated state    qn qn qn qn 

variables which are generated by EKBF, and the second step 4- 2- Deriving Individual Control Vectors of Cooperative
describes how to determine individual control vectors by Control Low
SOSMC. Finally, the cooperative control algorithm can be In this section, determining individual control vectors of
extracted by combining the individual control vectors and the cooperative control strategy based on second order sliding
basic control vectors. mode technique is supposed to asymptotically perform
4- 1- Deriving Basic Control Vectors of Cooperative Control position and attitude tracking of the quadrotors in order to
Law generate a robust solution. Now, to complete the cooperative
By using all force vectors of the coupled system in linear algorithm, using tracking errors of state variable(s), a
system (6), four equations for 4 quadrotors can be established switching sliding surface is considered for the whole
as follows: subsystem (rotational or transitional). The sliding manifolds
T
based on SOSMC are chosen as it follows [32, 33]:
 FB , M xB , M yB , M zB  = Xˆ u (12)
Sϕˆ α1ϕˆ (ϕ d − ϕˆ ) + α 2ϕˆ ϕd − ϕˆ
= ( )
where u ∈  includes 4 control input vectors for each n
4n

quadrotor: = ( )
Sθˆ α1θˆ θ d − θˆ + α 2θˆ θd − θˆ ( )
u  FBq ,1 , M x , M y , M z ….., FBq ,n , M x , M yB , M zB 
T
(13) Sψˆ α1ψˆ (ψ d −ψˆ ) + α 2ψˆ (ψ d −ψˆ )
=
 q ,n  (18)
S α ( x − xˆ ) + α x − xˆ ( )
Bq ,1 Bq ,1 Bq ,1 Bq ,n q ,n

= xˆ 1 xˆ d 2x d
Now, consider the following cost function to minimize the
control inputs [28]: S yˆ α1 yˆ
= ( y − yˆ ) + α ( y − yˆ )
d 2 yˆ d

=
J ∑(Λ Fi F + Λ Mxi M
2
q ,i
2
xq ,i + Λ Myi M 2
yq ,i + Λ Mzi M 2
zq ,i ) (14)
S zˆ α1zˆ
= ( z − zˆ ) + α ( z − zˆ )
d 2 zˆ d

i
in which, α1ϕˆ ,α 2ϕˆ ,…, α1zˆ , α 2 zˆ > 0 . The time derivatives of the
in which Λ Fi , Λ Mxi , Λ Myi , Λ Mzi are the weight of each of the sliding manifolds of rotational subsystem are obtained as it
control inputs. By considering the features of mathematical follows;
theory of Moore-Penrose inverse, point-wise minimization of
the function J will be accrued and an optimal control vector ( )
Sϕˆ = α1ϕˆ ϕd − ϕˆ + α 2ϕˆ ϕd − ϕˆ ( )
( ) ( )
can be written as [28 ]:  
Sθˆ α1θˆ
= θd − θˆ + α 2θˆ θd − θˆ (19)
des T
Γ Xˆ T ( Xˆ Γ −2 Xˆ T ) −1  FBdes , M xB
u = * −2 des des
, M yB , M zB  (15)
(
Sψˆ α1ψˆ ψ d −ψˆ + α 2ψˆ ψd −ψˆ
= ) ( )
where:
By generating S xˆ = −κ  sgn ( S xˆ ) − Υ xˆ S xˆ , w =ϕ θ ψ
with
Γ = Γu22 ,  = −κ
x  ( ) − Υx  [1, 2,3] , the associated control inputs are
ith xˆ = ϕˆ ,θˆ,ψˆ , and ∈
Γ = dig  Λ F 1 , Λ Mx1 , Λ My1 , Λ Mz1 ,..., Λ Fn , Λ Mxn , Λ Myn , Λ Mzn  , defined as:
 α1ϕˆ ˆ   Iy − Iz  jr ˆ 
by assuming the weight of basic control vectors =U2
Ix
l

 α 2ϕˆ
( )
ϕd − ϕˆ + ϕd − θψ ˆ
 Ix  Ix
+ θ Ω+
1
κ sgn ( Sϕˆ ) + Υ ϕˆ Sϕˆ  − ωϕ 
α 2ϕˆ  1 

Λ Fi =
Λ F , Λ Mxi =
Λ Myi =
Λ Mzi , we have,
= ( )
I y  α1θˆ  ˆ     Iz − Ix  jr 
θ d − θ + θ d − ϕψ  + ϕˆ Ω +
1 
( ) 
κ sgn Sθˆ + Υθˆ Sθˆ  − ωθ  (20)
( )
−1 U3  ˆ ˆ
Γ Xˆ T Xˆ Γ −2 Xˆ T
−2 U F , U Mz , U Mx , U My  .
= l  α 2θˆ  Iy  Ix α 2θˆ  2 
 qn qn qn qn 

I z  α1ψˆ 
Also, according to this fact that all quadrotors have a common= U4
l

 α
(ψ d −ψˆ ) +ψd − ϕθˆ ˆ  Ix Iz− Iy  + α12ψˆ [κ3 sgn ( Sψˆ ) + Υψˆ Sψˆ ] − ωψ 
role in lifting the object off the ground and equal partnership,  2ψˆ 
the body force F and yaw torque produced by each of them where κ1 ,κ 2 , κ 3 , Υ ϕˆ , Υ θˆ , Υψˆ > 0 . Now for transitional
is almost equal. Therefore, the basic control vectors which subsystem, the objective is to guarantee the state variables
include total force, yaw, pitch and roll moment generated for [ xˆ, yˆ , zˆ ] converge to the desired values [ xd , yd , zd ] ,
the coupled system structure are obtained as follows: respectively. The time derivatives of the sliding manifolds of
1  rotational subsystem are obtained,
 n [1, 0, 0, 0, …..,1, 0, 0, 0] 
   ÿ
 ¨ ¨

 1 0, 0, 0,1, ……, 0, 0, 0,1  S xˆ α1xˆ  x d − xˆ  + α 2 xˆ  x d − xˆ 
=
 U Fqn   n [ ]
  


   
 U Mzqn    Λ Mxy Λ Mxy  
   
1
ψ ψ … ψ ψ (16) ÿ ¨ ¨
= Λ  
S yˆ α1 yˆ  y d − yˆ  + α 2 yˆ  y d − yˆ  ,
 y , C , S , 0, , y , C , S , 0
 U Mxqn   Mxy ∑ yi2 + n  F
Λ
q1 q1 q1
ΛF
qn qn qn


 = (21)
 U   Λ F     
 Myqn  
 Λ Mxy Λ Mxy  
 1
 ÿ
  ¨ ¨

S zˆ α1zˆ  zd − zˆ  + α 2 zˆ  z d − zˆ  .
− xq1 , − Sψ q1 , Cψ q1 , 0, …, − xqn , − Sψ qn , Cψ qn , 0 
 Λ Mxy  ΛF ΛF  =
 Λ ∑ xi + n
2

 F

    

150
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100

By generating S xˆ = −κ  sgn ( S xˆ ) − Υ xˆ S xˆ , ( xˆ = xˆ, yˆ , zˆ ) and by as follows:

considering ∈ [ 4,5, 6] , the associated control inputs are m  α1zˆ   ¨
( ) 
ÿ
=Fqdes   zd − Xˆ 12 qn  + z d + g + κ 6 tanh S Xˆ11qn + Υ Xˆ11qn S Xˆ11qn − ω z 
defined as: ,n
µ3  α 2 zˆ   
m  α1xˆ   α1ϕˆ  
( )  ¨  Iy − Iz 
ÿ ÿ ÿ
¨ 1
=Ux  xd − xˆ + x d + κ 4 sgn ( S xˆ ) + Υ xˆ S xˆ  − ω x 
   ϕd − Xˆ 2 qn  + ϕ d − Xˆ 3qn Xˆ 5 qn   
µ1  α 2 xˆ α 2 xˆ , I α  Ix  
M xdes = x  2ϕˆ  ÿ 

m  α1 yˆ
( )  l  jr ˆ 1 
( ) 
q ,n
1 
κ sgn ( S yˆ ) + Υ yˆ S yˆ  − ω y  (22) + X 3qn Ω + κ tanh S Xˆ + Υ Xˆ S Xˆ − ωϕ 
¨
=Uy  y − yˆ + y d +  Ix α 2ϕˆ  
1 qn 

α 2 yˆ  5
1
µ2  α 2 yˆ d
1 qn 1 qn

 ,
m  α1zˆ   α1θˆ  ÿ
 ¨ ÿ ÿ
 Iz − Ix  
( ) 1 θd − Xˆ 4 qn  + θ d − Xˆ 1qn Xˆ 5 qn 
¨
=U1  zd − zˆ + z d + g + [κ sgn ( S zˆ ) + Υ zˆ S zˆ ] −ω z  
I y α 2θˆ   

(28)
µ3  α 2 zˆ α 2 zˆ 6 =    Iy 
, M ydes 
( )
ÿ
l  jr ˆ 1   −ω 
where κ 4 ,κ 5 , κ 6 , Υ xˆ , Υ yˆ , Υ zˆ > 0 . In addition, the terms µ1 , µ2 , µ3
q ,n

+ X 1qn Ω +

κ t anh S + Υ S
Xˆ 3 qn Xˆ 3 qn  θ
are:  Ix α 2θˆ  2 Xˆ 3 qn
 
 α1ψˆ  ÿ
 ¨ ÿ ÿ
 Ix − Iy  
µ2 = ( cosϕˆ sinθˆsinψˆ − sinϕ sinψ ) , µ = ( ϕ θ 1 ψ+ ϕ ψ )µ = ϕ
I
θˆ 
α
z  2ψˆ 
ψ d − Xˆ 6 qn  +ψ d − Xˆ 1qn Xˆ 3qn 
  Iz  

) ( cosϕˆ sinθˆcosψˆ + sinϕˆ sinψˆ ) , µ = cosϕˆ cosθˆ
(23) M zdes =  
ϕ ψ , µ1 = l  1
( )
q ,n
3
+ [κ 3 tanh S Xˆ + Υ Xˆ S Xˆ ] − ωψ 
 α 2ψˆ 
To synthesize a stabilizing control law by second order  
5 qn 5 qn 5 qn

sliding mode, the sufficient condition for stability of the In these equations Xˆ qn is the estimated state vector. Therefore:
system, i.e. V = ( S xˆ ) S xˆ . S xˆ < −. S xˆ , where ℒ is the positive
value ( > 0) , must be verified. The time derivative of V ( S x̂ ) =Ux
m  α1xˆ
 ( 
)
x − Xˆ 8 qn + xd +
µ1 α 2 xˆ d
1 
κ tanh S Xˆ
α 2 xˆ  4 ( )+Υ Xˆ 7 qn 
7 qn 

S Xˆ  − ω x 
satisfies S xˆ . S x̂ < 0 . Thus, a Lyapunov function can be written 
7 qn

1 (29)
as V = ∑ S xˆ T S xˆ , that one can describe as :
2 Uy =
m  α1 yˆ
 ( 
y − Xˆ 10 qn + 
µ2 α 2 yˆ d )
yd +
1 
α 2 yˆ  5 ( ) 

κ tanh S Xˆ 9 qn + Υ Xˆ 9 qn S Xˆ 9 qn  − ω y 

V ( S xˆ =
)
1 T
2
(
S xˆ S xˆ + S Tyˆ S yˆ + S zTˆ S zˆ + SϕTˆ Sϕˆ + SθTˆ Sθˆ + SψTˆ Sψˆ ) (24) and:
.
The chosen law for the attractive surface is the time derivative
of
= µ1 ( cosXˆ 1qn sinXˆ 3qn sinXˆ 5 qn + sinXˆ 1qn sinXˆ 5 qn )
 ( S ˆ ) = S ˆ . S ˆ = −κ S ˆ − Υ S 2ˆ ≤ 0
V x x x  x x̂ x
=
(25) µ2 ( cosXˆ 1qn sinXˆ 3qn sinXˆ 5 qn − sinXˆ 1qn sinX
ˆ
5 qn ) (30)
Therefore, according to (25) the time derivative of final
Lyapunov function V ( S xˆ ) is as: µ3 = cosXˆ 1qn cosXˆ 3qn
 −κ 4 S xˆ − Υ xˆ S xˆ2 − κ 5 S yˆ − Υ yˆ S y2ˆ − κ 6 S zˆ − Υ zˆ S zˆ2  4- 3- Final Cooperative Control Vectors
V ( S xˆ ) = 
Now, according to Eq. (17) the cooperative control vectors
 −κ1 Sϕˆ − Υ ϕˆ Sϕ2ˆ − κ 2 S ˆ − Υ ˆ S 2ˆ − κ 3 Sψˆ − Υψˆ Sψ2ˆ 
 θ θ θ 
2
for each independently designed quadrotor can be described
  ¨  

( l
)
ÿ
−κ 4 S xˆ − Υ xˆ  α1xˆ  xd − xˆ  + α 2 xˆ  x d −  cos ϕˆ sin θˆ cosψˆ + sin ϕˆ sinψˆ U1 + ω x    as follows:
     m 
 U Fq1Fqdes
,1 + U Mxq 1 M xq ,1 + U Myq 1 M yq ,1 + U Mzq 1 M zq ,1
des des des

 U coop Q1   
2
 ¨  
(
−κ 5 S yˆ − Υ yˆ  α1 yˆ ( yd − yˆ ) + α 2 yˆ  y d −  cos ϕˆ sin θˆ sinψˆ − sin ϕˆ sinψˆ
l
)
U1 + ω y   
U   U Fq 2 Fqdes 
,2 + U Mxq 2 M xq ,2 + U Myq 2 M yq ,2 + U Mzq 2 M zq ,2
des des des
   m 
 
2  coop Q2
=  (31)
¨  

( l
) U
 coop Q3   U Fq 3 Fq ,3 + U Mxq 3 M xq ,3 + U Myq 3 M yq ,3 + U Mzq 3 M zq ,3 
ÿ des des des des
−κ 6 S zˆ − Υ zˆ  α1zˆ  zd − zˆ  + α 2 zˆ  z d −  − g + cos ϕˆ cos θˆ U1 + ω z    (27)
     m   U coop Q   
 4 
 U Fq 4 Fqdes 
 ¨  
2
+ U Mxq 4 M xdes + U Myq 4 M ydes + U Mzq 4 M zdes
 ÿ
 ˆ ˆ  Iy − Iz  − jr θˆ Ω + l U + ω   
−κ1 Sϕˆ − Υ ϕˆ  α1ϕˆ  ϕd − ϕˆ  + α 2ϕˆ  ϕ d −  θψ
ÿ ÿ ÿ
 ,4 q ,4 q ,4 q ,4 
    ϕ 
  
2
      Ix  Ix Ix
5- Simulation Results
2
  ÿ
  ÿ̈ ÿ   Iz − Ix ÿ jr l  In order to illustrate the efficiency of the designed controller,
−κ 2 Sθˆ − Υθˆ  α1θˆ  θd − θˆ  + α 2θˆ  θ d −  ϕψ
ˆ ˆ  − ϕˆ Ω + U 3 + ωθ   
    a set of numerical simulations are carried out. The parameters
     Iy Ix y 
  ¨  ÿ  Ix − Iy  l 
2 for simulation of a sample cooperative quadrotors model
 ÿ

−κ 3 Sψˆ − Υψˆ  α1ψˆ ψ d −ψˆ  + α 2ψˆ ψ d −  ϕˆ θˆ   + U 4 + ωψ    are set as shown in Table 1 [30] and the factor of weight of
     Iz  Iz 
 
control inputs is set as Λ = 1 . Mxy

Thus, under the control laws extracted from SOSMC approach ΛF 400

(U1 ,….,U 4 ,U x ,U y ) , all the system state trajectories can reach Table.1 Structural parameters [30]
at and stay on the associated sliding surfaces, respectively, Parameter Definition Value Unit
and guarantees asymptotic stability of closed loop system.
Remark 2. Note that signum function which appears in m Mass of quadrotor 1.80 kg
second order sliding mode controllers and filters can cause l Arm length 0.42
−5
m
3.35 ×10
2
chattering phenomena, or high-frequency oscillations of jr Rotor inertia
−3
kg m
2.16 ×10
2
control variables. This problem can be avoided by replacing Ix X inertia
−3
kg m
2.16 ×10
2
discontinuous signum function with an appropriate continuous Iy Y inertia kg m
−2
4.52 ×10
2
approximation such as sign ( si ) ≈ tanh ( si / ε i ) . Iz Z inertia
−6
kg m
2.98 ×10
2
Finally, individual control vectors in (20) and (22) that are b Trust factor
−2
N .s
2.25 ×10
2
generated using SOSMC method, represented by parameters d Drag factor N m.s
which are equivalent of ( Fqdes des des des
. Thus the g gravity 9.81 m / s2
,n , M x , M y , M z ) q ,n q ,n q ,n
individual control vectors for cooperative strategy are defined

151
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100

Table.2 Controller parameters

Item Value Item Value Item Value Item Value Item value

α1xˆ 3.58 α1ϕˆ 8.05 Υ x̂ 6.32 κ1 3.04 ε1 0.0008

α 2 xˆ 5.05 α 2ϕˆ 13.65 Υ ŷ 4.12 κ2 1.24 ε2 0.0008
α1yˆ 3.40 α1θˆ 1.89 Υ ẑ 3.65 κ3 2.76 ε3 0.0008
α 2 yˆ 2.60 α 2θˆ 0.43 Υ ϕ̂ 4.34 κ4 3.86 ε4 0.0007
α1zˆ 3.60 α1ψˆ 4.26 Υθˆ 7.14 κ5 1.33 ε5 0.0007
α 2 zˆ 1.68 α 2ψˆ 5.01 Υψˆ 5.96 κ6 4.56 ε6 0.0007

Table.3 quadrotors’ initial position conditions

quadrotor1 quadrotor2 quadrotor3 quadrotor4
=x0 0= =
x0 2.5 =
x0 2.5 x0 0
=y0 0 =y0 0 = =
y3 2.5 y4 2.5

=z0 0 =z0 0 =z3 0=z4 0

Fig.3. Artificial wind gust generating setup Fig.3. The wind velocity component ω ( t ) is expressed with
respect to an inertial coordinate frame x, y, z . This disturbance
According to Item(4) Also in Eqs. (7)-(11) of the EKBF
has a time varying value which is introduced as follows:
estimator, Q ( t ) is considered as a diagonal matrix of
form Q = 10−3 *  7×7 ,10−4 *  5×5  , H ( xˆ ´k , tk ) = [ 12×12 ] and   π ( t − 20 )   π ( t − 20 )   π ( t − 20 )  
  0.5sin   + 0.2sin   + 0.06sin  
R k = 10−4 *  7×7 ,10−4 *  5×5  . Furthermore, state noise is set ω x ( t )    20   10   5 
  π ( − )   π ( − )   π ( − )  m
as w ( t ) = wgn 10 ,1,10 , linear  and measurement noise is
6 −4 ω ( t )  =
ù (t ) =
 y  
0.5sin 
t 30
 + 0.2sin 
t 30
 + 0.06sin 
t 30
  s (34)
assumed as v k = wgn 106 ,1,1, linear  and the Jacobean matrix ω
 z ( t )   20   10   5  
  π ( t − 40 )   π ( t − 40 )   π ( t − 40 )  
is set as: 0.5sin   + 0.2sin   + 0.06sin  
  20   10   5 

0 1 0 0 0 0 0 0 0 0 0 0
0 0 0 a1xˆ 6 0 a1xˆ 4 0 0 0 0 0 0 Remark 3. Genetic algorithm (GA) is a heuristic search
0 0 0 1 0 0 0 0 0 0 0 0 method used in artificial intelligence and computing. It is used
 0 a3 xˆ 6 0 0 0 a3 xˆ 2 0 0 0 0 0 0 for finding optimized solutions to search problems based on
0 0 0 0 0 1 0 0 0 0 0 0
 0 the theory of natural selection and evolutionary biology. After
F ( xˆ ´ ( t ) , t ) =  0 0 0 a2 xˆ 6 0 a1xˆ 4 0 0 0 0 0
0 0 0 0 0 0 0 1 0 0 0 0 (32) an initial population is randomly generated, the algorithm
0 0 0 0 0 0 0 0 0 0 0 0
0 evolves the following three basics: reproduction, crossover,
0 0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 0 0 0 0 and mutation. The most important feature of this method is a
0 0 0 0 0 0 0 0 0 0 0 1  way to transform the system output to the cost function. This

0 0 0 0 0 0 0 0 0 0 0 0 cost function is as it follows:
where ∞
1
 Xˆ T ( t ) ç Xˆ ( t ) + U T ( t ) ãU ( t )  dt
2 ∫0 
=J  (35)
I y − Iz Ix − I y Iz − Ix
=a1 = , a2 = , a3 (33)
Ix Iz Iy The GA operator values are determined after several
.
To investigate the effectiveness of proposed cooperative experiments. In this regard, the population size is set as 160,
controller, two cases are considered. The first case of grasping crossover is considered scatter and its value is 0.6, mutation
is a payload transporting by cooperative quadrotors from is set as 0.4 and type of probability is chosen as Gaussien.
different initial conditions of each UAVs for point-to-point Also η and γ are selected
= as η 10=3
[ 12×12 ] ,γ 102 [ 12×12 ] . The
path tracking. The second case considers grasping and then parameters of proposed controller, obtained by GA, have
manipulation by tracking a predefined trajectory. been presented in Table 2.
Hence, the wind field effect as a destructive factor is shown in

Fig. 4 Tracking simulation results of desired trajectories of position

152
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100
Fig. 5. Tracking simulation results of desired trajectories of angle (φ, θ, ψ)
Fig. 6. Error tracking of position (x, y, z)
Fig.7. Top view trajectory tracking for the Cross conf.
Fig.8. Global trajectory of the quadrotor position
5- 1- Point to point path tracking
Initial conditions of Euler angles of each quadrotor are set as
[ϕ ,θ ,ψ ] = [0.3, 0.5, 0.6] . For the first
T
simulation desired point
configuration is [ x, y, z ] = [ 2, 2, 2] . The external disturbance
T
that is wind field effect is considered according Eq.(34). The
payload on the end-effector is considered 2.5 kg while the
controller is designed for 1.80 kg which is the mass of end-
effector in order to examine the efficiency of the controller
in presence of payload uncertainty. The position and attitude
angle responses of all quadrotors in the presence of wind field
appear in Fig.4 and 5. Once more, the reference trajectories
are tracked within the first 5s. Also, Fig.5 demonstrates the
convergence of all the quadrotors attitude angles to 0 in a
153
short time. The position tracking errors of each quadrotor
are also shown in Fig.6 which illustrates that the position
errors converge to zero in a reasonable time. Moreover, the
fast convergence obtained for the tracking error ensures the
solution of the quadrotor path tracking problem. The top view
and global trajectory of quadrotors in 3D space are shown
in Fig.7 and 8. Therefore, the numerical simulation results
for point-to-point path tracking in a three dimensional space
showed the superior performance of the controller proposed
in this study.
5- 2- Predefined trajectory tracking
A set of numerical simulations are executed to evaluate the
efficiency of the controller proposed for a case that end-
effector must track a predefined trajectory. The important aim
of the simulation in this subsection is to track the following
desired trajectory by the group of quadrotors:
 t +π 
xdqn =
0.05sin   , ydqn =
−0.05 , zdqn =
−0.032t + 0.8 (36)
 2 
The position and attitude angle responses of the system in
presence of wind field effect in predefined trajectory tracking
are shown in Figs.9 and 10. The proposed controller stabilizes
the attitude angle at 0 rad in a short time. It can be seen that
the control inputs are smooth and free of any chattering.
The tracking trajectory of all quadrotors appear in Fig.10. It
is shown that even though the position and attitude of the
quadrotors are affected by the abruptly changed reference
positions and angles, the controller is able to drive all these
state variables back to the new reference position and angle
within seconds which is a mandatory characteristic to control
a system such as the quadcopter. The top view trajectory
tracking and trajectory of all quadrotors in 3D space are
shown in Fig.11 and 12. We see that the system performs well
and tracks the desired trajectory in three-dimensions. Results
indicate that quadrotors lift up and down the object, and
stable grasp has been achieved. Moreover, the aerodynamic
forces and moments and air drag are taken into account in
the controller design which demonstrate the robustness and
effectiveness of the designed cooperative control scheme.
The behavior of the sliding variables, shown in Fig.13, follows
the expectations as all these variables converge to their sliding
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100
Fig.9 Tracking simulation results of desired trajectories of angle (ϕ, θ, ψ)
Fig.10 Tracking simulation results of desired trajectories of position (x, y, z)
Fig.11 Top view trajectory tracking for the Cross conf.
Fig.12 Global trajectory of the quadrotor position
surfaces. Moreover, the position and velocity tracking errors
of the system state variables are perfectly explained by the
fluctuations of these sliding variables. Finally, Fig.14 shows
the stability rotor speed response of a Quadrotor during
hovering. The control inputs are smooth and the chattering
phenomenon is decreased using the proposed controller. The
amplitudes of the control inputs are also practically realizable
and small. In a practical sense, the energy reduction enforces
154
a longer autonomous flying period for the quadrotor which is
still a major problem for this kind of mobile robots.
6- Conclusion and Future Works
In this paper, the problem of controlling multiple quadrotors
that cooperatively manipulate and transport a payload
in three dimensions was properly addressed. An optimal
decentralized control algorithm based on second order sliding
mode controller (SOSMC) approach using extended kalman-
bucy filter (EKBF) as an observer has been proposed for a
group of cooperative quadrotors. First, a model for a single
quadrotor and then a team of quadrotors rigidly attached to
a payload were developed. Individual robot control laws
were introduced with respect to the payload to stabilize the
payload along three-dimensional trajectories. This method
offers an efficient and systematic procedure to solve a non-
linear closed-loop optimal robust control problem. In order to
validate the efficacy and optimality of the proposed method,
simulation results were presented. Simulation results suggest
that our method demonstrates better performance in the
presence of external disturbances and parametric uncertainties
such as wind field effects. Qualities of high accuracy, small
position errors, minimum influence of the nonlinearities on
the performance of cooperative UAVs, and versatility to a
variety of maneuvers are other advantages of our proposed
method. As a future work, the authors seek to develop a multi-
stage sliding controller to maintain the states at desired values
while using the Load Sharing approach among multiple
agents in order to improve design performance gained in this
paper.
Reference
[1] M.H. Korayem, A. Zehfroosh, H. Tourajizadeh,
S. Manteghi, Optimal Motion Planning of Non-
linear Dynamic Systems in the Presence of Obstacles
and Moving Boundaries Using SDRE: Application
on CableSuspended Robot, Nonlinear Dyn, 76(2)
(2014)1423-1441.
[2]  K. Das, R. Fierro, V. Kumar, J.P. Ostrowski, J.Spletzer,
C.Taylor, A vision-based formation control framework,
IEEE Transactions on Robotics and Automation, 18(5)
(2002) 813-825.
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100

Fig. 13 The sliding variables (Sφ, Sθ, Sψ, Sx, Sy , Sz)

Fig. 14 The controllers (U1, U2, U3, U4)

[3] J. Sanchez, R. Fierro, Sliding mode control for robot
formations, In: Proceedings of the 2003 IEEE International
Symposium on Intelligent Control, Houston, USA, 2003.
[4] O.A.A. Orqueda, R. Fierro, Robust vision-based
nonlinear formation control, In: Proceedings of the 2006
American Control Conference, Minneapolis, USA, 2006.
[5] T.J. Tarn, A.K. Bejczy, X. Yun, Coordinated Control
of Two Robot Arms, In: Proceedingog 1986 IEEE
International Conference Robotics and Automation , San
Francisco, USA, 1986.
[6] S.C. LIU, D.L. Tan, G.J. Lui, Robust Leader-follower
Formation Control of Mobile Robots Based on a Second
Order Kinematics Model, Acta Automatica Sinica, 33(9)
(2007) 947–955.
[7] R. Babaei, A.F. Ehyaei, A Novel UAV Cooperative
Approach for Industrial Inspec on and Monitoring,
In: Proceeding of 31st International Power System
Conference (PSC), Tehran, Iran,2016.
[8] P. Cheng, J. Fink, S. Kim, V. Kumar, Cooperative
towing with multiple robots, Algorithmic Foundation of

155
 R. Babaie, A. F. Ehyaei, AUT J. Model. Simul., 50(2) (2018) 147-156, DOI: 10.22060/miscj.2018.14252.5100
Robotics VIII. Springer Tracts in Advanced Robotics, 57
( 2008) 101-116.
[9] N. Michael, J. Fink, V. Kumar, Cooperative manipulation
and transportation with aerial robots, Autonomous
Robots , 30(1) (2011) 73-86.
[10] J. Fink, N. Michael, S. Kim, V. Kumar, Planning and
control for cooperative manipulation and transportation
with aerial robots, Int. J. Rob. Res, 30(1) (2010) 324–334.
[11]  M. Bernard, K. Kondak, I. Maza, A. Ollero, Autonomous
transportation and deployment with aerial robots for
search and rescue missions, Journal of Field Robotics,
28(6) (2011) 914–931.
[12] G. Wang, X. Nian, H. Wang, Cooperative control of
discrete-time linear multi-agent systems with fixed
information structure, J Control Theory , 10(3) (2012)
403–409.
[13] A.R. Patel, M.A. Patel, D.R. Vyas, Modeling and
analysis of quadrotor using sliding mode control, In:
Proceeding of the 44th IEEE Southeastern Symp. on
System Theory, Jacksonville , 2012.
[14] L. Besnard, Y.B. Shtessel, B. Landrum, Quadrotor
vehicle control via sliding mode controller driven
by sliding mode disturbance observe , Journal of the
Franklin Institute, 349 (2)(2012) 658-684.
[15] B. Sumantri, N. Uchiyama, S. Sano, Y. Kawabata,
Robust tracking control of a quad-rotor helicopter
utilizing sliding mode control with a nonlinear sliding
surface, J. System Design and Dynamics, 7(2)(2013)
226-241.
[16] B. Sumantri, N. Uchiyama, S. Sano, Least square
based sliding mode control for a quad-rotor helicopter,
In: Proceeding of IEEE/SICE Int. Symp. on System
Integration, Kobe ,2013.
[17] J.J. Xiong , E.H. Zheng, Position and attitude tracking
control for a quadrotor UAV, ISA Transaction, 53(3)
(2014) 725-731.
[18] R. Babaei, A. FEhyaei, Robust Control Design of
a quadrotor UA Based on Incremental Hierarchical
Sliding Mode Approach, In: 25th Iranian Conference on
Electrical Engineering (lCEE), Tehran, Iran, 2017.
[19] L. Derafa, A. Benallegue, L. Fridman, Super twisting
control algorithm for the attitude tracking of a four rotors
UAV, Journal of the Franklin Institute , 349 (2) (2012)
685-699.
[20] L.L.Vegan, B.C.Toledo, and A.G.Loukianov, Robust
block second order sliding mode control for a quadrotor,
Journal of the Franklin Institute , 349 (2) (2012) 719-739
[21] M. Bouchoucha, S. Seghour, M. Tadjine, Classical
and second order sliding mode control solution to an
attitude stabilization of a four rotors helicopter: from
theory to experiment , In: Proceedings of the 2011 IEEE
Please cite this article using:
R. Babaie, A. F. Ehyaei, Cooperative Control of Multiple quadrotors for Transporting a Common Payload, AUT J. Mod.
Simul., 50(2) (2018) 147-156.
DOI: 10.22060/miscj.2018.14252.5100
156
international conference on mechatronics , Istanbul,
Turkey , 2011
[22] I. Eker, Second-order sliding mode control with
experimental application, ISA Transaction, 49(3) (2010)
394–405.
[23] S. Mondal, C. Mahanta, A fast converging robust
controller using adaptive second order sliding mode, ISA
Transaction, 51 (6) (2012) 713–721.
[24] Z.Q. Guo, J.X. Xu, T.H. Lee, Design and implementation
of a new sliding mode controller on an underactuated
wheeled inverted pendulum, Journal of the Franklin
Institute , 351(4) (2014) 2261–2282.
[25] L. Besnard, Y.B. Shtessel, B. Landrum , Quadrotor
vehicle control via sliding mode controller driven by
sliding mode disturbance observer, Journal of the
Franklin Institute, 349 (2) (2012) 658–84.
[26] C. Coza, C. Nicol, C.J.B. Macnab ,A. Ramirez-Serrano
, Adaptive fuzzy control for a quadrotor helicopter robust
to wind buffeting, Journal of Intelligent and Fuzzy
Systems , 22(5)(2011) 267–83.
[27] R. Babaie , A.F. Ehyaei, Robust optimal motion planning
approach to cooperative grasping and transporting using
multiple UAVs based on SDRE , Transaction of the
Institute of Measurement and Control , 39(9)(2017)1391-
1408.
[28] D. Mellinger, M. Shomin, N. Michael, V. Kumar ,
Cooperative Grasping and Transport Using Multiple
quadrotors, Distributed Autonomous Robotic Systems.
Springer Tracts in Advanced Robotics, 83( 2013) 545-
558.
[29] M. Rhudy, Y. Gu, M.R. Napolitano, An Analytical
Approach for Comparing Linearization Methods in EKF
and UKF, International Journal of Advanced Robotic
Systems, 10 (208) (2013).
[30] M. Rhudy, Y. Gu , Understanding Nonlinear Kalman
Filters Part I: Selection of EKF or UKF, Interactive
Robotics Letters, Link: http://www2.statler.wvu.edu/~irl/
page13.html, 2013.
[31] R. Babaie, A.F. Ehyaei, Robust Backstepping Control of
a quadrotor UAV Using Extended Kalman Bucy Filter,
International Journal of Mechatronics, Electrical and
Computer Technology, 5(16) (2015) 2276- 2291.
[32] K. Runcharoo, V. Srichatrapimuk, Sliding Mode Control
of quadrotor, In: Proceeding of International Conference
of Technological Advances in Electrical, Electronics and
Computer Engineering (TAEECE), Konya, 2013.
[33] L.Wang, C. He, P. Zhu, Adaptive Sliding Mode Control
for quadrotor Aerial Robot with Ι Type Configuration,
International Journal of Automation and Control
Engineering , 3(1) (2014) 20-26.