Aerospace Science and Technology 126 (2022) 107351

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Unknown input observer-based appointed-time funnel control for

quadrotors
Wenhao Zhang a,b , Xingling Shao a,b,∗ , Wendong Zhang a,b , Jiaping Qi c , Haoze Li a,b
a
Key Laboratory of Instrumentation Science & Dynamic Measurement, Ministry of Education, North University of China, Taiyuan 030051, China
b
National Key Laboratory for Electronic Measurement Technology, School of Instrument and Electronics, North University of China, Taiyuan 030051, China
c
Computing & Innovation, The University of Adelaide, South Australia 5005, Australia

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, an unknown input observer-based appointed-time funnel control policy is proposed
Received 15 September 2021 for quadrotors suffering from environmental disturbances and parametric uncertainties. To improve
Received in revised form 13 December 2021 the robustness of quadrotors, unknown input observers (UIOs) are respectively devised in trajectory
Accepted 10 January 2022
and attitude levels to online identify unknown system dynamics via utilizing filtering manipulations,
Available online 29 January 2022
Communicated by Xiande Fang
possessing a concise structure, low calculation consumption and asymptotic error convergence. Then, an
appointed-time funnel control consisting of a modified funnel variable is proposed to regulate trajectory
Keywords: tracking error profiles within preassigned boundary without incurring singularity problem, where a
Appointed-time continuous piecewise function is embedded to specify the funnel envelop, such that the stringent
Funnel control temporal constraints can be arbitrarily prescribed by operators, independent of initial states and system
Unknown input observer arguments. All error signals involved in quadrotor dynamics are demonstrated to be ultimately uniformly
Quadrotors bounded (UUB) and trajectory tracking errors are constrained to designed funnel. Eventually, simulations
and comparisons validate the effectiveness of proposed control approach.
© 2022 Elsevier Masson SAS. All rights reserved.

1. Introduction serving as prevailing tools to recover the unknown dynamics,

are typically employed to accommodate the negative effect in
Over the past several years, worldwide research scholars pay in- a feedforward manner, and fruitful viable disturbance observers
creasing attentions to quadrotor unmanned aerial vehicle due to its have been exploited [9–17]. In [9], a composite disturbance re-
notable features, comprising of simple structure, easy maintenance jection control policy is proposed for quadrotors by incorporating
and high degree of agility, giving rise to a broad application in an extended state observer (ESO). However, the observer band-
the fields of surveillance, rescue, aerial photography and plant pro- width needs to be prudently tunned to make a compromise be-
tection [1–5]. However, it is challenging for quadrotors to derive tween the convergence speed and peaking phenomenon. A func-
a reliable and safe flight controller due to the subsequent con- tion approximator-based adaptive neural control policy with a
cerns. One is that quadrotor dynamics is characterized by under- deep reinforcement learning for quadrotors is investigated in [10].
actuation, strong coupling and nonlinearity, rendering that tradi- Note that the implementation of function approximator consist-
tional linear feedback control rules are infeasible. Another is that ing of multiple weight updating is rather time-consuming and
in practical flight process, quadrotors inevitably experience envi- demanding for hardware module, inevitably yielding a notable con-
ronmental disturbances containing aerodynamic interference and trolling deviation for quadrotor system with strong maneuvering.
wind gusts. Consequently, it is meaningful yet nontrivial to realize Based on the disturbance estimation of sliding mode observers, an
a high-quality trajectory tracking simultaneously considering the equivalent-input-disturbance control is introduced in [11]. In [12],
previous conditions [6–8]. a high-order sliding mode observer-based supertwisting control is
Aiming at reinforcing system robustness and adaptiveness of reported for quadrotors to ensure a precise tracking. In [13], an
quadrotors in the event of uncertainties, disturbance observers, adaptive terminal sliding mode disturbance observer-based robust
tracking control method is developed with consideration of extra-
neous disturbances. However, the chattering phenomenon cannot
*
Corresponding author at: National Key Laboratory for Electronic Measurement
be fully surmounted for sliding model observers [11–13] originat-
Technology, School of Instrument and Electronics, North University of China, Taiyuan
030051, China. ing from the embedding of sign function, meanwhile the infor-
E-mail address: [email protected] (X. Shao). mation on upper bound of the differential of unknown dynam-

https://doi.org/10.1016/j.ast.2022.107351
1270-9638/© 2022 Elsevier Masson SAS. All rights reserved.
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

Nomenclature

p = [x, y , z] T Position vector l Length from each rotor to the center of quadrotor
v = [ẋ, ẏ , ż] T Linear velocity vector mass
 = [φ, θ, ψ]T Angle vector c Force-to-moment factor
ω = [φ̇, θ̇, ψ̇]T Angular rates vector fp = [ f x , f y , f z ] T Parametric uncertainties caused by aerody-
namic coefficients in trajectory subsystem
() Coupling term
fa = [ f φ , f θ , f ψ ] T Parametric uncertainties caused by aerody-
uT Quadrotor total thrusts
namic coefficients in attitude subsystem
m Quadrotor mass dp = [dx , d y , d z ] T Environmental disturbances in trajectory sub-
Gz = [0, 0, g ] T Gravity acceleration vector system
 
J = diag J φ , J θ , J ψ Matrix of moment of inertia da = [dφ , dθ , dψ ] T Environmental disturbances in attitude sub-
uω = [u φ , u θ , u ψ ] T Control inputs system

ics is supposed to be available, which is clearly inconsistent with appears in control actions and disturbance estimations, herein
real practice. Fortunately, based on invariant manifold principle, using invariant manifold principle instead of finite-time sta-
unknown input observer (UIO) [18] is constructed to assure ex- bility, UIOs are incorporated to recover continuous uncer-
ponential convergence towards to zero by enforcing filtering op- tainty estimation for position and attitude levels without ac-
erations upon the available system dynamics, preserving a lower quiring the upper bound of disturbance derivative, respec-
computational burden and a more convenient tuning process, since tively. Contrast to aforementioned function approximations-
merely one filtering constant rather than a series of arguments, is based quadrotor control strategies [10,15] experiencing com-
involved. With above superiorities, UIO has been successfully ap- putational explosion and sluggish weight convergence process,
plied to control automotive engines [39] and servo systems [40]. by imposing simple filtering calculation on available system
However, recent UIO-based control consequences [39,40] can only states and adjusting filtering constant solely, an asymptotic
assure an error bounded tracking, and satisfactory transient per- estimation with a reduced computational burden can be en-
formance cannot be prescribed by users, due to the lack of explicit dowed by the introduced UIO, thus improved robustness and
inclusion of convergence time and overshoot metrices in control easy parameter tuning can be simultaneously achieved for
derivation. quadrotor tracking.
To realize constrained transient and steady-state profile gov- • Unlike the existing exponential decaying based FC methods
erned by predefined function boundary, funnel controls (FCs) [19– [19–26] subject to control singularity and infinite convergence
26] are exploited by guaranteeing the boundedness of the funnel time limitations, here a modified FC consisting of a continu-
variable in terms of tracking error. For instance, in [19], an output ous piecewise funnel boundary and novel funnel variable is
feedback FC is introduced for single-input-single-output nonlinear proposed that realizes an appointed-time prescribed perfor-
systems subject to input saturation. In [20], to regulate the con- mance without incurring singularity, i.e., trajectory controlling
vergence process of tracking errors as a prior, a fuzzy echo state errors can approach to the pregiven zone within a preassigned
neural network-based FC is proposed for nonlinear systems. In convergence time designated by users, irrespective of control
[21], a FC-based adaptive control is applied to uncertain nonlinear arguments and initial states, such characteristic is paramount
multi-input-multi-output systems, such that the tracking disagree- for implementing aerial missions with stringent time require-
ment can be preserved within the preselected funnel region. By ments.
virtual of FC, tracking errors for networked systems can decay to
the predetermined specifications even with external disturbances The remainders of this paper are arranged as follows. Quadro-
and input saturation [22]. In [23], a neural-network-based FC is tor dynamics description is introduced in Section 2. Section 3 gives
designed by using a modified funnel variable to make the servo the main results, and the stability analysis is expounded in Sec-
mechanism achieve an ultimately uniformly bounded (UUB) sta- tion 4. Section 5 gives the simulations results. Conclusion is drawn
bility in face of dead-zone. However, it may encounter potential in Section 6.
singularity problems caused by the non-differential property of
funnel variable at the origin, leading to an unstable tracking result.
2. Quadrotor dynamics description
More importantly, for the reported literatures [19–26], an expo-
nential decreasing function is typically embedded as the funnel
boundary, thus one can prevent tracking errors from violating fun- Consider a trajectory tracking problem of quadrotors under
nel boundary in an asymptotic convergent sense, i.e., theoretically lumped disturbances, where quadrotors are driven by the rotor
as time tends to infinity, tracking errors could be made to approach thrusts T i (i = 1, 2, 3, 4) generated from the motor actuation. To
to the origin, clearly such phenomenon is not beneficial for oper- facilitate the description of flight states, an earth-fixed coordinate
ating quadrotors to realize a time-urgent assignment. Therefore, it E = {xe , y e , ze } and a body-fixed coordinate B = {xb , y b , zb } are
is preferable to develop a modified FC capable of avoiding control established, where the simplified sketch of quadrotors is shown in
singularity and achieving appointed-time prescribed performance Fig. 1. According to related results [27–29], the model of quadrotor
for quadrotors. dynamics can be formulated as
Enlightened by the previous discussions, an unknown input
observer-based appointed-time funnel control is proposed for ⎧
⎪ ṗ = v
quadrotors enduring lumped disturbances, the innovations of pro- ⎪
⎪
posed control method are outlined as below:
⎪
⎨ v̇ = (u T /m − Gz ) + fp + dp
(1)
⎪
⎪ ˙ = ω
• Compared with current quadrotor control schemes depending ⎪
⎪
⎩
on sliding mode observer [11–13], where chattering widely ω̇ = J−1 uω + fa + da
2
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

estimates via establishing the mapping from available states to

lumped disturbances. An appointed-time funnel control mecha-
nism is exploited to enforce transient and steady-state constraints
on trajectory controlling errors with enhanced convergence per-
formances. The diagram of proposed control strategy is drawn in
Fig. 2.

3.1. Unknown input observer

In this section, referring to the design principle of UIO [37–39],

Fig. 1. Simplified sketch of quadrotors. the following filtering operations for the available states v, uv in
quadrotor trajectory dynamics are introduced:
where the coupling term () is defined as () = [x ,  y , z ] T 
τ1 v̇f + vf = v, vf (0) = [0, 0, 0]T
with x = cos(ψ) sin(θ) cos(φ) + sin(ψ) sin(θ),  y = sin(ψ) × (2)
sin(θ) cos(φ) − cos(ψ) sin(φ), z = cos(θ) cos(φ). The total thrust τ1 u̇fv + ufv = uv , ufv (0) = [0, 0, 0]T
u T = T 1 + T 2 + T 3 + T 4 . The control input uω = [u φ , u θ , u ψ ] T
f f f
related to the rotor thrusts is defined as u φ = lT 1 − lT 3 , u θ = where vf = [ẋ f , ẏ f , ż f ] T , ufv = [u x , u y , u z ] T , uv = [x u T /m,
lT 2 − lT 4 , u ψ = −cT 1 + cT 2 − cT 3 + cT 4 .  represents the kine-  y u T /m, z u T /m − g ] T = [u x , u y , u z ] T , and τ1 denotes the non-
matic relation between the angles and the angular velocity, given negative filtering constant to be governed.
by Based on the quadrotor dynamic model (1) and filtering opera-
⎛ ⎞ tions (2), the auxiliary variable μ is designed as
1 tan θ sin φ tan θ cos φ
⎜ ⎟ 1    
=⎝0 cos φ − sin φ ⎠ . μ= v − vf − ufv + Dp (3)
τ1
0 secθ sin φ sec θ cos φ
where Dp = [ D x , D y , D z ] T , μ = [μ x , μ y , μ z ] T .
Assumption 1. [30–33]: The time derivatives of lumped distur-
bances Dp = fp + dp and Da = fa + da are hypothesized to be upper Lemma 1. [40]: For any given τ1 ∈ (0, +∞), the auxiliary variable μ is
bounded, and
 there
 exist
 unknown
 nonnegative constants D̄ P , D̄ a , ultimately bounded and satisfies:
leading to Ḋp  ≤ D̄ P , Ḋa  ≤ D̄ a .   
1   
lim lim v − vf − ufv + Dp =0 (4)
Remark 1. Assumption 1 is a traditional precondition in the re- τ1 →0 t →∞ τ1
ported disturbance observer designs [30–33] and can be readily    
which indicates v − vf /τ1 − ufv + Dp = 0 is an invariant manifold.
verified from the following aspects: Owing to limited onboard
power, the frequency spectrums of linear velocity and angular rate
Note that the invariant manifold μ = 0 in Lemma 1 exhibits a
 of fp = −kp v,
are considerably restrained, thus thetime derivative
fa = −ka ω with kp = diag kx , k y , k z , ka = diag kφ , kθ , kψ being mapping between the filtered variables vf , ufv and lumped distur-
the bounded damping matrix, could be deemed to be bounded. Be- bances Dp . Based on this, a UIO for position level is established
sides, the external disturbances including turbulence cannot main- as
tain over a long time span. Consequently, Assumption 1 is reason- 1  
able from engineering practice. Additionally, note that the upper D̂p = v − vf − ufv (5)
τ1
bound is not required to be known, greatly relaxing the strict con-
straint inherent in sliding mode observer theory [11–13]. where D̂p = [ D̂ x , D̂ y , D̂ z ] T represents the estimation of lumped
disturbances.
Control objective: This paper aims at realizing an unknown input Then, by enforcing first-order filtering 1/ (τ1 s + 1) on original
observer-based appointed-time funnel control for quadrotors, en- velocity subsystem, one has
abling:
1 1 1
v̇ = uv + Dp (6)
1) With the aid of UIO, the lumped disturbances involved in τ1 s + 1 τ1 s + 1 τ1 s + 1
trajectory and attitude dynamics can be estimated with a re-
where s denotes the Laplace operator.
markable accuracy, and related estimation errors can be regu-
By combining (2) and (6), it can be derived that
lated to converge to the origin exponentially.
2) Under the proposed controller, all the error variables are pre- 1  
served to be UUB, additionally, trajectory controlling devia- v̇f = v − vf = ufv + Dfp (7)
τ1
tion can be prescribed to a user-defined funnel envelop in an
appointed-time manner, meanwhile the adjustment of settling where Dfp = Dp / (τ1 s + 1) and obviously, it produces Dfp = D̂p .
time is free from quadrotor initial states. Therefore, the estimation error D̃p = Dp − D̂p can be rewritten in
the frequency domain:
3. Main results
τ1 s
D̃p = Dp − Dfp = Dp (8)
In this section, an UIO-based appointed-time funnel control is τ1 s + 1
proposed for quadrotors to achieve robust geometric path track-
ing, which decouples the design into position and attitude loop. Theorem 1. For the quadrotor dynamics (1) and UIO (5), the estimation
With the aid of tuning a constant gain in auxiliary state filter- errors of lumped disturbances D̃p = [ D̃ x , D̃ y , D̃ z ] T are convergent expo-
ing, UIOs are embedded to offer continuous and exact disturbance nentially to the following compact region:

3
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

Fig. 2. Configuration of the proposed control method.

  
 
D̃p  ≤ D̃ 2i (0)e −t /τ1 + τ12 D̄ 2p , i = x, y , z (9) Then, in line of UIO principle, the disturbance estimation D̂a
can be designed as
and thus Dp = D̂p for τ1 → 0 or D̄ p → 0.
1  
D̂a = ω − ωf − J−1 ufω (16)
Proof. According to (8), the time derivative of estimation errors τ2
can be derived as where τ2 represents the nonnegative filtering constant.
˙ Dp − Dfp
D̃p = Ḋp − Ḋfp = Ḋp − Remark 2. Compared to sliding model observers [11–13] subject to
τ1
estimation chattering, herein by enforcing low-pass filtering upon
1
=− D̃p + Ḋp (10) the available system dynamics, and based on explicit relationship
τ1 between filtered states and disturbances derived from invariant
manifold rule, UIOs are incorporated in position and attitude lev-
The Lyapunov function in terms of D̃p is selected as below:
els to allow for continuous observation with exponential decaying
1 and convenient parameter tuning, while the heavy calculation cost
VD = D̃Tp D̃p (11) caused by a set of weight updating encountered in neural approx-
2
imators [10,15] can be fully circumvented. Moreover, as seen from
whose derivative along the solution of (10) is calculated as
Theorem 1, a smaller filtering coefficient contributes to a more ac-
1 curate estimate, however, since filtering coefficient relates to the
V̇ D = − D̃Tp D̃p + D̃Tp Ḋp filter bandwidth, an excessively minor value may give rise to a
τ1
higher sensitivity to measurement noises. Thus, it is recommended
1 1 τ1 D̄ 2p to compromise between estimation precision and noise elimination
≤− D̃Tp D̃p + D̃Tp D̃p +
τ1 2τ1 2 ability.
1 τ1 D̄ 2p
=− VD + (12) 3.2. Appointed-time funnel controller design
τ1 2
Integrating the inequality (12) over [0, t] results in
Step 1: Let the trajectory tracking error be e1 = [e x , e y , e z ] T = p − r
− τt 1 with r = [r x , r y , r z ] T being the reference signal. To stabilize tra-
V D (t ) ≤ V D (0)e 1 + τ12 D̄ 2p (13) jectory tracking deviation within expected performance behavior
2
and surmount the deficiency of existing funnel control [34] in
Thus, we have
    prescribing finite convergence time, thus instead of using expo-
 
D̃p  = 2V D (t ) ≤ D̃ 2i (0)e −t /τ1 + τ12 D̄ 2p , i = x, y , z (14) nential decaying funnel function, a piecewise continuous func-
tion B i (t ) is developed to serve as the funnel boundary enabling
It is obvious that the estimation errors are convergent exponen- − B i (t ) < e i (t ) < B i (t ), ∀t ≥ 0, i = x, y , z:
tially to the set τ1 D̄ p when time goes to infinity, and especially,
⎧
provided that τ1 → 0 or D̄ p → 0, UIO can offer an asymptotic es- ⎨ 1  
timation against real value. B i (t ) = [( T i − t ) / T i ] 1−εi ϕi,0 − ϕi,∞ + ϕi,∞ , 0 ≤ t ≤ T i
The proof of Theorem 1 is completed. ⎩ B (t ) = ϕ , t > Ti
i i ,∞

Similarly, a series of filtering manipulation for attitude dynam-

(17)
ics is imposed to generate the auxiliary state variables ω f , ufω : where εi ∈ (0, 1), i = x, y , z, ϕi ,0 represents the initial value and
 satisfies −ϕi ,0 = − B i (0) < e i (0) < B i (0) = ϕi ,0 . ϕi ,∞ denotes the
τ2 ω̇f + ωf = ω, ωf (0) = [0, 0, 0]T maximum steady-state error and holds 0 < ϕi ,∞ < ϕi ,0 . T i is the
(15)
τ2 ufω + ufω = uω , ufω (0) = [0, 0, 0]T user-defined time.

4
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

Theorem 2. Given any T i , εi , ϕi ,∞ , ϕi ,0 , i = x, y , z satisfying the afore- Combining with the modified funnel variable (20), the virtual
mentioned design condition, the funnel function B i (t ) and its derivatives control law α 1 = [αx1 , α y1 , αz1 ] T is designed as
with regard to time are both continuous over the time domain [0, +∞).
α 1 = −k1 γ −1 z + ṙ − ζ (23)
Proof. For any t ∈ [0, T i ], the time derivative of B i (t ) can be com- where k1 = diag{kx1 , k y1 , k z1 } represents the positive gain vector
puted as for position subsystem. z = [ zx , z y , z z ] T , γ = diag{γx , γ y , γz } and
  εi ζ = [ζx , ζ y , ζz ] T .
ϕi,0 − ϕi,∞ Ti − t 1−εi
Next, designate the velocity error as η1 = [ηx , η y , ηz ] T = v −
Ḃ i (t ) = −
T i (1 − ε i ) Ti α 1 , with α 1 being the desired signal, its differential along (1) is
1  1 − ε i calculated as follows:
=− ϕi,0 − ϕi,∞
T i (1 − ε i )
  εi η̇1 = uv + Dp − α̇ 1 (24)
Ti − t 1−εi  ε i
× ϕi,0 − ϕi,∞ According to [41,42], note that the system dynamics, i.e., po-
Ti sition and angle states, are coupled originating from the under-
1  1 − ε i  ε i actuated property for quadrotors, showing that it is nontrivial to
=− ϕi,0 − ϕi,∞ B i (t ) − ϕi ,∞
T i (1 − ε i ) control the trajectory and attitude levels independently. Then, a
≤ 0, 0 ≤ t ≤ T i (18) virtual controller law α 2 associated with angle reference r =
[rφ , rθ , rψ ] T is selected to stabilize linear velocity tracking error η 1 :
Especially, one has ⎧  
⎪
⎪ α = cos(rψ ) sin(rθ ) cos(rφ ) + sin(rψ ) sin(rθ ) u T /m
B i (t ) = ϕi ,∞ , Ḃ i (t ) = 0, t ≥ Ti (19) ⎨ x2  
α y2 = sin(rψ ) sin(rθ ) cos(rφ ) − cos(rψ ) sin(rφ ) u T /m (25)
Thus, the funnel function and its differential are all continuous ⎪
over [0, +∞).
⎩ α = cos(r ) cos(r ) u /m − g
⎪
z2 θ φ T
The proof of Theorem 2 is completed.
where α 2 = [αx2 , α y2 , αz2 ] T . rφ , rθ , rψ denote the desired roll,
pitch and yaw respectively.
To constrain position controlling errors within pregiven size, a
Then, taking the inverse transformation of (25), the total thrust
modified funnel variable is designated as
u T and reference signals rφ , rθ can be calculated as follows:
ei
zi =  , i = x, y , z (20) ⎧   2
B 2i − e 2i ⎪
⎪ u = m (αx2 )2 + α y2 + (αz2 + g )2
⎨ T
 
rφ = arcsin αx2 sin(rψ ) − α y2 sin(rψ ) m/u T (26)
The time derivative of (20) can be derived as ⎪
⎪
⎩  
  12 1
 − 12   rθ = arctan αx2 cos(rψ ) + α y2 sin(rψ ) m/ (αz2 + g )
ė i B 2i − e 2i + 2
B 2i − e 2i 2 Ḃ i B i − 2ė i e i e i
żi = Based on (24), to stabilize the velocity error at the origin, the
B 2i − e 2i
virtual control law α 2 is designed as
 − 1  − 3  
2 2
= ė i B 2i − e 2i − e i B 2i − e 2i Ḃ i B i − 2ė i e i α 2 = −k2 η1 + α̇ 1 − D̂p (27)
1  
where k2 = diag{kx2 , k y2 , k z2 } with ki2 , i = x, y , z being positive
=  ė i B 2i − ė i e 2i − e i Ḃ i B i + ė i e 2i
3 controller gain.
B 2i − e 2i
 
e2 e i Ḃ i Step 2: Let the angle tracking error be e2 = [e φ , e θ , e ψ ] T =  − r ,
=  i 3 ė i − (21) the time derivative of e2 is derived as
Bi
B 2i − e 2i
 ˙ − ṙ
ė2 =  (28)
3
i = ei / − e 2i and ζi = −e i Ḃ i / B i ,
2
Given the definitions of γ B 2i To realize an asymptotic tracking of attitude angles, the follow-
(21) can be rewritten in a concise form: ing virtual control law is established:
żi = γi (ė i + ζi ) (22) α 3 = −k3 e2 + ṙ (29)

Remark 3. Different from the derivation for conventional fun- where k3 = diag{kφ 1 , kθ 1 , kψ 1 } is a positive gain vector for attitude
nel variable zi (t ) = e (t )/ ( B i (t ) − |e (t )|), experiencing the potential subsystem.
control singularity when the tracking error e (t ) tends to zero [20], Defining ωa = ω , the angular rate tracking errors can be com-
the non-differential property can be effectively addressed by intro- puted as η2 = ωa − α 3 . Combining with (1), the derivative of
ducing a modified funnel variable (20). In addition,  the reported
 η2 = [ηφ , ηθ , ηψ ]T with respect to time can be calculated as
exponential decreasing funnel bound B i (t ) = ρi ,0 − ρi ,∞ e −li t +
ρi,∞ applied in [35,36] results in an infinite convergence time for ˙ ω + J−1 uω + D∗a − α̇ 3
η̇2 =  (30)
tracking errors, hindering a guaranteed fast response for quadro- By incorporating the estimation results in (16), the practical
tors. Differently, here the appointed-time funnel control mecha-
toque action for angular rates subsystem is designed as
nism consisting of a continuous piecewise function and modified
 
funnel variables can enable tracking errors approach to preas- ˙ ω − D̂∗a
uω = J−1 −k4 η2 + α̇ 3 −  (31)
signed zone within a user-specified time T i and overcome con-
trol singularity inherent in traditional funnel control, such that an where Da∗ = Da = [ D ∗φ , D ∗θ , D ∗ψ ] T , D̂a∗
= D̂a = [ D̂ ∗φ , D̂ ∗θ , D̂ ∗ψ ] T ,
appointed-time prescribed performance can be realized for trajec- k4 = diag{kφ 2 , kθ 2 , kψ 2 } represents the controller gain to be reg-
tory tracking. ulated.

5
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

4. Stability analysis where D̃a = Da∗ − D̂a∗ .

Substituting (36) into (35) results in
To facilitate the subsequent stability analysis for quadrotors,
11  2
considering the time derivative of tracking errors involved in the V̇ ≤ −λmin (k1 ) z 2
+ λmax (γ ) η1 
+ λmax (γ ) z 2

quadrotor dynamics, i.e., funnel variables (21), linear velocity track- 2 2

 2 1  2 1  
2

ing errors (24), attitude tracking errors (28) and angular rate track-
− λmin (k2 ) η1  + η1  + D̃p  + ηT1 uv
ing errors (30), the entire error dynamics can be integrated in a 2 2
compact formulation by utilizing some fundamental derivation: 1  2 1
T
− η α 2 − λmin (k3 ) e2 + η 2 
2
+ e2 2
⎧ 1
2 2
⎪
⎪ żi = −ki1 zi + γi ηi  2 1  2 1  2
⎪
⎪  
⎪
⎨ η̇i = −ki2 ηi + D i − D̂ i + u i − αi2 − λmin (k4 ) η2  + η2  + D̃a  (37)
2 2
(32)
⎪
⎪ ė j = −k j3 e j + η j where λmin (·), λmax (·) stand for the minimum and maximum value
⎪
⎪
⎪
⎩ η̇ = −k η + D ∗ − D̂ ∗ of a matrix, respectively.
j j4 j j j According to [41], for any given u T , αi2 ∈ R,  ∈ R3 , the in-
where i = x, y , z; j = φ, θ, ψ . terconnection term uv (, u T ) − α 2 satisfies uv (, u T ) − α 2 ≤
 
δ e2 with δ = diag δx , δ y , δz being a positive definite matrix.
Furthermore, the following result can be reached by virtual of
Theorem 3. Taking the quadrotor dynamic model (1), UIOs (5), (16), and
Young’s inequality:
the control policy containing (23), (27), (29), (31) into account, given
that the initial trajectory tracking error is preassigned to satisfy the con- 1  2 1
dition − B i (0) < e i (0) < B i (0), i = x, y , z, all the error signals related to ηT1 (uv − α 2 ) ≤ λmax (δ) η1  + λmax (δ) e2 2
(38)
2 2
the quadrotor dynamics are UUB and the trajectory tracking errors can
be prescribed to the user-defined funnel envelop in an appointed-time Thus, the inequality (37) can be integrated as
manner, provided that the subsequent restrictions for controller gains  
1 2
hold: V̇ ≤ − λmin (k1 ) − λmax (γ ) z
2
1  
κ p = λmin (k1 ) − λmax (γ ) > 0, 1 1 1  2
2 − λmin (k2 ) − λmax (γ ) − λmax (δ) − η 
1
2 2 2
1
κ v = λmin (k2 ) − λmax (γ ) − λmax (δ) −
1 1
> 0,  
1 1
2 2 2 − λmin (k3 ) − λmax (δ) − e2 2
1 1 2 2
2
κ = λmin (k3 ) − λmax (δ) − e2 > 0,  2 1    
 2 1  2
2 2
− (λmin (k4 ) − 1) η2  + D̃p  + D̃a  (39)
κω = λmin (k4 ) − 1 > 0. (33) 2 2
Recalling Theorem 1, for τ1 ∈ (0, +∞), there exists an upper
Proof. Take the following Lyapunov function for the entire quadro- bound τ1 D̄ P for the estimation error D̃p , indicating that the es-
tor dynamics: timation error D̃a in attitude dynamics can also be inferred to be
1     less than an unknown positive constant as time goes to infinity:
V = z2i + ηi2 + e 2j + η2j (34)  
2  
i =x, y , z j =φ,θ,ψ lim D̃a  ≤ λmax ()τ2 D̄ a (40)
t →∞
whose derivative with respect to time is derived by invoking (32):
Therefore, (40) can be rewritten as
   
V̇ = zi żi + ηi η̇i + e j ė j + η j η̇ j  
1 2
i =x, y , z j =φ,θ,ψ V̇ ≤ − λmin (k1 ) − λmax (γ ) z
  2
= zi (−ki1 zi + γi ηi )  
1 1  2 1
i =x, y , z j =φ,θ,ψ − λmin (k2 ) − λmax (γ ) − λmax (δ) − η 
1
  2 2 2
   
+ ηi −ki2 ηi + D i − D̂ i + u i − αi2 + e j −k j3 e j + η j 1 1
  − λmin (k3 ) − λmax (δ) − e2 2
2 2
+ η j −k j4 η j + D ∗j − D̂ ∗j (35)  2
− (λmin (k4 ) − 1) η2  + τ1 D̄ P + λmax ()τ2 D̄ a (41)
Then, the following conditions fulfill using Young’s inequality:  
Given the definitions of κ = min 2κ p , 2κ v , 2κ , 2κω > 0, ϑ =
     2 
γi zi ηi  ≤ 1 λmax (γ ) z 2
+ η  , τ1 D̄ P + λmax ()τ2 D̄ a > 0, we have
1
2
i =x, y , z
V̇ ≤ −κ V + ϑ (42)
   1 
e j η j  ≤ 1 e2 2
+ η 2 ,
2 It can be further derived by solving the inequality (42):
2 2
j =φ,θ,ψ
   1 
 1
 2
 ϑ 
ηi D̃ i  ≤ η1  + D̃p  ,
2 0 ≤ V (t ) ≤ 1 − e −κ t + V (0)e −κ t (43)
2 2 κ
i =x, y , z
   1 
 1
 2
 Hence, there exists an upper bound for the error signals
η j D̃ j  ≤ η2  + D̃a  ,
2 z, η1 , e2 , η2 aforementioned in the quadrotor dynamics, and the
(36)
2 2 following results can be derived as time goes to infinity:
j =φ,θ,ψ

6
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

⎧ √
⎪ z ≤ 2ϑ/κ
⎪
⎪  
⎨ η  ≤ √2ϑ/κ
⎪
1
√ (44)
⎪
⎪ e2 ≤ 2ϑ/κ
⎪
⎪
⎩ 
η  ≤ √2ϑ/κ
2

Based on (44), given a nonnegative scalar  , one has

z2i ≤ 2ϑ/κ <  , i = x, y , z (45)

Fig. 3. Three-dimensional trajectory tracking.
Substituting (20) into (45) yields

e 2i
< (46)
B 2i − e 2i
Subsequently, it is derived that

e 2i < B2 (47)
1+ i
It follows that

 
|e i | < Bi < Bi < Bi (48)
1+ 1+
Thus, (48) can be further written as

− B i (t ) < e i (t ) < B i (t ), i = x, y , z (49)

The proof of Theorem 3 is completed.

Remark 4. To realize the convergence of error signals and as-

sure the trajectory tracking converge to the funnel envelop in an Fig. 4. Trajectory tracking errors and derivations of funnel variable. (For interpreta-
appointed-time manner, the restrictions (33) imposed on the con- tion of the colors in the figure(s), the reader is referred to the web version of this
article.)
troller gains should be satisfied. In addition, the tuning rules of the
relevant parameters are given as follows: ⎧   T
⎪ −0.3t
⎨ 0, 0, 9 1 − e , t≤9
1) For UIOs, the upper bound of estimation errors is determined r= [10 (1 − cos (2π (t − 9)/23)) , 5 (sin 4π (t − 9)/23) , (50)
⎪
⎩ 
by the filtering constants τ1 , τ2 . Based on (9), the estimated
9 1 − e −0.3t ] T , t > 9
accuracy can be considerably improved using minor filtering
constants, however, a higher sensitivity to measurement noises The initial position and angle states are set as p(0) =
may induce. Thus, it is preferable to make a tradeoff between [−2, 3, 1] T m, (0) = [0.4, 0.3, 0.2] T rad, and the desired yaw an-
estimation precision and noise filtering. gle is given as rψ = 0 rad.
2) Note that the user-defined time T i for the appointed-time fun-
nel control is utilized to regulate the convergence time, such Case 1: Efficacy verification of proposed control method
that quadrotor trajectory error can reach its final state before
To demonstrate the feasibility for the proposed control al-
T i . It should be mentioned that an excessively small T i in-
gorithm, simulation results are provided in Figs. 3–8. Three-
evitably yields a larger control authority during the transient,
dimensional tracking response is depicted in Fig. 3, indicating that
thus input saturation and convergence speed should be prop-
quadrotor is able to follow the spatial path with a fast speed and
erly concerned. And ϕi ,0 is forced to satisfy −ϕi ,0 < e i (0) <
a good precision. Fig. 4 gives the trajectory controlling error under
ϕi,0 , a proper parameter selection of ϕi,∞ is utilized to realize
proposed appointed-time funnel control, it is clear that trajectory
a reasonable size for the tracking error at steady-state.
errors never violate the preassigned funnel boundary during entire
operational time, and the differential of funnel variable is smooth
5. Simulation results and bounded without incurring control singularity. Additionally, to
illustrate the utility of funnel mechanism in prescribing conver-
In this section, the relevant simulations and comparisons are gence time, Fig. 5 plots trajectory tracking errors using different
conducted to confirm the feasibility and effectiveness for the pro- initial position states, showing that appointed-time convergence
posed control algorithm, which are executed in MATLAB/SIMULINK can be fulfilled once the user-defined time argument T i is deter-
platform, including S-function based quadrotor model (1), UIOs (5), mined, despite of initial states. Position and attitude responses,
(16) and M-file based control policy (26), (32), (35), (37), and a as well as disturbance estimates offered by UIOs are provided
fixed-step solver is used to sample and calculate the entire control in Figs. 6–7, respectively. And it is intuitively stated that bene-
algorithms with the fundamental sample time being 0.001 s. fiting from accurate disturbance observation and accommodation
Referring to the selection of quadrotor model parameters in of UIOs, the quadrotor is capable of tracking defined reference pre-
[42], the related values of model arguments and lumped distur- cisely and promptly in translational and rotational levels under
bances are listed in Table 1, and the associated parameters of strong uncertainties. Fig. 8 describes the limited control actions
controllers are collected in Table 2 inferred from Remark 4. that steer quadrotor motions.
To verify the proposed control algorithm capable of achieving
an appointed-time trajectory tracking, the trajectory reference sig- Case 2: Comparative study against continuous sliding mode con-
nal is considered as follows: troller [43]

7
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

Table 1
Model arguments and lumped disturbances.

Sections Values

Model parameters m = 2 kg, l = 0.4 m, c = 0.05, J = diag {0.16, 0.16, 0.32} kg m2 ,

kp = diag {0.01, 0.01, 0.01} N m2 , ka = diag {0.012 , 0.012, 0.012} N m2

Lumped disturbances dx = 2 (sin(t ) + sin(1.5t )), d y = 2 (cos(t ) + sin(0.5t ) − cos(0.8t )), d z = 2 sin(t ),
dφ = 0.2(sin(t ) + sin(0.5t )), dθ = 0.2 (cos(0.5t ) − cos(0.8t )), dψ = 0.2 sin(t ) cos(0.5t )

Table 2
Design parameters of proposed method.

Module Value

UIOs τ1 = 0.02; τ2 = 0.01

Appointed-time funnel control T x = T y = T z = 2; εx = ε y = εz = 0.5

ϕx,0 = 0.5, ϕ y,0 = 0.6, ϕz,0 = 0.4; ϕx,∞ = ϕ y,∞ = ϕz,∞ = 0.5;
k1 = diag {1.2, 1.2, 1.1}; k2 = diag {4, 4, 4};
k3 = diag {6, 6, 6}; k4 = diag {10, 10, 10}

Fig. 5. Trajectory tracking errors with various initial states.

Fig. 7. Estimation results for lumped disturbances using UIOs.

Fig. 6. Tracking performance of trajectory and attitude dynamics.

Fig. 8. Control inputs.

Aiming to reveal the superiority for the proposed control al-
gorithm, a comparative study between the appointed-time funnel
Sliding mode observer:
control and continuous sliding mode control [43] is given.  
⎧
Correspondingly, the design procedure and gain selection of in- ⎪
⎪ p̂˙ = v̂ + K̂1 β1 D̃p ,
volved continuous sliding mode control [43] consisting of sliding ⎪
⎨  
mode observer and nonsingular terminal sliding mode controller v̂˙ = (u T /m − Gz ) + D̂p + K̂2 β2 D̃p , (51)
is formulated as follows: ⎪
⎪  
⎪
⎩ D̂˙ = K̂ β D̃ ,
p 3 3 p
8
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

Table 3
Quantitative comparison between proposed method and continuous sliding mode controller [43].

Index Notation Proposed method Continuous sliding mode

controller [43]

Convergence time ex 1.701 2.402

ey 1.675 1.754
ez 1.412 2.191

Tracking accuracy (standard derivation) ex 0.0098 0.0170

ey 0.0314 0.0384
ez 0.0370 0.0513

Integrated absolute of control inputsa Qφ 28.01 813.2

Qθ 31.09 816.3
Qψ 101.4 101.7
a
Integrated absolute of control inputs is computed by the sum of the
! t  absolute
 value of control inputs
during the simulation time, which is mathematically expressed as Q j = 0 u j  dσ , j = φ, θ, ψ .

with
⎧ " #  1 1 1

⎪
⎪ K̂1 = diag k̂x1 , k̂ y1 , k̂ z1 = diag 3λ̂x , 3λ̂ y , 3λ̂z
3 3 3
⎪
⎪
⎪
⎪ " #  
⎪
⎪ 1 1 1
⎪
⎪ = = 3 3 3
⎪ 2
⎪
K̂ diag k̂ x2 , k̂ , k̂
y2 z2 diag 1 .53 λ̂ x , 1 .53 λ̂ y , 1 .53 λ̂ z
⎪
⎪ " #  
⎪
⎪ 1 1 1
⎪
⎨ K̂3 = diag k̂x3 , k̂ y3 , k̂ z3 = diag 1.1λ̂x3 , 1.1λ̂ y3 , 1.1λ̂z3
⎪    2
⎪
⎪  3
⎪
⎪ β 1 D̃p = D̃p  sign(D̃p )
⎪
⎪
⎪
⎪    1
⎪
⎪  3
⎪
⎪ β2 D̃p = D̃p  sign(D̃p )
⎪
⎪
⎪
⎪    0
⎩ β D̃ =  
D̃p  sign(D̃p )
3 p

(52)
where λ̂x = 11, λ̂ y = 10, λ̂z = 8.
Besides, the controllers [43] for trajectory loop are designed as Fig. 9. Estimation errors and tracking errors of trajectory loop using [43].

 3
s1 = ev + ks1 ep  2 sign(ep ),
1
v̄ = vs − ks2 s1 3 sign(s1 ),
0
v̇s = −ks3 s1 sign(s1 ), uv = v̄ + r̈ − D̂p (53)

and for attitude loop:

3 1
s2 = eω + ks4 ea 2 sign(ea ), ω̄ = ωs − ks5 s2 3 sign(s2 ),
0
ω̇s = −ks6 s2 sign(s2 ), uω = ω̄ + r̈ − D̂a (54)
where ev , eω denote the linear and angular velocity tracking er-
rors and ep , ea represent the position and angle tracking er-
rors. The controller gains are set as ks1 = diag {0.45, 0.45, 0.5},
ks2 = diag {3, 3, 3}, ks3 = {1, 1.1, 1.1}, ks4 = diag {7, 7, 4}, ks5 =
diag{10, 10, 10}, ks6 = {0.3, 0.3, 0.7}. Moreover, the quantitative
comparison between appointed-time funnel control and [43] is
collected in Table 3. Fig. 9 shows the disturbance estimation er-
rors and tracking errors in trajectory loop, it is evident that sliding
mode observer imposes severe chattering in disturbance estimates
due to the utilization of discontinuous functions, such effect can
also be detected from the high-frequency oscillated control in-
puts shown in Fig. 10. In addition, compared to [43], the pro- Fig. 10. Control inputs using [43].
posed scheme is superior in terms of shorter convergence time,
reduced tracking deviation as well as less control consumption,
which can be attributed to continuous and exact estimation of UIO, To demonstrate the merits of UIO in achieving exponential con-
appointed-time guaranteed behavior mechanism of modified fun- vergence to the origin and reduced computational cost, the pre-
nel control. vailing tools to deal with unknown dynamics online, such as radial
basis function neural network (RNFNN) [44] and sliding mode ob-
Case 3: Contrastive study among UIO, radial basis function NN [44] server (SMO) [43] are carried out for quadrotor trajectory level to
and sliding mode observer [43] make a comparison. And to assure fairness of comparisons, related

9
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

Table 4 [2] X.L. Shao, W.H. Zhang, W.D. Zhang, Improved prescribed performance anti-
Estimation performance against RBFNN [44] and SMO [43]. disturbance control for quadrotors, Appl. Math. Model. 97 (2021) 501–521.
[3] H. Ghadiri, M. Emami, H. Khodadadi, Adaptive super-twisting non-singular ter-
Index Notation UIO RBFNN [44] SMO [43] minal sliding mode control for tracking of quadrotor with bounded distur-
bances, Aerosp. Sci. Technol. 112 (2021) 106616.
Estimation error D̃ x 0.0125 0.9350 0.0273
[4] S.O. Ariyibi, O. Tekinalp, Quaternion-based nonlinear attitude control of
D̃ y 0.0099 0.5740 0.0215 quadrotor formations carrying a slung load, Aerosp. Sci. Technol. 105 (2020)
D̃ z 0.0071 0.5537 0.0212 105995.
[5] S.A. Emami, A. Banazadeh, Simultaneous trajectory tracking and aerial manip-
Computational burdena 3.14% 29.53% 4.00% ulation using a multi-stage model predictive control, Aerosp. Sci. Technol. 112
a (2021) 106573.
Computational burden is calculated by the proportion of accumulated executing
[6] H. Zhang, B.F. Song, F. Li, et al., Multidisciplinary design optimization of an elec-
time for estimators to the entire simulation time.
tric propulsion system of a hybrid UAV considering wind disturbance rejection
capability in the quadrotor mode, Aerosp. Sci. Technol. 110 (2021) 106372.
arguments are selected such that the involved three disturbance [7] P. Tang, F.B. Zhang, J.C. Ye, et al., An integral TSMC-based adaptive fault-tolerant
estimators can reach consensus in convergence time. control for quadrotor with external disturbances and parametric uncertainties,
Estimation performance against RBFNN [44] and SMO [43] is Aerosp. Sci. Technol. 109 (2021) 106415.
[8] P. Tang, D.F. Liu, D. Zheng, et al., Observer based finite-time fault tolerant
shown in Table 4, illustrating that the considered UIO can realize
quadrotor attitude control with actuator faults, Aerosp. Sci. Technol. 104 (2020)
the best estimation ability with the least estimation error and cal- 105968.
culation cost, which is consistent with aforementioned analysis in [9] K. Zhao, J.H. Zhang, D.L. Ma, et al., Composite disturbance rejection attitude
Section 3.1, since exponential decaying towards to zero can be as- control for quadrotor with unknown disturbance, IEEE Trans. Ind. Electron.
sured by selecting a sufficiently small filtering gain and only a set 67 (8) (2020) 6894–6903.
[10] Y.D. Wang, J. Sun, H.B. He, et al., Deterministic policy gradient with integral
of simple filtering manipulation is involved rather than multiple
compensator for robust quadrotor control, IEEE Trans. Syst. Man Cybern. Syst.
weight updating inherent in neural learning schemes, such features 50 (10) (2020) 3713–3725.
can be simultaneously fulfilled using UIO. Oppositely, RBFNN en- [11] W.J. Cai, J.H. Shen, M. Wu, et al., Disturbance suppression for quadrotor UAV
counters computational explosion raised by a great deal of hidden using sliding-mode-observer-based equivalent-input-disturbance approach, ISA
layer nodes, leading to a more demanding requirement on compu- Trans. 92 (2019) 286–297.
[12] J.A.C. Gonzalez, O. Salas-Pena, J. De Leon-Morales, Observer-based super twist-
tational resource of quadrotor control system. On the other hand,
ing design: a comparative study on quadrotor altitude control, ISA Trans. 109
although SMO reduces calculation burden significantly, it exhibits (2021) 307–314.
harmful chattering in disturbance estimates as depicted in Fig. 9, [13] L.Q. Dou, X.T. Su, X.Y. Zhao, et al., Robust tracking control of quadrotor via on-
which cannot be affordable for quadrotor practical implementation. policy adaptive dynamic programming, Int. J. Robust Nonlinear Control 31 (7)
(2021) 2509–2525.
[14] X.L. Shao, Y. Shi, Neural-network-based constrained output-feedback control for
6. Conclusion
MEMS gyroscopes considering scarce transmission bandwidth, IEEE Trans. Cy-
bern. (2022), https://doi.org/10.1109/TCYB.2021.3070137.
In this paper, an unknown input observer-based appointed-time [15] B. Wang, Y.Y. Shen, Y.M. Zhang, Active fault-tolerant control for a quadrotor
funnel control is proposed for quadrotors. By merely selecting a fil- helicopter against actuator faults and model uncertainties, Aerosp. Sci. Technol.
tering constant, with the aid of invariant manifold theory, UIOs are 99 (2020) 105745.
[16] X.H. Yue, X.L. Shao, W.D. Zhang, Elliptical encircling of quadrotors for a dy-
designed to offer a continuous and precise disturbance estimate via
namic target subject to aperiodic signals updating, IEEE Trans. Intell. Transp.
establishing the mapping between available quadrotor dynamics Syst. (2022), https://doi.org/10.1109/TITS.2021.3127722.
and lumped disturbances. Then, an appointed-time funnel bound [17] J.J. Sun, J. Wang, P. Yang, et al., Model-free prescribed performance fixed-time
consisting of a continuous piecewise function is proposed to as- control for wearable exoskeletons, Appl. Math. Model. 90 (2021) 61–77.
sure the tracking errors converge to the envelope, and a modified [18] J. Na, B.R. Jing, Y.B. Huang, et al., Unknown system dynamics estimator for
motion control of nonlinear robotic systems, IEEE Trans. Ind. Electron. 65 (5)
funnel variable is proposed to surmount the singularity problem.
(2020) 3850–3859.
And it should be emphasized that the trajectory tracking errors [19] N. Hopfe, A. Ilchmann, E.P. Ryan, Funnel control with saturation: nonlinear SISO
can reach preassigned steady-state before a pregiven time, which systems, IEEE Trans. Autom. Control 55 (9) (2010) 2177–2182.
is uncorrelated with the initial quadrotor states. Ultimately, simu- [20] S.I. Han, J.M. Lee, Fuzzy echo state neural networks and funnel dynamic surface
lations and comparisons exhibit the effectiveness and superiority control for prescribed performance of a nonlinear dynamic system, IEEE Trans.
Ind. Electron. 61 (2) (2014) 1099–1112.
of the proposed control rule.
[21] T. Berger, H.H. Le, T. Reis, Funnel control for nonlinear systems with known
Consider that the overall updating and calculation of associated strict relative degree, Automatica 87 (2018) 345–357.
controller are performed in a time-triggered fashion, i.e., a great [22] X.L. Shao, Y. Shi, W.D. Zhang, et al., Neurodynamic approximation-based quan-
amount of periodic states and control inputs are released over tized control with improved transient performances for microelectromechani-
sensor-to-controller and controller-to-actuator ends, yielding a sig- cal system gyroscopes: theory and experimental results, IEEE Trans. Ind. Elec-
tron. 68 (10) (2021) 9972–9983.
nificant onboard resource consumption. Thus, in the future, it is
[23] S.B. Wang, H.S. Yu, J.P. Yu, et al., Neural-network-based adaptive funnel control
of great values to exploit an event-driven anti-disturbance control for servo mechanisms with unknown dead-zone, IEEE Trans. Cybern. 50 (4)
policy for quadrotors to realize reduced sampling and transmission (2020) 1383–1394.
cost, as well as reinforced adaptiveness against uncertain environ- [24] X.W. Bu, Air-breathing hypersonic vehicles funnel control using neural approx-
ment. imation of non-affine dynamics, IEEE/ASME Trans. Mechatron. 23 (5) (2018)
2099–2108.
[25] T. Berger, S. Otto, T. Reis, et al., Combined open-loop and funnel control for
Declaration of competing interest underactuated multibody systems, Nonlinear Dyn. 95 (3) (2019) 1977–1998.
[26] D. Liberzon, S. Trenn, The bang-bang funnel controller for uncertain nonlin-
The authors declare that they have no known competing finan- ear systems with arbitrary relative degree, IEEE Trans. Autom. Control 58 (12)
cial interests or personal relationships that could have appeared to (2013) 3126–3141.
[27] X.H. Wang, B.J. Shirinzadeh, M.H. Ang, Nonlinear double-integral observer
influence the work reported in this paper.
and application to quadrotor aircraft, IEEE Trans. Ind. Electron. 62 (2) (2015)
1189–1200.
References [28] X.L. Shao, B. Tian, W. Yang, Fixed-time trajectory following for quadrotors via
output feedback, ISA Trans. 110 (2021) 213–224.
[1] K. Liu, R.J. Wang, X.D. Wang, et al., Anti-saturation adaptive finite-time neural [29] X.L. Shao, L.X. Xu, W.D. Zhang, Quantized control capable of appointed-time
network based fault-tolerant tracking control for a quadrotor UAV with external performances for quadrotor attitude tracking: experimental validation, IEEE
disturbances, Aerosp. Sci. Technol. 115 (2021) 106790. Trans. Ind. Electron. 69 (5) (2022) 5100–5110.

10
W. Zhang, X. Shao, W. Zhang et al. Aerospace Science and Technology 126 (2022) 107351

[30] S.B. Wang, L. Tao, Q. Chen, et al., USDE-based sliding mode control for servo [38] L.X. Xu, X.L. Shao, W.D. Zhang, USDE-based continuous sliding mode control for
mechanisms with unknown system dynamics, IEEE/ASME Trans. Mechatron. quadrotor attitude regulation: method and application, IEEE Access 9 (2021)
25 (2) (2020) 1056–1066. 64153–64163.
[31] A. Castillo, P. Garcia, Predicting the future state of disturbed LTI systems: a [39] A. Stotsky, I. Kolmanovsky, Application of input estimation techniques to charge
solution based on high-order observers, Automatica 124 (2021) 109365. estimation and control in automotive engines, Control Eng. Pract. 10 (12)
[32] K.X. Li, C. Peng, Z.Q. Deng, et al., Field dynamic balancing for active magnetic (2002) 1371–1383.
bearings supporting rigid rotor shaft based on extended state observer, Mech. [40] J. Na, S.B. Wang, Y.J. Liu, et al., Finite-time convergence adaptive neural net-
Syst. Signal Process. 158 (2021) 107801. work control for nonlinear servo systems, IEEE Trans. Cybern. 50 (6) (2020)
[33] C.C. Hua, K. Wang, J.N. Chen, et al., Tracking differentiator and extended state 2568–2579.
observer-based nonsingular fast terminal sliding mode attitude control for a [41] R. Falcon, H. Rios, M. Mera, et al., Attractive ellipsoid-based robust control for
quadrotor, Nonlinear Dyn. 94 (1) (2018) 343–354. quadrotor tracking, IEEE Trans. Ind. Electron. 67 (9) (2020) 7851–7860.
[34] A. Ilchmann, E.P. Ryan, P. Townsend, Tracking control with prescribed tran- [42] X.L. Shao, X.H. Yue, J. Li, Event-triggered robust control for quadrotors with
sient behaviour for systems of known relative degree, Syst. Control Lett. 55 (5) preassigned time performance constraints, Appl. Math. Comput. 392 (2021)
(2006) 396–406. 125667.
[35] H.P. Wang, Y.C. Zou, P.X. Liu, et al., Robust fuzzy adaptive funnel control of [43] H. Rios, R. Falcon, O.A. Gonzales, et al., Continuous sliding-mode control strate-
nonlinear systems with dynamic uncertainties, Neurocomputing 314 (2018) gies for quadrotor robust tracking: real-time application, IEEE Trans. Ind. Elec-
299–309. tron. 66 (2) (2019) 1264–1272.
[36] C.G. Liu, H.Q. Wang, X.P. Liu, et al., Adaptive finite-time fuzzy funnel control for [44] Y. Wu, Y.G. Pan, C. Mou, et al., Quantized adaptive finite-time bipartite NN
nonaffine nonlinear systems, IEEE Trans. Syst. Man Cybern. Syst. 51 (5) (2021) tracking control for stochastic multiagent systems, IEEE Trans. Cybern. 51 (6)
2894–2903. (2021) 2870–2881.
[37] J. Na, A.S. Chen, G. Herrmann, et al., Vehicle engine torque estimation via un-
known input observer and adaptive parameter estimation, IEEE Trans. Veh.
Technol. 67 (1) (2018) 409–422.

11