JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.

1 (1-7)
Aerospace Science and Technology ••• (••••) ••••••

1 67
Contents lists available at ScienceDirect
2 68
3 69
4 Aerospace Science and Technology 70
5 71
6 72
7
www.elsevier.com/locate/aescte 73
8 74
9 75
10 76
11
12
Multivariable adaptive control based consensus flight control system 77
78
13 for UAVs formation ✩ 79
14 80
15 a b,∗ a b 81
Ziyang Zhen , Gang Tao , Yue Xu , Ge Song
16 82
a 83
17 Nanjing University of Aeronautics and Astronautics, Nanjing, 210016, China
b
18 University of Virginia, Charlottesville, 22904, USA 84
19 85
20 86
21 a r t i c l e i n f o a b s t r a c t 87
22 88
Article history: Formation flight contributes to improving the attack, reconnaissance and survival ability of the multiple
23 89
Received 31 December 2016 unmanned aerial vehicles (UAVs). This paper studies a multivariable adaptive control based consensus
24 Received in revised form 28 June 2019 90
flight method for UAVs formation. A majority of existing research is focused on the leader-following
25 Accepted 12 August 2019 91
consensus problem assuming that only the parameters of followers are uncertain. However, they do not
26 Available online xxxx 92
consider the leader dynamic uncertainty and the unknown external disturbances. Therefore, this paper
27 addresses the problem of the UAVs consensus flight control with parametric uncertainties and unknown 93
Keywords:
28 external disturbances for both the leader and follower. A multivariable model reference adaptive control 94
Multivariable adaptive control
29 Consensus flight control (MRAC) based consensus flight control scheme is designed for UAVs formation, which enables the 95
30 UAVs formation follower UAV to track the leader UAV. The stability of the multivariable MRAC based consensus flight 96
31 control system is analyzed. Simulation results show that the proposed adaptive consensus flight control 97
32 scheme has stronger robustness and adaptivity than the fixed control scheme. 98
33 © 2019 Elsevier Masson SAS. All rights reserved. 99
34 100
35 101
36 102
37 103
1. Introduction At present, specific formation strategies of formation flight
38 104
mainly include: leader-follower formation, behavior-based forma-
39 105
With the wide use of high-tech science and technology in the tion, virtual structure formation and artificial potential field for-
40 106
field of aviation, development of unmanned aerial vehicles (UAVs) mation. Several formation control algorithms have been proposed,
41 107
has made a breakthrough [1]. After decades of development, the such as: feedback linearization (Li, 2001) [9]; H∞ control (Ren,
42 108
UAVs have been widely used in military and civil applications. 2003) [10]; PID control (Zuo, 2004) [11]; linear quadratic (LQ)
43 109
However, a single UAV has many limitations to complete the com- control (Giulietti, 2000, 2005) [12,13]; adaptive control (Semsar,
44 110
plex tasks. For example, when a single UAV is performing a recon- 2006) [14]; neural network based dynamic surface control (Peng,
45 111
naissance mission, it may fail to observe the target from different 2013) [15]. However, the formation control methods in [9–13]
46 112
directions, due to the sensor’s detection limitation. Therefore, the are based on the system model with known parameters, while
47 113
cooperativity among UAVs is usually indispensible. In order to im- the method in [14] only considers the unknown leader command
48 114
prove the efficiency and broaden the range of applications, the and disturbance. Ref. [15] considered the uncertain local dynamics
49 115
concept of UAVs formation flight was put forward, which enables and uncertain leader dynamics, but it was mainly a neural net-
50 116
UAVs to complete the consensus flight, aerial refueling, cooperative work method for multiple autonomous surface vehicles formation.
51 117
combat missions and so on [2]. In military applications, the UAVs Refs. [16,17] designed some adaptive formation control algorithms,
52 118
formation flight can expand investigation view and improve the however, they were for the second-order agents in an unknown
53 119
rate of task completion. In civil applications, it has wide prospect
54 spatiotemporal flowfield. 120
in the fields of meteorological exploration, forest fire prevention
55 When a group of UAVs are keeping a desired special distance 121
and emergency rescue. In these applications, several key technical
56 and flying in a stable, specified structure, formation flight problem 122
problems should be addressed, such as cooperative path planning
57 can be equal to a consensus problem. Consensus means that the 123
[3,4], cooperative mission planning [5], formation relative naviga-
58 followers eventually reach an agreement on the state or output of 124
tion [6], formation control [7] and collision avoidance [8].
59 a leader. Recently, the leader-following system with general linear 125
60 dynamics has been investigated. Ni et al. [18] studied a consen- 126
61 ✩
Fully documented templates are available in the elsarticle package on CTAN. sus leader-following problem for the identical followers and leader. 127
62 * Corresponding author. This method can also be applied in the formation of multiple 128
63 E-mail address: gt9s@virginia.edu (G. Tao). UAVs. Being different from [18], Liu et al. [19] solved a consensus 129
64 130
https://doi.org/10.1016/j.ast.2019.105336
65 131
1270-9638/© 2019 Elsevier Masson SAS. All rights reserved.
66 132
 JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.2 (1-7)
2 Z. Zhen et al. / Aerospace Science and Technology ••• (••••) ••••••

1 problem for the leader-following systems with different dynamics Ẋlon = A lon Xlon + B lon U lon (3) 67
2 under external disturbances. Peng et al. proposed the distributed Ẋlat = A lat Xlat + B lat U lat (4)
68
3 adaptive control schemes respectively using the virtual structure 69
4 strategy and the leader-following strategy for the uncertain nonlin- where Xlon = [ V , α , q, θ] T denote velocity, angle of attack, 70
5 ear multi-agent systems in [20] and [21]. In many cases, the model pitch angular rate and pitch angle, U lon = [δe , δ T ] T represent 71
6 parameters of both the leader and followers are unknown, thus the control surfaces deflections of elevator and throttle, X lat = 72
7 it is difficult to control the formation flight by a fixed controller. [β,  p , r ] T denote sideslip angle, roll angular rate and yaw an- 73
8 Adaptive control is a control method capable of dealing with un- gular rate, U lat = [δa , δr ] T represent the control surfaces deflec- 74
9 certain systems to ensure desired control performance under the tions of aileron and rudder, respectively. A lon and Alat are unknown 75
10 matching conditions [22,23]. A majority of existing research is fo- matrices which are decomposed from A, while B lon and B lat are 76
11 cused on the leader-following consensus problem assuming that unknown matrices which are decomposed from B. The detailed ex- 77
12 the parameters of follower agents are uncertain, while few litera- pressions of these matrices can be found in [24,25].  denotes the 78
13 ture consider the leader and the follower dynamic uncertainty at deviation from the equilibrium point. 79
14 the same time. The longitudinal and lateral controllers are usually designed 80
15 Therefore, we focus on the consensus flight control problem separately. Therefore, for design convenience of the controllers, an 81
16 of the leader-following UAVs system with parametric uncertain- unified state-space model for UAV’s longitudinal motion or lateral 82
17 ties and unknown external disturbances. Different from the results motion is established, given by 83
18 in the literature, the main contributions of this paper are as fol- 84
19 lows: This problem has not been addressed in the literature. In ẋ = Ax + Bu (5) 85
20 the real engineering environment, it is difficult to get accurate Here, for the longitudinal controller design, x denotes X lon , A de- 86
21 UAV dynamic parameters. Being different from [11,14], both the notes A lon , B denotes B lon and u denotes U lon ; For the lateral 87
22 parametric uncertainties of the lead UAV and wing UAV are consid- controller design, x denotes X lat , A denotes A lat , B denotes B lat 88
23 ered here. Different with [24], the unknown external disturbances and u denotes U lat . 89
24 in both the lead UAV and wing UAV are considered. Therefore, a 90
25 state feedback state tracking multivariable model reference adap- 2.2. Consensus flight control problem 91
26 tive control (MRAC) scheme is presented, which ensures the con- 92
27 sensus flight control system stable, rejects the effect of the distur- The consensus flight control problem of UAVs in formation 93
28 bances and makes the wing UAV track the lead UAV asymptotically. flight can be described as the leader-following problem. In this 94
29 The remainder of this paper is organized as follows. In Section 2 problem, both the leader and the follower have uncertain parame- 95
30 we consider the parametric uncertainties and unknown exter- ters and unknown external disturbances. 96
31 nal disturbances, to establish dynamic models of consensus flight Consider a linear time-invariant follower plant in state-space 97
32 problem. In Section 3, we develop an adaptive control scheme for form 98
33 leader-following consensus flight problem of uncertain UAVs sys-  99
ẋ(t ) = Ax(t ) + Bu (t ) + Bd(t )
34 tem. In Section 4, we present the simulation results of the UAVs in q (6) 100
d(t ) = d0 + β=1 dβ f β (t )
35 formation flight to illustrate the effectiveness of the adaptive con- 101
36 trol method. Finally, we discuss some conclusions and the future where x(t ) ∈ R n , u (t ) ∈ R p , A and B are unknown parameter ma- 102
37 work in Section 5. 103
trices, the disturbance d(t ) is matched to the control input u (t )
38 104
through the same matrix B, with d0 ∈ R p and dβ ∈ R p being un-
39 2. Consensus flight control problem of UAVs formation 105
known constant vectors, and f β (t ), β = 1, 2, . . . , q are some known
40 106
bounded basis functions, for some q ≥ 0. The leader dynamic sys-
41 In this section, the consensus flight control problem of leader- 107
tem is given by
42 following UAVs formation with the parameter uncertainties and 108

43 unknown external disturbances are formulated. ẋm (t ) = A m xm (t ) + B m um (t ) + B m dm (t ) 109
qm (7)
44 dm (t ) = dm0 + β= 1 dmβ f mβ (t )
110
45 2.1. UAV dynamic model 111
46 where xm (t ) ∈ R n , um (t ) ∈ R m , A m and B m are an unknown pa- 112
47 rameter matrices, and xm (t ) and um (t ) are bounded and available 113
The UAV is generally a nonlinear system with six degrees
48 for measurement, dm (t ) is matched to the control input um (t ) 114
of freedom, which can be linearized according to the small-
49 through the same matrix B m , with dm0 ∈ R m and dmβ ∈ R m being 115
disturbance theory, expressed as [24,25] unknown constant vectors, and f mβ (t ), β = 1, 2, . . . , qm are some
50 116
known bounded basis functions, for some qm ≥ 0. Here, d(t ) and
51
E Ẋ = Ã X + B̃U (1) 117
52 dm (t ) represent the unknown external disturbances. Some distur- 118
53 where E, Ã and B̃ are Jacobian matrices; X = [ V , β, α , θ, p , q, r ] T bance of UAVs is the interference of the airflow, which can be 119
54 is the state vector. For E = diag (m, mV ∗ , 1, 1, mV ∗ , 1, 1), det ( E ) = described as a constant matrix. More general disturbances can not 120
55 m3 V ∗2 , here m and V ∗ denote the mass and balanced velocity be denoted as constant vectors, they are usually in the form of (6) 121
56 and (7) [26]. 122
of the UAV, and both of them are not equal to zero. Therefore,
57 det ( E ) = m3 V ∗2 = 0, E is nonsingular. Then (1) can be rewritten in 123
58 Remark 1. In the leader-following UAVs system, xm (t ) denotes the 124
the form of standard linear state equation
59 states of the lead UAV, um (t ) represents the control input of the 125
lead UAV. They are all bounded and can be obtained by the wing
60 Ẋ = A X + BU (2) 126
61 UAV through the measurement or the communication. 127
62 where A = E −1 Ã and B = E −1 B̃ contain the UAV’s stability deriva- The control objective is to design a bounded state feedback con- 128
63 tives. trol law u (t ) including a disturbance compensator to make the 129
64 Generally, based on the earth-surface inertial reference frame, follower system state x(t ) bounded and track the leader system 130
65 the linear model (2) is decoupled to a longitudinal model and a xm (t ) asymptotically, i.e., lim (x(t ) − xm (t )) = 0 and reject the ef- 131
t →∞
66 lateral model fect of unknown external disturbances. 132
 JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.3 (1-7)
Z. Zhen et al. / Aerospace Science and Technology ••• (••••) •••••• 3

1 ė (t ) = A e e (t ), e (0) = x(0) − xm (0) (11) 67

2 68
3
which implies that lim e (t ) = 0 exponentially. 69
t →∞
4 When the parameters of A, B, A m , B m are unknown, an adap- 70
5 tive controller is designed as 71
6 72
7 u (t ) = k1T (t )(x(t ) − xm (t )) + k2 (t )um (t ) + k3T (t )xm (t ) + k5 (t ) 73
8 (12) 74
9 75
10
where k1 (t ), k2 (t ), k3 (t ) and k5 (t ) are the estimates of k∗1 , k∗2 , k∗3 76
11 Fig. 1. A MRAC scheme for UAVs formation
and k∗5 , respectively. 77
12
The design objective is to choose adaptive laws to update these 78
13 2.3. MRAC framework estimates so that the control objective is still achievable even if 79
14 there are parameter uncertainties and unknown external distur- 80
15 Given the consensus flight control problem of the UAVs in for- bances. Therefore, we choose the adaptive laws as 81
16 mation flight, an advanced control method is indispensable. MRAC 82
17 methods have been applied in the aircraft flight control fields [27]. k̇1T (t ) T T
= − S B e P e (t )e (t ) T
(13) 83
18 Therefore, in this paper, a novel MRAC scheme is proposed to k̇2 (t ) = − S T B e T P e (t )um T (t ) (14) 84
19 solve this problem, whose schematic diagram is shown in Fig. 1. 85
20 The control system is mainly composed of the reference model, k̇3T (t ) = − S T B e T P e (t )xm T (t ) (15) 86
21 the flight control system and the controlled object. The proposed k̇50 (t ) = − S T B e T P e (t ) (16) 87
22 MRAC scheme has an adaptation capacity to deal with parame- 88
23 ter uncertainties and unknown external disturbances for both the k̇5β (t ) = − S T B e T P e (t ) f β T (t ), β = 0, 1, 2, . . . , q (17) 89
24 leader and the follower. T T T
k̇5mβ (t ) = − S B e P e (t ) f mβ (t ), β = 0, 1, 2, . . . , qm (18)
90
25 91
26 3. Adaptive consensus flight control design where P = P T > 0, satisfies A e T P + P A e = − Q < 0, for any cho- 92
27 sen Q ∈ R n×n being constant and Q = Q T > 0. S ∈ R p × p satisfies 93
28 In order to meet the control objectives, we will solve the UAVs Assumption 2. 94
29 consensus flight control problem using the multivariable MRAC 95
30 scheme. Theorem 1. The adaptive controller (12), with the adaptive laws (13)- 96
31 (18), applied to the linear system (6) guarantees that all closed-loop sig- 97
32 3.1. Adaptive control scheme for consensus flight nals are bounded and the tracking error e (t ) = x(t ) − xm (t ) goes to zero 98
33 as t goes to infinity. 99
34 For the leader-following UAVs with unknown system matrices 100
35 and external disturbances, in order to reject the effect of the dis- 101
Proof. According to (6), (7) and (12), we obtain the tracking error
36 turbances so that the desired consensus flight control performance 102
equation
37 can be achieved, we make some assumptions first. 103
∗−1
38
ė (t ) = A e e (t ) + B e k4 [k̃1T (t )e (t ) + k̃2 (t )um (t ) + k̃3T (t )xm (t ) 104
39 Assumption 1. [24,28,29] There exist four parameter matrices k∗1 ∈ 105
40 R n× p , k∗2 ∈ R p ×m , k∗3 ∈ R n× p and k∗4 ∈ R p × p which satisfy: + k̃5 (t )] (19) 106
41 107
where the parameter errors are
42 A e = A + Bk1∗ T , B m = Bk∗2 108
(8)
43 A m = A + Bk3∗ T , B e = Bk∗4 ∗
k̃i (t ) = ki (t ) − ki , i = 1, 2, 3 (20)
109
44 110
45 where A e ∈ R n×n is a stable matrix and B e ∈ R n× p is a matrix. k̃50 (t ) = k50 (t ) − k∗50 (21) 111
46 ∗ 112
Assumption 2. [24,28,29] There is a known matrix S ∈ R p × p such
k̃5β (t ) = k5β (t ) − k5β , β = 1, 2, . . . , q (22)
47 113
48 that k∗4 S is symmetric and positive definite: M s = k∗4 S = (k∗4 S ) T > k̃5mβ (t ) = k5mβ (t ) − k∗5mβ , β = 1, 2, . . . , qm (23) 114
49 0. 115
50
For convenience of derivation, the time constant t is omitted in 116
51 Assumption 1 is the plant-model state matching condition. As- the following proof. In order to prove the adaptive controller (12) 117
52 sumption 2 is needed to design the adaptive law for the controller with adaptive laws (13)-(18) to ensure e = x − xm bounded, we 118
53 parameters. choose a positive definite function as 119
54 A nominal controller with a disturbance compensator is chosen 120
55 as
V (ec ) = e P e + tr [k̃1 M s k̃1 ] + tr [k̃2T M s−1 k̃2 ] + tr [k̃3 M s−1 k̃3T ]
T −1 T
121
q
56 T
+ tr [k̃50 M s−1 k̃50 ] + β=1 tr [k̃5Tβ M s−1 k̃5β ] (24) 122
57 u ∗ (t ) = k1∗ T (x(t ) − xm (t )) + k∗2 um (t ) + k3∗ T xm (t ) + k∗5 (t ) (9) qm 123
+ −1
β=1 tr [k̃5mβ M s k̃5mβ ]
T
58 124
where k∗1 T , k∗2 T and k∗3 T satisfy the Assumption 1, and
59 125
with tr [ M ] denoting the trace of a square matrix M.
k∗5 (t ) = k∗2 dm (t ) − d(t )
60 126
Calculate the time derivative of V (e c ) as
61 qm ∗ q 127
= (k∗2 dm0 − d0 ) + β= 1 k2 dmβ f mβ (t ) − β=1 dβ f β (t ) (10) −1 ˙ T ˙ ˙
T
k̃1 ] + 2tr [k̃2T M s−1 k̃2 ] + 2tr [k̃3 M s−1 k̃3T ]
128
V̇ = 2e P ė + 2tr [k̃1 M s
62
63 ∗ qm ∗ q ∗ 129
= k50 + β=1 k5mβ f mβ (t ) + β=1 k5β f β (t ) ˙ q ˙
64 T
+ 2tr [k̃50 M s−1 k̃50 ] + 2 β=1 tr [k̃5Tβ M s−1 k̃5β ] (25) 130
65 with k∗50 = k∗2 dm0 − d0 , k∗5mβ = k∗2 dmβ , and k∗5β = −dβ . Then the qm 131
+ 2 β= −1 ˙
1 tr [k̃5mβ M s k̃5mβ ]
T
66 tracking error satisfies 132
 JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.4 (1-7)
4 Z. Zhen et al. / Aerospace Science and Technology ••• (••••) ••••••

1 Substituting (19) into (25), we have that each wing UAV needs to interact its motion information with 67
2 the adjacent UAVs and may not get the information from the lead 68
3 V̇ = −e T Q e + 2e T P B e k4∗−1 k̃1T e + 2e T P B e k4∗−1 k̃2 um UAV directly. 69

−1 ˙ T The distributed control strategy can be described by direct

4 70
5 +2e T P B e k4∗−1k̃3T xm + 2e T P B e k4∗−1k̃5 + 2tr [k̃1 M s k̃1 ] graphs, with different kinds of formation communication topology 71
(26)
˙ ˙ ˙
+2tr [k̃2T M s−1k̃2 ] + 2tr [k̃3 M s−1k̃3T ] + 2tr [k̃50 M s−1 k̃50 ] structures. Each follower in the direct graph can get information
6 T 72
7 directly from a nominal leader, which can be an adjacent commu- 73
q ˙ qm −1 ˙
8 +2 β=1 tr [k̃5Tβ M s−1k̃5β ] + 2 β= 1 tr [k̃5mβ M s k̃5mβ ]
T nicated UAV or the lead UAV. Thus, the proposed adaptive control 74
9 scheme can be applied in the followers, to make each follower 75
10 Using the definition M s = k∗4 S = (k∗4 S ) T > 0 and the properties that asymptotically track its nominal leader. The convergence of local 76
11 tr [ M 1 M 2 ] = tr [ M 2 M 1 ], tr [ M 3 ] = tr [ M 3T ], we obtain tracking error for each UAV will contribute to achieving the con- 77
12 vergent global consensus performance. 78
˙
13 e T P B e k4∗−1 k̃1T e = tr [k̃1 M s−1 S T B eT P ee T ] = −tr [k̃1 M s−1 k̃1T ] (27) 79
14 4. Simulation study 80
˙
15 e T P B e k4∗−1 k̃2 um = tr [k̃2T M s−1 S T B eT P eum
T
] = −tr [k̃2T M s−1k̃2 ] (28) 81
16
∗−1 T −1 T ˙ In this section, we use MATLAB R2014a to simulate the leader- 82
17
T
e P B e k4 k̃3 xm = tr [k̃3 M s S B eT
P exmT
] = −tr [k̃3 M s−1k̃3T ] (29)
following UAVs system, to verify the effectiveness of the multivari- 83
qm
e T P B e k4∗−1 k̃5 = e T P B e k4∗−1 [k̃50 + β=
18 able MRAC scheme, by comparing with the fixed control. 84
1 k̃5mβ f mβ
19 q 85
20 + β=1 k̃5β f β ] 86
˙ q ˙ (30) 4.1. Control problem and method
21 = −tr [k̃50
T
M s−1 k̃50 ] − β=1 tr [k̃5Tβ M s−1 k̃5β ] 87
22 qm −1 ˙ 88
23
− β=1 tr [k̃5m T
β M s k̃5mβ ]
Controlled plant. A linear model of a real UAV called Silver
89
Fox is developed, including the longitudinal and lateral dynamics,
24 90
Substituting (27)-(30) into (26), we get based on the aircraft parameters and the aerodynamic data in [30].
25 91
The fuselage of the Silver Fox is 1.8 m long and the aircraft weighs
26
V̇ = −e T (t ) Q e (t ) ≤ −qm e (t ) 22 ≤ 0, Q = QT >0 (31) 8.6 kg. The Silver Fox can carry small cameras and global position-
92
27 93
28 where qm > 0 is the minimum eigenvalue of Q . From the above ing system (GPS) receiver. It was originally developed by Advanced 94
29 results, the desired properties of the proposed adaptive laws are Ceramics Research (ACR). The linear model of the UAV which con- 95
30 obvious: sists of the longitudinal and lateral equations are respectively given 96
31 (i) V > 0 and V̇ ≤ 0 imply that the equilibrium state (e = 0, by 97
32 k̃1 = 0, k̃2 = 0, k̃3 = 0, k̃50 = 0, k̃5β = 0, k̃5mβ = 0) of the closed- ⎡ ⎤ ⎡ ⎤ 98
33 loop system (13)-(19) is uniformly stable and the system state  V̇ −0.0785 6.0293 −1.6485 −9.7783 99
(e (t ) = 0, k̃1 (t ) = 0, k̃2 (t ) = 0, k̃3 (t ) = 0, k̃50 (t ) = 0, k̃5β (t ) = 0, ⎢ α̇ ⎥ ⎢ −0.0489 −3.9919 −0.7386 0.0326 ⎥
⎢ ⎥ ⎢ ⎥
⎣ q̇ ⎦ = ⎣ −0.0003 −96.9781 −260.2504
34 100
35 k̃5mβ (t ) = 0) is uniformly bounded, which gives the boundedness 0 ⎦ 101
36 of x(t ), k1 (t ), k2 (t ), k3 (t ), k50 (t ), k5β (t ) and k5mβ (t ) and in turn of θ̇ 0 0 1 0 102
37 the boundedness of ė (t ); ⎡ ⎤ ⎡ ⎤ 103
 V −2.1657 1.4976  
38 (ii) Eq. (31) implies e (t ) ∈ L 2 . With e (t ) ∈ L 2 L ∞ and ė (t ) ∈ ⎢ α ⎥ ⎢ −0.575 −0.0052 ⎥ δe 104
39 L ∞ , applying Barbalat Lemma [22], we conclude that lim e (t ) = 0. ×⎣⎢ ⎥ + ⎢ ⎥ (32) 105
40
t →∞ q ⎦ ⎣ −95.5596 0 ⎦ δ T
106
41
θ 0 0 107
Remark 2. The wing UAV in the leader-following system with ⎡ ⎤ ⎡ ⎤⎡ ⎤
42
uncertain parameters and unknown external disturbances can β̇ −0.1798 0.069 −0.9976 β 108
43 ⎣ ṗ ⎦ = ⎣ −22.4565 −8.213 2.0046 ⎦ ⎣  p ⎦ 109
44
track the lead UAV asymptotically when using multivariable MRAC 110
method. The effect of the unknown external disturbances can ṙ 15.0747 −0.6578 −0.7095 r
45 ⎡ ⎤ 111
46
be rejected effectively when applying adaptive disturbance rejec- 0.0000 0.0873   112
δa
47
tion strategy. Moreover, for the situation without external distur- + ⎣ 99.5144 2.4034 ⎦ (33) 113
bances, an adaptive control u (t ) = k1T (t )(x(t ) − xm (t )) + k2 (t )um (t ) + δr
48 −7.9397 −10.1124 114
49 k3T (t )xm (t ) with adaptive laws (13)-(15) will also make the wing 115
50 UAV track the lead UAV asymptotically. Control problem and objective. For the UAVs consensus flight 116
51 control system with uncertainties for both the lead UAV and the 117
52 Remark 3. For the proposed MRAC method, the convergence of wing UAV, we use the multivariable MRAC method to make the 118
53 control parameters estimation may not be guaranteed. However, state of the wing UAV track the state of the lead UAV. The objective 119
54 the boundedness of control parameters estimation and the attenu- of the consensus flight control based on the multivariable MRAC is 120
55 ation of their variations can be achieved. to design a bounded state feedback control law u (t ) to make the 121
56 wing UAV state x(t ) bounded and track the lead UAV state xm (t ) 122
57 3.2. Extension to multi-UAV system asymptotically, i.e. lim (x(t ) − xm (t )) = 0. 123
t →∞
58 Matching condition. For the multivariable MRAC based consen- 124
59 This work focus on the consensus flight control problem for one sus flight control method, the design conditions as specified by 125
60 lead UAV and one wing UAV formation. The information of lead- Assumption 1 and Assumption 2 need to be satisfied. The sys- 126
61 er’ states, control inputs and follower’s states are necessary for the tem matrices of lead UAV and the wing UAV are the same, thus 127
62 consensus controller design. When there exist more than two wing A m = A , B m = B. However, they may have different initial states, 128
63 UAVs in the multi-UAV system, a suitable communication topology different control inputs, or different external disturbances. In the 129
64 is necessary. The information interaction among UAVs can adopt aircraft flight control field, Assumption 1 and Assumption 2 are 130
65 the distributed control strategy, due to less information interac- normally satisfied based on a linear quadratic (LQ) method, the 131
66 tions and conflicts than the centralized control strategy. It means details of which can be found in [31–34]. Therefore, the matching 132
 JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.5 (1-7)
Z. Zhen et al. / Aerospace Science and Technology ••• (••••) •••••• 5

1 67
2 68
3 69
4 70
5 71
6 72
7 73
8 74
9 75
10 76
11 77
12 78
13 79
14 80
15 81
16 82
Fig. 2. Follower states  V and α vs. leader states with adaptive control vs. fixed control.
17 83
18 84
19 85
20 86
21 87
22 88
23 89
24 90
25 91
26 92
27 93
28 94
29 95
30 96
31 97
32 98
33 99
34 Fig. 3. Follower states q and θ vs. leader states with adaptive control vs. fixed control. 100
35 101
36 102
37 103
38 104
39 105
40 106
41 107
42 108
43 109
44 110
45 111
46 112
47 113
48 114
49 115
50 116
51 Fig. 4. Follower states β and  p vs. leader states with adaptive control vs. fixed control. 117
52 118
53 conditions are satisfied for the adaptive consensus flight control acting on the leader and the follower are dm (t ) = [1.8, 0.05] + T 119
54 system design. 1 ∗ [1, 0.02] T and d(t ) = [3, 0.1] T + 5 ∗ [2, 0.03] T , respectively. For 120
55 121
Control methods and initial parameters. For comparison, two the lateral control channel, the external disturbances acting on the
56 122
control methods are adopted, one is the proposed adaptive con- leader and the follower are dm (t ) = [0.8, 0.5] T + 1 ∗ [2, 1] T and
57 123
trol, the other is the fixed control which exactly is the nominal d(t ) = [1, 0.7] T + 3 ∗ [0, 2] T , respectively. The adaptive longitudinal
58 124
control described by (9). The fixed control parameters can be ob- and lateral controllers of the wing UAV are respectively designed
59 125
60
tained according to the matching conditions, assuming that the by 126
61 system matrices are known. For the longitudinal control channel, 127
62 the leader’s initial state is X mlon = [5 m/s, 0◦ , 0◦ /s, 0◦ ] T , the fol- T
U lon = k1lon T
( Xlon − Xmlon ) + k2lon U mlon + k3lon X mlon + k5lon (34) 128
63 lower’s initial state is X lon = [0 m/s, 0◦ , 0◦ /s, 2◦ ] T , and the control 129
64 input is U mlon = [5◦ , 0.05] T , respectively. For the lateral control T
U lat = k1lat T
( Xlat − Xmlat ) + k2lat U mlat + k3lat X mlat + k5lat (35) 130
65 channel, they are X mlat = [1◦ , 0◦ /s, 0◦ /s] T , Xlat = [0◦ , 0◦ /s, 0◦ /s] T , 131
66 and U mlat = [−2◦ , −1◦ ] T , respectively. The external disturbances where kilon , kilat (i = 1, 2, 3, 5) are updated by (13)-(18). 132
 JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.6 (1-7)
6 Z. Zhen et al. / Aerospace Science and Technology ••• (••••) ••••••

1 67
2 68
3 69
4 70
5 71
6 72
7 73
8 74
9 75
10 76
11 77
12 78
13 79
14 80
15 81
16 82
17 Fig. 5. Follower state r vs. leader state with adaptive control vs. fixed control. 83
18 84
19 85
20 86
21 87
22 88
23 89
24 90
25 91
26 92
27 93
28 94
29 95
30 96
31 97
32 98
33 99
34 100
35 101
36 102
37 103
38 104
39 105
40 106
41 107
42 108
43 109
44 110
45 111
46 112
47 113
48 114
49 115
50 Fig. 6. Follower control input vs. leader control input with adaptive control. 116
51 117
52 118
4.2. Simulation results and discussion still asymptotically track the states of the lead UAV when using
53 119
the adaptive control scheme. Therefore, it is found that the effect
54 The responses results of the flight states under the adaptive 120
of the unknown external disturbances is rejected by applying the
55 control and fixed control are shown in Fig. 2–4. Fig. 5 shows the 121
adaptive control scheme.
56 control surfaces deflections of elevator, throttle, aileron and rudder 122
Comparison of the adaptive control and fixed control. We get
57 under adaptive control (see also Fig. 6). 123
58
that the proposed multivariable MRAC scheme provides substan- 124
Performance analysis of consensus flight control. The final val-
tially improved performance over the fixed control method under
59 ues of tracking errors are as follows: (i) Fixed control:  V e = 125
the same flight conditions. The reason is that the fixed control is
60 1.3046 m/s, αe = −1.0830◦ , βe = 0.9053◦ ,  p e = 3.1228◦ /s, 126
61 qe = 0.0226◦ /s, θe = −1.1288◦ , re = 0.1203◦ /s; (ii) Adap- obtained based on the known system parameters, which are not 127
62 tive control:  V e = 0.2004 m/s, αe = 0◦ , βe = 0◦ ,  p e = accurate enough under the system uncertainties. While the adap- 128
63 0◦ /s, qe = 0◦ /s, θe = 0◦ , re = 0◦ /s. We get that almost state tive control is based on the UAV model with unknown parameters, 129
64 tracking errors converge to zero. It implies that although there whose control gains can be updated online through the adaptive 130
65 are unknown external disturbances for both the lead UAV and the laws. Furthermore, the final values of state tracking errors un- 131
66 wing UAV, the flight states X lon and Xlat of the wing UAV can der fixed control cannot converge to zero. While the final values 132
 JID:AESCTE AID:105336 /FLA [m5G; v1.260; Prn:22/08/2019; 8:19] P.7 (1-7)
Z. Zhen et al. / Aerospace Science and Technology ••• (••••) •••••• 7

1 of state tracking errors under adaptive control converge to zero [11] B. Zuo, Y. Hu, Uav tight formation flight modeling and autopilot designing, in: 67
2 asymptotically. It means that the proposed multivariable MRAC Proceedings of the 5th World Congress on Intelligent Control and Automation, 68
vol. 32(1), 2004, pp. 180–183.
3 method can reject unknown external disturbances and achieve 69
4 better robustness than the fixed control. Therefore, the proposed [12] F. Giulietti, L. Pollini, M. Innocenti, Autonomous formation flight, IEEE Control 70
Syst. Mag. 25 (12) (2000) 34–45, https://doi.org/10.1109/37.887447.
5 multivariable MRAC scheme will be more suitable for the real ap- 71
[13] F. Giulietti, M. Innocenti, M. Napolitano, L. Pollini, Dynamic and control issues
6 plications. of formation flight, Aerosp. Sci. Technol. 36 (9) (2005) 65–71, https://doi.org/
72
7 10.1016/j.ast.2004.06.011. 73
8 5. Conclusion [14] E. Semsar, K. Khorasani, Adaptive formation control of uavs in the pres- 74
9 ence of unknown vortex forces and leader commands, in: Proceedings of 75
10 When a group of UAVs are flying in a stable and specified struc- the 2006 American Control Conference Minneapolis, June 14–16, 2006, 2006, 76
pp. 3563–3569.
11 ture, formation flight can be equal to a consensus flight problem. 77
[15] Z. Peng, D. Wang, Z. Chen, X. Hu, W. Lan, Adaptive dynamic surface control
12 Therefore, in order to address the consensus flight control problem 78
for formations of autonomous surface vehicles with uncertain dynamics, IEEE
13 of the multiple UAVs with parameter uncertainties and external Trans. Control Syst. Technol. 21 (2) (2013) 513–520, https://doi.org/10.1109/ 79
14 disturbances, a multivariable adaptive control scheme is designed. TCST.2011.2181513. 80
15 Simulation study exhibits that the formation flight stability and [16] Y. Chen, Z. Wang, Y. Zhang, C. Liu, Q. Wang, A geometric extension design for 81
16 tracking properties can be better achieved by the multivariable spherical formation tracking control of second-order agents in unknown spa- 82
tiotemporal flowfields, Nonlinear Dyn. 88 (2017) 1173–1186, https://doi.org/10.
17 MRAC scheme, comparing with the fixed control method. 83
1016/j.jfranklin.2017.07.020.
18 This work addresses the velocity and attitudes consensus con- [17] Y. Chen, Y. Zhang, Z. Wang, An adaptive backstepping design for formation 84
19 trol problem of the UAVs in formation flight. In fact, the control tracking motion in an unknown eulerian specification flowfield, J. Franklin Inst. 85
20 scheme is also suitable for the trajectory tracking control problem 354 (2017) 6217–6233, https://doi.org/10.1007/s11071-016-3303-2. 86
21 of the UAVs formation, due to the full-state tracking strategy. How- [18] W. Ni, D. Cheng, Leader-following consensus of multi-agent systems under 87
ever, the consensus fight control problem under unknown external fixed and switching topologies, Syst. Control Lett. 59 (3) (2010) 209–217.
22 88
[19] Y. Liu, Y. Jia, Adaptive leader-following consensus control of multi-agent sys-
23 disturbances with general forms still needs to be investigated in 89
tems using model reference adaptive control approach, IET Control Theory
24 future. Appl. 6 (13) (2012) 2002–2008, https://doi.org/10.1049/iet-cta.2011.0649. 90
25 [20] Z. Peng, D. Wang, H. Zhang, G. Sun, H. Wang, Distributed model reference 91
26 Declaration of competing interest adaptive control for cooperative tracking of uncertain dynamical multi-agent 92
27 systems, IET Control Theory Appl. 7 (8) (2013) 1079–1087, https://doi.org/10. 93
None declared. 1049/iet-cta.2012.0765.
28 94
[21] Z. Peng, D. Wang, H. Zhang, G. Sun, Distributed neural network control for
29 adaptive synchronization of uncertain dynamical multiagent systems, IEEE 95
30
Acknowledgements 96
Trans. Neural Netw. Learn. Syst. 25 (8) (2014) 1508–1519, https://doi.org/10.
31 1109/TNNLS.2013.2293499. 97
32
This work was supported by the National Natural Science Foun- [22] G. Tao, Multivariable adaptive control: a survey, Automatica 50 (2014) 98
dation of China (grant 61304223, 61673209, 61533008), the Aero- 2737–2764, https://doi.org/10.1016/j.automatica.2014.10.015.
99
33
nautical Science Foundation (grant 2016ZA52009), and the Na- [23] G. Tao, Adaptive Control Design and Analysis, John Wiley and Sons Inc.
100
34
tional Natural Science Foundation of USA (grant ECCS1509704). [24] Y. Xu, Z. Zhen, Multivariable adaptive distributed leader-follower flight control
35 for multiple uavs formation, Aeronaut. J. 121 (1241) (2017) 877–900, https:// 101
36 doi.org/10.1017/aer.2017.42. 102
References
37 [25] Z. Zhen, D. Wang, Q. Kang, Uav flight trajectory control based on information 103
38 fusion control method, in: Proceedings of 2010 IEEE Chinese Guidance, Naviga- 104
[1] H. Pu, Z. Zhen, M. Xia, Flight control system of unmanned aerial vehicle, Trans.
tion and Control Conference, 2010, pp. 337–341.
39 Nanjing Univ. Aeronaut. Astronaut. 32 (1) (2015) 1–8, https://doi.org/10.16356/ 105
[26] L. Wen, G. Tao, H. Yang, Y. Yang, Aircraft turbulence compensation using adap-
40 j.1005-1120.2015.01.001. 106
tive multivariable disturbance rejection techniques, J. Guid. Control Dyn. 38 (5)
41 [2] D. Bennet, C. McInnes, M. Suzuki, K. Uchuyama, Autonomous three-dimensional 107
(2015) 954–963, https://doi.org/10.2514/1.G000658.
formation flight for a swarm of unmanned aerial vehicles, J. Guid. Control Dyn.
42 [27] J. Guo, G. Tao, Y. Liu, A multivariable mrac scheme with application to a non- 108
34 (6) (2011) 1899–1908, https://doi.org/10.2514/1.53931.
43 [3] K. Obermeyer, P. Oberlin, S. Darbha, Sampling-based path planning for a vi- linear aircraft model, Automatica 47 (2011) 804–812, https://doi.org/10.1016/j. 109
44 sual reconnaissance unmanned air vehicle, J. Guid. Control Dyn. 35 (2) (2012) automatica.2011.01.069. 110
619–631, https://doi.org/10.2514/1.48949. [28] G. Song, Adaptive Control for Distributed Leader-Following Consensus of Multi-
45 111
[4] Z. Zhen, C. Gao, F. Zheng, J. Jiang, Cooperative path replanning method for mul- Agent Systems, Thesis of Master of Science, University of Virginia.
46 112
tiple unmanned aerial vehicles with obstacle collision avoidance under timing [29] Y. Xu, Z. Zhen, J. Chen, Adaptive control for leader-following consensus prob-
47 constraints, Proc. Inst. Mech. Eng., G J. Aerosp. Eng. 229 (10) (2015) 1813–1823, lem, in: IEEE Chinese Guidance, Navigation and Control Conference, August 113
48 https://doi.org/10.1177/0954410014561703. 12–14, 2016, pp. 1654–1659. 114
49 [5] C. Gao, Z. Zhen, H. Gong, A self-organized search and attack algorithm for [30] C. Chua, Generic UAV Modeling to Obtain Its Aerodynamic and Control Deriva- 115
multiple unmanned aerial vehicles, Aerosp. Sci. Technol. 54 (2016) 229–240, tives, Thesis of Master of Science, Naval Postgraduate School, California, 2008,
50 116
https://doi.org/10.1016/j.ast.2016.03.022. pp. 804–812.
51 [6] Z. Zhen, Q. Hao, C. Gao, J. Jiang, Information fusion distributed navigation for 117
[31] E. Lavretsky, R. Gadient, I. Gregory, Predictor-based model reference adaptive
52 uavs formation, in: Proceedings of 2014 IEEE Chinese Guidance, Navigation and control, J. Guid. Control Dyn. 33 (4) (2010) 1195–1201, https://doi.org/10.2514/ 118
53 Control Conference, 2014, pp. 1520–1525. 1.46849. 119
[7] C. Ou, X. Wang, Z. Zhen, Monte carlo approach to the analysis of uavs control [32] E. Lavretsky, Combined/composite model reference adaptive control, IEEE Trans. 120
54
system, in: Proceedings of 2014 IEEE Chinese Guidance, Navigation and Control Autom. Control 54 (11) (2009) 2692–2697, https://doi.org/10.1109/TAC.2009.
55 121
Conference, 2014, pp. 458–462. 2031580.
56 [8] S. Lorenz, F. Walter, Collision-avoidance framework for small fixed-wing un- 122
[33] L. Crespo, M. Matsutani, J. Jang, T. Gibson, A. Annaswamy, Design and veri-
57 manned aerial vehicles, J. Guid. Control Dyn. 37 (4) (2014) 1323–1328, https:// 123
fication of an adaptive controller for the generic transport model, in: AIAA
58 doi.org/10.2514/1.G000226. 124
Guidance, Navigation, and Control Conference, vol. 47, August 10–13, 2009,
[9] S. Li, J. Boskovic, R. Mehra, Globally stable automatic formation flight control
59 pp. 1–22. 125
in two dimensions, in: AIAA Guidance, Navigation, and Control Conference and
60 Exhibit, vol. 32(1), 2001, pp. 176–182. [34] Q. Sang, G. Tao, Adaptive control of piecewise linear systems: the state track- 126
61 [10] W. Ren, R. Bear, A decentralized scheme for spacecraft formation flying via ing case, IEEE Trans. Autom. Control 57 (2) (2012) 522–528, https://doi.org/10. 127
virtual structure approach, Proc. Am. Control Conf. 32 (1) (2003) 1746–1751, 1109/TAC.2011.2164738.
62 128
https://doi.org/10.1109/ACC.2003.1239847.
63 129
64 130
65 131
66 132