Professional Documents
Culture Documents
Automatica
journal homepage: www.elsevier.com/locate/automatica
article info a b s t r a c t
Article history: This paper proposes a novel inner–outer loop control methodology to tackle the trajectory tracking
Received 23 July 2021 problem for input-saturated quadrotors. The outer-loop, encapsulating the position tracking dynamics,
Received in revised form 28 July 2023 is rendered input-to-state stable by designing a nested saturated control law with integrative action
Accepted 19 December 2023
and applying small-gain arguments. The control strategy comprises an inner-loop that relies on a
Available online 25 January 2024
hybrid controller with integral action, designed based on the modified Rodrigues parameters attitude
Keywords: description, to solve the attitude tracking problem. The hybrid formulation of the controller benefits
Autonomous vehicles from the unique properties of the mentioned attitude description. Furthermore, it also provides a
Nonlinear control systems favorable framework to overcome the global stabilizing continuous feedback topological obstruction.
Global stability For any given initial state, the resulting double-loop control architecture asymptotically tracks a
Input saturation position trajectory that satisfies some assumptions, while minimizing the distance to the desired
Trajectory tracking attitude and satisfying the saturation limits of the thrust and torque inputs. The simulation results
demonstrate the potential of the strategy. The experimental test carried out with a commercially avail-
able quadcopter and a motion capture system highlighted the proposed solution tracking capabilities,
validating thereby the global saturated trajectory tracking controller. The proposed strategy is the first
saturated MRP-based hybrid solution for quadrotor trajectory tracking.
© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
https://doi.org/10.1016/j.automatica.2024.111521
0005-1098/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
even unstable behavior when saturation occurs. In the majority quaternions. Moreover, compared to controllers designed on the
of the quadrotor control solutions comprising bounded actuation rotation group, MRP-based strategies do not require an additional
techniques, the motivation behind the design of a saturated con- discrete control state stemming from a switching mechanism
trol law stems from the singularity caused by a null thrust value. between multiple potential functions from a synergistic family
Devising a saturated positioning controller, apart from limiting and their respective feedback laws, as in Mayhew and Teel (2011),
the thrust input magnitude, ensures the attitude references well- or between multiple error functions, as in Lee (2015), to obtain a
posedness (Casau, Sanfelice, Cunha, Cabecinhas, & Silvestre, 2015; global stability result. The stability of the resulting double-loop
Martins, Cardeira, & Oliveira, 2021b; Naldi et al., 2017). For this control architecture is studied by capitalizing on the position
reason, it is easier to find works reporting only thrust force control law ISS property. The simulation results demonstrate the
saturation than works that simultaneously address the generation capacity of the strategy to track a given trajectory under the
of a priori bounded thrust and torque inputs. Cao and Lynch influence of bounded constant disturbances, to perform consecu-
(2015) devised a saturated inner–outer loop control architecture tive flip maneuvers, and to overcome considerable initial attitude
for quadrotors that yields a local asymptotic stability result. The errors, illustrating, thereby, the global nature of the tracking
solution resorts to Euler angles to describe the attitude, does not controller. The solution was experimentally validated by resorting
account for the derivatives of the attitude references, assumes a to an off-the-shelf quadrotor and a motion capture system.
bounded angular velocity, and relies on some linear approxima- The main contributions of this paper are the design and exper-
tions to study the stability of the resulting architecture. Shao, Sun, imental validation of a novel control solution for quadrotors that
Yao, Liu, and Wu (2021) developed a saturated adaptive sliding yields a global asymptotic trajectory tracking result while verify-
mode control for attitude and altitude stabilization assuming ing specified bounds on the thrust and torque inputs. The devised
bounded states and relying on Euler angles. Wang and Liu (2018) solution comprises a saturated MRP-based hybrid controller and a
proposed a hierarchical control scheme that generates bounded nested saturated position controller, with both controllers having
thrust and torque inputs. The quaternion-based controller results an integrative action. The saturated MRP-based hybrid controller
from a backstepping approach with a Nussbaum function and globally uniformly asymptotically stabilizes the attitude error
does not yield an asymptotic stability result. Furthermore, the dynamics origin and is robust to small measurement noise and
unwinding phenomenon and the impact of the attitude error in bounded constant disturbances. The outer-loop saturated control
the position tracking system are not addressed. law renders the position tracking system input-to-state stable.
In this paper, a novel saturated control strategy is proposed In the absence of external inputs, the position controller glob-
for trajectory tracking subjected to thrust and torque input con- ally uniformly stabilizes the origin of this system. The position
control law integrative action deals effectively with bounded con-
straints for quadrotors. Inspired by the methodology proposed
stant perturbations. The resulting control solution fully addresses
in Isidori et al. (2003, Appendix C), a position control law was
the coupling effect between the position and attitude tracking
developed based on nested saturation design to render the po-
systems, prevents the singularity caused by null thrust values,
sition error dynamics input-to-state stable (ISS) while verifying
accounts for the computation and impact of the derivatives of
the thrust saturation limits. The resulting controller globally uni-
the attitude references, and overcomes the attitude topological
formly asymptotically stabilizes the position error dynamics in
obstruction. Furthermore, the authors derived bounds, in terms of
the absence of external inputs. In the present work, since the an-
the control gains and parameters and independent of the initial
gular velocity and angular acceleration references result from the
conditions of the vehicle, for the thrust and torque inputs. As a
position controller and its derivatives, the saturation functions
result, for any initial state, the control structure tracks a given
are required to be sufficiently smooth, instead of only differen-
position trajectory, verifying some mild assumptions, in the face
tiable as in Isidori et al. (2003, Appendix C). Further, the spectrum
of constant disturbances and with the guarantee of complying
of functions here considered encompasses the smooth saturation
with user-specified saturation limits for the thrust and torque.
functions more frequently used in control applications, namely,
Consequently, the proposed strategy is more comprehensive than
the hyperbolic tangent, inverse tangent, or sigmoid function. Con-
the saturated solutions found in the literature and, to the best of
versely, these standard saturation functions do not fulfill the the authors’ knowledge, is the first solution for quadrotors ad-
condition σ (s) = sgn(s), for |s| ≥ 1 imposed in Isidori dressing simultaneously thrust and torque saturation with global
et al. (2003, Appendix C). Moreover, the controller here proposed asymptotic trajectory tracking properties.
includes a saturated integrative action to deal more effectively This paper is organized as follows: some preliminaries and the
with bounded external disturbances. Specifically, this inclusion notation used are presented in Section 2; the physical model is
eliminates the steady-state error caused by additive constant detailed and the control problem is formulated in Section 3; the
disturbances. In contrast, the controller proposed in Naldi et al. saturated position tracking controller is designed in Section 4;
(2017) lacks this capacity. As a result of these crucial differences, the saturated MRP-based hybrid controller for the attitude error
the necessary conditions for ISS derived are distinct from Isidori dynamic system is devised and the global asymptotic stability
et al. (2003, Appendix C) and Naldi et al. (2017, p.2). A satu- result for the full tracking system is proved in Section 5; the simu-
rated MRP-based hybrid controller with integrative action was lation results obtained with the proposed solution are presented
designed for the attitude tracking system. The hybrid formulation and discussed in Section 6; in Section 7, the implementation is
provides a suitable framework to capture the MRP discontinuity addressed and the results attained in the experimental valida-
and allows exploiting an inherent characteristic of this represen- tion with a commercially available quadcopter are displayed and
tation to automatically drive the quadrotor through the short- analyzed; lastly, some concluding remarks are drawn in Section 8.
est rotational direction (Junkins & Schaub, 2009). This feature
prevents the unwinding phenomenon and is one of the main 2. Notation and preliminaries
advantages against quaternion-based solutions, which require ad-
ditional control states (cf. Mayhew et al. (2011)) to guarantee this Rn represents the n-dimensional Euclidean space; R≥0 ex-
behavior. Hence, MRP-based solutions yield less intricate control presses the set of non-negative real numbers; N symbolizes the
structures. Furthermore, another relevant advantage stems from set of natural numbers; K Bn denotes the closed ball of radius
the MRP description only requiring three parameters to describe K centered at the origin of Rn ; Rn×m denotes the set of n × m
the attitude of a rigid body. Therefore, the MRP are a more matrices; Sn = {x ∈ Rn+1 : x⊤ x = 1} symbolizes the n-
compact attitude representation than rotation matrices and unit dimensional unit sphere; R̄n = Rn ∪ {∞} denotes the Alexandroff
2
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
compactification (Dugundji, 1966, p. 246) of Rn ; F : X ⇒ Y can be obtained from the original set by resorting to the map
represents the set-valued map F from X to Y ; C̄ denotes the Υ : R̄3 ↦→ R̄3 :
closure of the set C; dom V symbolizes the domain of the function
⎧
⎨−ϑ∥ϑ∥−2 , for ϑ ∈ R3 \ {0}
V ; V −1 (µ) expresses the µ-level set of the function V , which is
ϑs = Υ (ϑ) = ∞ , for ϑ ∈ {0} (3)
the set of points {x ∈ domV : V (x) = µ}; In ∈ Rn×n represents the ⎩ 0 , for ϑ ∈ {∞}
n-dimensional identity matrix; given the functions f : X ↦ → Y and
g : Y ↦ → Z , g◦f = g (f) denotes the composite function g of f; ei ∈ Both the original and the shadow MRP respect the following
R3 denotes a vector of zeros except for the ith entry which is 1; kinematic equation (Junkins & Schaub, 2009)
the operator ⊙ denotes the Hadamard product; ∥·∥ represents the {
(1−∥ϑ∥2 )I3 +2[ϑ]× +2ϑϑ⊤
Euclidean norm; for s ∈ Rn , ∥s∥∞ = ∥(s1 , . . . , sn )∥∞ = maxi |si | ω , for ϑ ∈ R3
ϑ̇ = T(ϑ)ω = 4 (4)
represents the L∞ -norm; for s ∈ R, sgn(s) represents the sign ∞ , for ϑ ∈ {∞}
function, which satisfies sgn(0) = 0 and sgn(s) = |s|s−1 ∀ s ̸ = 0;
[ω]× is such that [ω]× s = ω × s for each s, ω ∈ R3 , where × The mapping Rϑ (ϑ) : R̄3 ↦ → SO(3)
denotes the cross product; for a given square matrix A ∈ Rn×n , {
8[ϑ]2× +4(1−∥ϑ∥2 )[ϑ]×
tr (A) symbolizes the trace and λmax (A) denotes the maximum I3 + , for ϑ ∈ R3
Rϑ (ϑ) = (1+∥ϑ∥2 )2 (5)
eigenvalue. The class K, K∞ , and KL comparison functions used I3 , for ϑ ∈ {∞}
throughout the manuscript are in accordance with Khalil and
Grizzle (2002, Definition 4.2 and Definition 4.3). Given a vector maps a given ϑ to the corresponding rotation matrix and has
s ∈ Rn and a closed set A ⊂ Rn , ∥s∥A denotes the distance of the property Rϑ (ϑ) = Rϑ (ϑs ). For further details, the reader is
s to A and is given by ∥s∥A := infx∈A ∥s − x∥. The asymptotic referred to Junkins and Schaub (2009).
bound ∥f(t)∥a = lim supt →+∞ ∥f(t)∥ is defined for a function A hybrid system H is characterized by the data (C, F, D, G)
f(t) : R≥0 ↦ → Rm . The saturation functions considered in this work and its model can be represented by
are aligned with the following definition: {
ẋ ∈ F (x) , x∈C
H (6)
Definition 1. The mapping σ : Rm ↦ → Rm is an independent
x+ ∈ G (x) , x+ ∈ D
symmetric function, i.e., σ (s) ≜ [σ (s1 ) · · · σ (sm )], where each σ The hybrid system evolves according to the set-valued map F :
is a smooth strictly increasing function satisfying the following Rn ⇒ Rn while in the flow set C ⊂ Rn and instantaneously
properties with M > 0: changes under the set-valued map G : Rn ⇒ Rn while in the
jump set D ⊂ Rn . A solution x(t , j) to H, with t and j denoting,
σ (0) = 0 (1a)
respectively, ordinary time and jump time, is a function x :
dom x ↦ → Rn , where dom x ⊂ R≥0 × N is a hybrid time domain.
si σ (si ) > 0 ∀ si ̸ = 0 (1b) For further details, see Goebel, Sanfelice, and Teel (2012).
0 < σ̇ (si ) ≤ min(1, M |2si |−1 ) (1d) In this paper, envisioning a global trajectory tracking capacity,
a controller for quadrotors is designed. To this end, an inertial co-
ordinate frame {I } and a body-fixed coordinate frame {B}, whose
σ̈ (si ) ≤ 0, for si ≥ 0 (1e)
origin is coincident with the center of mass of the quadrotor, are
defined, and the dynamics of such vehicles are considered to be
σ̈ (si ) ≤ M −1 (1f) governed by the following set of differential equations:
Concerning rigid-body attitude description, R represents an ṗ = v, v̇ = −ge3 + Re3 Tm−1 + bp (7a)
element of the three-dimensional special orthogonal SO(3) and
the vector q ∈ S3 denotes the unit quaternion defined by the
pair (q0 , q1 ), where q0 ∈ R and q1 ∈ R3 correspond, respectively, Ṙ = R [ω]× , Jω̇ = Jω × ω + τ + dϑ (7b)
to the scalar and vector components. The map R : S3 ↦ → SO(3) 3
where p ∈ R represents the position of the aerial vehicle in the
maps a given quaternion to the corresponding rotation matrix. inertial frame, v ∈ R3 denotes the velocity in the inertial frame,
3
The double-valued{ inverse map Q} : SO(3) ⇒ S is characterized g ∈ R corresponds to the gravity acceleration, T ∈ Ω T ⊂ R>0
as Q(R) = q ∈ S3 : R(q) = R . In addition to the previous symbolizes the thrust magnitude, R ∈ SO(3) is the rotation matrix
representations, the MRP vector, ϑ ∈ R̄3 , can also be used to from the body-fixed to the inertial frame, m ∈ R is the total
parameterize the attitude. Each ϑ has a shadow MRP associated, mass of the quadrotor, ω ∈ R3 represents the angular velocity
ϑs ∈ R̄3 . Both ϑ and ϑs are related to a given unit quaternion expressed in the body-fixed frame, τ ∈ Ω τ ⊂ R3 represents the
through moment, J ∈ R3×3 corresponds to the quadrotor diagonal tensor
{ q1 of inertia, bp ∈ R3 models an unknown exogenous disturbance
, for q ∈ S3 \ {S}
ϑ = ϕ(q) = 1+q0
(2a) vector bounded by
∞ , for q ∈ {S}
{ −q ∥bp ∥∞ < bMp , (8)
1
, for q ∈ S3 \ {N}
ϑ = ϕ (q) =
s s 1−q0
(2b) 3
and dϑ ∈ R represents an exogenous disturbance vector whose
∞ , for q ∈ {N} entries are unknown, constant, and bounded by ∥dϑ ∥∞ < dMϑ .
where S = (−1, 0, 0, 0) and N = (1, 0, 0, 0) are, respectively, the The sets Ω T and Ω τ model the actuators limitation and are
south and north poles of the three-dimensional sphere. Observe given by Ω T := {T ∈ R>0 : T < Tmax } and Ω τ := {τ ∈
R3 : |e⊤ 1 τ| ≤ τϕmax , |e2 τ| ≤ τθmax , |e3 τ| ≤ τψmax }, where
⊤ ⊤
that ϕs (q) = ϕ(−q). In virtue of the original and shadow sets
being singular for different rotations, judiciously switching be- Tmax and (τϕmax , τθmax , τψmax ) represent, respectively, the maximum
tween the original and shadow sets yields a minimal non-singular attainable thrust magnitude and maximum attainable absolute
attitude representation (Junkins & Schaub, 2009). The shadow set value for each direction of τ . The relation between the inputs
3
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
where
2M 2
b
σ (M)2 Mp
+ b Mp + Kj
Λu̇ =
∗
g − Ka3 − 2bMp
2M 2 bMp
( )
2M 3 σ (M)
σ (M)2 σ (M)
+ 2
+ 4M
+ Ks
Λ∗ü = . □
g − Ka3 − 2bMp
The condition expressed in (10a) is crucial to avoid generat-
ing a null thrust value, which results in a singularity inherent
to the physics of the problem since the quadrotor goes into
freefall regardless of the attitude. Concerning conditions (10b)–
(10e), these inequalities guarantee that the trajectory r can be
tracked without violating the actuation constraints. Note that
Fig. 1. Schematic representation of the reference frames and the forces and
higher values for the trajectory parameters K∗ lead to a more
moments generated by each rotor.
restrictive range for the control actuation, which, in turn, implies
a slower asymptotic convergence to the reference trajectory. In
light of being an underactuated vehicle, the quadrotor is unable
T and τ and the thrust generated by each propeller is described to perform an arbitrary trajectory. In this regard, to prevent a
by: scenario of incompatibility between the desired rotation matrix
Rr and the thrust direction imposed by the position tracking, the
⎡ ⎤⎡ ⎤
[ ] 1 1 1 1 T1
T ⎢L −L −L L ⎥ ⎢T2 ⎥ following optimization problem (Frazzoli, Dahleh, & Feron, 2000)
=⎣ (9)
τ −L −L L L ⎦ ⎣T3 ⎦
Rd tr I3 − R−
( 1
)
= argmin r R
c1 −c2 c3 −c4 T4 R∈SO(3) (11)
where Ti ∈ (0, Timax ) and Timax ∈ R>0 , with i = {1, 2, 3, 4}, s.t. Re3 = α
denote the thrust and the maximum thrust generated by the
where α ∈ R3 denotes the thrust direction obtained from the
ith propeller, L ∈ R>0 denotes the perpendicular distance of
position control, is resorted to. The unique solution of (11) is
any given propeller to the body-fixed frame axis XB or YB , ci ∈
equivalent to finding the ‘‘closest’’ feasible rotation matrix, Rd ∈
R>0 defines a lumped parameter that relates the steady-state
SO(3), in the sense of the cost function, that verifies the restriction
thrust and yaw moment generated by a propeller in free air. For
Rd e3 = α. To this end, the degree of freedom associated with
further details, please see Martins, Cardeira, and Oliveira (2021a).
the rotation around the vector α is exploited. Note that, apart
In Fig. 1, the reference frames and the forces and yaw moments
from the condition detailed in Assumption 2, Rr can be arbitrarily
generated by each rotor are schematically represented.
defined. From Casau et al. (2015), it follows that
Let the mapping r(t) : R≥0 ↦ → Ω , given by r(t) :=
(3)
(pd , ṗd , p̈d , pd , Rr , ωr )(t), define the reference trajectory encom- Rd = Rs (θ, β) Rr (12)
passing the desired position, pd , and attitude, Rr , and the re-
where Rs ∈ SO(3) results from the Rodrigues’ rotation formula:
spective derivatives. To ensure the well-posedness of the attitude
references and establish the actuation bounds, the trajectory r(t) Rs (θ, β) = I3 + [β]× + (1 + cos (θ))−1 [β]2×
verifies the following assumption:
with β := [Rr e3 ]× α and θ = sin−1 (∥β∥) as the axis and the
Assumption 2. The trajectory r(t) is characterized by: angle of rotation, respectively. Note that the conditions stated in
Assumption 2 yield cos (θ ) = e⊤ 3 Rr α ̸ = −1 ∀ t ≥ 0. In geometric
⊤
{
r ∈ Ω ⊂ R × R3 × Ω a × Kj B3 × SO(3) × Kω B3
3
terms, Rs rotates Rr around an axis that is perpendicular to Rr e3
(3) and α, by an angle θ , so that Rd e3 and α are coincident. The
ṙ ∈ (ṗd , p̈d , pd , Ks B3 , Rr [ωr ]× , Kω̇ B3 )
control problem can finally be stated as follows:
with
Problem 3. Design bounded inputs T ∈ Ω T and τ ∈ Ω τ to
1 p̈d , e2 p̈d ) ∈ Ka1,2 B , |e3 p̈d | < Ka3 }
Ω a = {p̈d ∈ R3 : (e⊤ ⊤ 2 ⊤
globally asymptotically stabilize the set
and Ka1,2 , Ka3 , Kj , Ks , Kω , Kω̇ > 0 and where Ω satisfies e⊤
3 Rr (t)e3 A = {(r, x) ∈ Ω × χ : p = pd , R = Rd }
≥ 0 ∀ t ≥ 0 and is compact. Furthermore, the following
conditions hold with x := (p, v, R, ω) ∈ χ := R3 × R3 × SO(3) × R3 and r
verifying the conditions detailed in Assumption 2, for the closed-
g − Ka3 > 2bMp σ M
(M)
(10a) loop system that results from the application of the designed
( √ ) control law to the system (7).
1
Tmax > m g + (Ka21,2 + Ka23 ) 2 + 2 3bMp σ M
(M)
(10b) To tackle Problem 3, saturated position and attitude con-
trollers are devised for the tracking error dynamics that result
τϕmax > e⊤
1 Je1 νω̇ + |e3 Je3 − e2 Je2 |νω + dMϑ
∗ ⊤ ⊤ ∗
(10c) from considering the reference trajectory r(t) and the evolution
of the vehicle state vector x(t). Note that the upper bounds on
τθmax > e⊤ the thrust and torque detailed in Problem 3 can be specified by
2 Je2 νω̇ + |e3 Je3 − e1 Je1 |νω + dMϑ
∗ ⊤ ⊤ ∗
(10d)
the user.
τψmax > e⊤
3 Je3 νω̇ + |e1 Je1 − e2 Je2 |νω + dMϑ
∗ ⊤ ⊤ ∗
(10e)
4. Nested saturated position tracking
with
)2 Let the position and velocity tracking errors be defined by
3 Λ∗u̇
(
+ Kω2 ( ( )2 ) 12
νω∗ = , νω̇∗ = 3 Λ∗ü + Λ∗u̇ Kω + Kω̇2 , p̃ = p − pd , ṽ = v − ṗd (13)
2
4
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
From (7a), the error dynamics are expressed as follows: result directly from the relation between the saturation levels Mv
and Mp . For further details, the reader is referred to Sepulchre,
p̃˙ = ṽ, ṽ˙ = −ge3 + Re3 Tm−1 + bp − p̈d (14) Janković, and Kokotović (1997).
From the position control, it is intended to obtain the magnitude In Theorem 4, the stability result is formalized by proving
and direction of the thrust force required to perform the desired that the referred nested manifolds are invariant and applying
trajectory r(t). To this end, the rotation matrix R and the thrust small gain arguments. Although the method followed in Isidori
force T are considered inputs and a control vector up ∈ R3 \ {0} et al. (2003) inspired the proof of Theorem 4, the inequalities
satisfying guaranteeing the reachability of the nested manifolds and the
fulfillment of the small gain condition are distinct due to the vicis-
up := Re3 Tm−1 (15) situdes of the sufficiently smooth saturation functions considered.
is designed. In this way, the thrust input can be computed Furthermore, in contrast to Isidori et al. (2003), the following
through proof resorts to ISS Lyapunov functions.
T := m∥up ∥ (16) Theorem 4. Let the conditions expressed in Assumption 2 hold for
all t ≥ 0. Suppose that the control law design parameters verify the
To avoid the generation of non-positive thrust values, comply
following conditions
with the input saturation limit, and render the position error
dynamics ISS, a position control law was devised based on a Mv > Mp (22a)
nested saturation design. Let z1 = p̃ and consider the position
error kinematics (14). A saturated smooth-state feedback control σv (Mv ) > 2 max bMp + Mζ , kp Mv k−
{ ( 1
)}
v + Mp (22b)
law ṽ∗ is designed with the intent of asymptotically stabilizing
the origin of p̃˙ = ṽ∗ . In this direction, consider the virtual control
kv > (σp Mp )−1 max 4kp Mp2 (σp Mp )−1 , Mv
( ) { ( ) }
ṽ∗ = −σ p (kp p̃), with kp > 0 and where σ p is a saturation (22c)
function in line with Definition 1, with Mp as saturation level. By
ρa
{ }
adding and subtracting the virtual control input to the kinematic Mv + Mζ < max ρa1 , √2 (22d)
3m
equation in (14) and defining z2 = ṽ − ṽ∗ , one arrives to the
following change of coordinates: with
( ( ) 1)
z1 = p̃, z2 = ṽ + σ p (kp p̃) (17) ρa1 = g − Ka3 , ρa2 = Tmax − m g + Ka21,2 + Ka23 2 .
Given (14) and (15), ż2 satisfies: Then, the position error dynamics (21) are Input-to-State Stable
without restrictions on the initial state and restriction (8) on the
ż2 = up − ge3 − p̈d + bp + σ̇ p (kp z1 ) ⊙ kp ż1 (18) input b)p . Moreover, T ∈ Ω T for r ∈ Ω and each solution
p̃, ṽ, ζ (t ) defined for t ≥ 0.
(
Let the control law up be defined as
5
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
v + σp Mp
1
∥y1 ∥∞ ≤ kp Mv k−
( ( ))
Then, it follows that any trajectory z(t) starting in R3 × R3 × R3 \ (34)
Ω 2 enters Ω 2 in finite time. Hence, there exists a time instant Let Vz2 : R3 × R3 ↦ → R≥0 be defined by
t ∗ > 0 such that ∥z2 (t)∥∞ < Mv k−v
1
∀ t ≥ t ∗ . For z(t) ∈ Ω 2 \Ω 1 ,
in virtue of (22c), the following inequality is verified 3 ∫ kv e⊤ z ∫ kζ e⊤ ζ
∑ i 2 kv i
Thus, for z(t) ∈ Ω 1 ∩ Ω 2 , the asymptotic bound on the initial state and restriction (8) on the input bp . In this
direction, according to Isidori et al. (2003, Definition B.2.2.), the
i z2 ∥a , kp Φ1 ∥ei z2 ∥a
∥ei ż1 ∥a ≤ 2 max ∥e⊤ , for i = 1, 2, 3
⊤
{ ( ⊤ )}
trajectory )(z1 , z2 )(t) is asymptotically bounded by ∥(z1 , z2 )(t)∥a ≤
Φz ∥bp ∥a , where Φz is a class K function. Considering the
(
follows from (33). The latter expression is equivalent to ( trans-
formation (17), for (z1 , z2 ) = (0, 0) one has x̃p = p̃, ṽ =
)
2Mp
∥e⊤
i ż1 ∥a ≤ i z2 (t)∥a , for i = 1, 2, 3
( ) ∥e⊤ (0, 0). Moreover, an asymptotically bounded trajectory (z1 , z2 ) (t)
σp Mp implies an asymptotically bounded trajectory x̃p (t):
Hence, the interconnection terms y1 = σ̇ p (kp z1 ) ⊙ kp ż1 and y2 = ∥x̃p (t)∥a ≤ Φp ∥bp ∥a
( )
(38)
z2 , for z(t) ∈ Ω 1 ∩ Ω 2 , fulfill the asymptotic bounds
where Φp is a class K function. Hence, (21) is also ISS without
i y1 (t)∥a ≤ Φ3 ∥ei y2 (t)∥a ,
∥e⊤ for i = 1, 2, 3
( ⊤ )
restrictions on the initial states and restriction (8) on the dis-
turbance bp . Concerning the bounds on T , applying the triangle
i y2 (t)∥a ≤ max Φ2 ∥ei y1 (t)∥a ,
∥e⊤
{ ( ⊤ )
inequality to the norm of the control law (19) yields
Φ2 ∥ei bp (t)∥a , for i = 1, 2, 3
( ⊤ )}
∥up ∥ ≤ ∥σ v (kv z2 ) ∥ + ∥σ ζ kζ ζ ∥ + ge3 + ∥p̈d ∥
( )
(39)
2kp Mp
where Φ3 (s) = σp (Mp )
s is a class K function as well. In further- Given the properties detailed in Definition 1, (10a), (16), and
ance of proving that the system (20) is input-to-state stable, the (22d), one has
functions Φ3 and Φ2 must satisfy
√ 1
0 < T < m(g + 3(Mv + Mζ ) + (Ka21,2 + Ka23 ) 2 ) (40)
Φ3 ◦ Φ2 (r) < r , for r > 0 (36) Thus, it follows from (10b) and (22d) that T ∈ Ω T for r ∈ Ω and
In this direction, the composition Φ3 ◦ Φ2 (r) yields each solution (x̃p , ζ )(t) defined for t ≥ 0. ■
Φ3 ◦ Φ2 (r) = 4kp Mp Mv (kv σp Mp σv (Mv ))−1 r In light of (22), Mv and kv have a positive lower-bound,
( )
whereas Mp , kp , Mζ can be set with an arbitrarily small positive
Mp
Since σp and σv are in line with Definition 1, σp (Mp )
= Mv
σv (Mv )
= value. In more detail, for any selection of the control gains and
M parameters, one has
σ (M )
. Thus, given the small gain condition contained in (22c), the
inequality (36) holds. Furthermore, it also implies Mv > 2M
b ,
σ (M) Mp
kv > M
σ (M)
Φ2 ◦ Φ3 (r) < r , for r > 0 (37) In addition, note that (22) implies
( { ρa
} )( )−1
To apply the small gain theorem, the origin of (20) is required Mζ < max ρa1 , √2 − 2M
b
σ (M) Mp
1+ 2M
σ (M)
3m
to be stable in the absence of inputs. Thereby, profiting from the
existence of a time τ such that z(t) ∈ Ω 1 ∩ Ω 2 ∀ t ≥ τ , consider Compared to the ISS conditions outlined in Isidori et al. (2003)
bp = 0 and let Vz : R3 × R3 × R3 × ↦ → R≥0 be defined by and Naldi et al. (2017), due to not imposing σ (M) = M in
Vz (z) := c1 Vz1 (z1 ) + c2 Vz2 (z2 , ζ), with c1 = 4k3p and c2 = k−1
v .
this work, the inequalities (22b)–(22c) depend on the value of
The function Vz is continuous, positive-definite with respect to its the saturation functions at the respective saturation level rather
origin, and radially unbounded. Given (1d), the time derivative V̇z than depending exclusively on the saturation levels. Furthermore,
satisfies considering |s| ≥ σ (|s|) hinders the derivation of the class K
3 functions of the asymptotic bounds, (33) and (35), since extract-
ing an upper bound for |s| from an inequality involving σ (|s|)
∑
i z1 ||ei z2 | − c1 ei z1 σp kp ei z1
c1 |e⊤ ⊤ ⊤ ⊤
( )
V̇z ≤
becomes less straightforward. Given this drawback, the deriva-
i=1
( ⊤ )2 tion leverages the concave function σ (|s|), leading to the distinct
i z2 σv kv ei z2 − σv kv ei z2
+ kp e⊤
( ⊤ )
condition (22c). Lastly, due to the inclusion of the saturated
+ kp |σp kp e⊤ integrative action, the convergence to the positively invariant
( ) ( ⊤ )
i z1 ||σv kv ei z2 |
nested manifolds is influenced, resulting in one of the necessary
For z(t) ∈ Ω 1 ∩ Ω 2 , by combining the bound (32) with the inequalities for the reachability of unsaturated regions, (22b),
properties (1b) and (1d), and the condition (22c), the previous being dependent on its saturation level. It is worth noting that
equation yields the saturated integrative term allows compensating effectively
3
[ ] for constant additive bounded disturbances. On the other hand,
∑ 4k2p −kp
z ∗⊤
ei e⊤ ∗ the solutions proposed in Isidori et al. (2003) and Naldi et al.
V̇z ≤ − 3 i z ≤ 0
i=1
−kp 4 (2017) lack this feature, thus failing to eliminate the steady-state
( ) ( ) error caused by such disturbances.
where e⊤ ∗ ⊤
i z = [|σp kp ei z1 | |σv kv e⊤ ⊤
i z2 |] . The function V̇z
is negative for z(t) ∈ Ω 1 ∩Ω 2 . Therefore, in light of the trajectory 5. Saturated global asymptotic tracking for the full system
reaching the positively invariant set Ω 1 ∩ Ω 2 for any given initial
state z(0), it follows from Khalil and Grizzle (2002, Theorem 4.8) The position control designed in Section 4 determines the
that the origin of the system (20), in the absence of inputs, is thrust force magnitude and direction required to track the trajec-
globally uniformly stable. Since the subsystems Σ 1 and Σ 2 are tory. Accordingly, the vector up is reassigned to up := Rd e3 Tm−1 .
ISS without restrictions on the initial state and restriction (8) on With this redefinition, the position error dynamics reshape into
the input bp , the trajectories of (20) are defined for all t ≥ 0,
˙
⎧
and the restrictions (30) and (34) are fulfilled for all t ≥ τ , the ⎨p̃ = ṽ
⎪
interconnection terms y1 and y2 verify the asymptotic bounds ˙ṽ = (R − Rd ) ∥up ∥e3 + up − ge3 − p̈d (41)
(33) and (35). Furthermore, according to Isidori et al. (2003, Theo- ⎪
ζ̇ = σ v kv (ṽ + σ p (kp p̃))
⎩ ( )
rem B.3.2.), by combining these asymptotic bounds with the small
gain properties (36) and (37), and the global uniform stability The interconnection between the positional and the angular con-
result for bp = 0, it follows that (20) is ISS without restriction trol loops originates the term Γ = (R − Rd ) ∥up ∥e3 , which, in
7
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
(21), was initially represented by( bp). Bearing in mind (15), the As previously stated, the modified Rodrigues parameters are
constraint of (11) is defined as α up := up ∥up ∥−1 . In this way, resorted to parameterize the attitude. Let R̃ ∈ SO (3) denote the
Rd can be computed using (12) and, based on Martins et al. rotation matrix error given by
(2021b, Eqs. (27)–(30)), the angular velocity and acceleration
R̃ = R⊤
dR (45)
references are given by:
−1 and that satisfies R̃ = Rϑ (ϑ̃), where ϑ̃ ∈ R̄3 represents the
e1 ]⊤ R⊤
[ ]
[−e2 d u̇p ∥up ∥
ωd = (42) MRP error. The evolution of the MRP error satisfies the kinematic
3 ωr
e⊤ ˙
equation ϑ̃ = T(ϑ̃)(ω̃), where ω̃ ∈ R3 denotes the angular velocity
[
[−e2 e1 ]⊤ ([ωd ]2× e3 − R⊤ −1 ] error computed through
d üp ∥up ∥ )
ω̇d = (43)
3 ω̇ r
e⊤ ω̃ = ω − R̃⊤ ωd (46)
Since the thrust direction does not constrain the rotation around To compute ϑ̃ from R̃, the hybrid dynamic path-lifting algorithm
itself, the third component of ωd and ω̇d result from ωr . In described in Appendix is used. To better illustrate an inherent
Lemma 5, bounds for ωd and ω̇d are derived. mechanism of the MRP representation, consider α = 0 to with-
draw the hysteretic behavior. In this way, the bound ∥ϑ̃∥ ≤ 1,
Lemma 5. There exist constants Λω , Λω̇ ∈ R>0 such that corresponding to the MRP error vector associated with the short-
∥ωd (t)∥ ≤ Λω and ∥ω̇d (t)∥ ≤ Λω̇ hold for t ≥ 0. est principal rotation available (Junkins & Schaub, 2009, p. 120),
is guaranteed. This equivalence enables dealing effectively with
Proof. Let α, β, φ ∈ R>0 be defined as follows tumbling situations in which the quadrotor has performed a prin-
cipal rotation beyond ±180◦ away from the angular reference. In
α = max |σ̇ (s)σ (s)|, β = max |σ̈ (s)σ (s)2 |, practical terms, this translates into the quadrotor completing the
s∈R s∈R
revolution instead of attempting to force it back. Hence, this ad-
φ = max |σ̈ (s)σ (s)| vantageous feature prevents the unwinding phenomenon without
s∈R
requiring the introduction of a logic state in the control structure,
The time derivative of (19) yields:
as in quaternion-based solutions (cf. Casau et al. (2015) and
(3)
u̇p = −σ̇ v (kv z2 ) ⊙ kv ż2 − σ̇ ζ (kζ ζ ) ⊙ kζ ζ̇ + pd (44) Mayhew et al. (2011)). Further, compared to solutions developed
directly on SO(3) (cf. Lee (2015) and Mayhew and Teel (2011)), by
Given the characteristics of the saturation function, one notes that designing a control law that exploits this inherent mechanism, it
the maximum value for ∥u̇p ∥∥up ∥−1 occurs when the saturation is possible to attain a global stability result without resorting to
functions σ v and σ ζ are operating within the respective unsatu- multiple potential or error functions.
rated region. With this in mind and recalling√ Assumption 2, (1d), A saturated hybrid control methodology relying on Lyapunov
(22b), and (22c), the bound ∥u̇p ∥∥up ∥−1 ≤ 3Λu̇ is derived, with stability was designed considering the position control presented
in Section 4. Let the output of the MRP-based controller,
kv (αv + 2−1 σv (Mv ) + kp αp ) + kζ σv (Mv ) + Kj τ (ϑ̃, ω̃, ϵ̃, τ f ) : R̄3 × R3 × R3 × R3 ↦→ R3 , be defined by saturated
Λu̇ = ,
g − Ka3 − σv (Mv ) − σζ (Mζ ) feedback control law
Then, considering (42), the norm ∥ωd ∥ verifies τ = −T(ϑ̃)⊤ σ ϑ (kϑ ϑ̃) − σ ω (kω ω̃) − σ ϵ (kϵ ϵ̃) − τ f (47)
with kϑ , kω , kϵ > 0 and where σ ϑ , σ ω , and σ ϵ are saturation
√
∥ωd ∥ ≤ 3Λ2u̇ + Kω2 = Λω
functions, with the properties detailed in Definition 1, with Mϑ ,
The time derivative of (44) yields: Mω , and Mϵ , respectively, as saturation levels, ϵ̃ ∈ R3 is an integral
state that verifies ϵ̃˙ = kϑ ω̃ and τ f ∈ R3 is defined as follows
üp = −k2v σ̈ v ⊙ ż2 ⊙ ż2 − kv σ̇ v ⊙ z̈2
τ f = [Jω̄]× ω̄ − JR̃⊤ ω̇d , with ω̄ = R̃⊤ ωd (48)
− k2ζ σ̈ ζ ⊙ ζ̇ ⊙ ζ̇ − kζ σ̇ ζ ⊙ ζ̈ + p(4)
d
In light of 4T(ϑ̃)⊤ ϑ̃ = (1 + ∥ϑ̃∥2 )ϑ̃ (Junkins & Schaub, 2009),
For readability, the arguments of the derivatives of the saturation
when the saturation function σ ϑ operates within its unsatu-
functions in (5) were omitted. As in the computation of the previ-
rated region, T(ϑ̃)⊤ σ ϑ (kϑ ϑ̃) can be well approximated by a vector
ous bound, the maximum value for ∥üp ∥∥up ∥−1 occurs when σ v
collinear with the principal axis of rotation and proportional to
and σ ζ operate within the respective unsaturated region. In this
the principal angle of rotation.
direction, considering the conditions detailed in Assumption
√ 2,
The attitude error dynamic model is constructed within the
(1d), (1f), one arrives to the bound ∥üp ∥∥up ∥−1 ≤ 3Λü , with
hybrid system framework to encapsulate the switches between
σv (Mv ) σv (Mv ) the original and the shadow MRP error sets and the subsequent
k2v (βv + 2φv ( + kp αp ) + 1
( + kp αp )2 )
effects. Let rh = (Rd , ωd ) ∈ Ω h ⊂ SO(3) × Λω B3 , xh :=
2 Mv 2
Λü =
g − Ka3 − σv (Mv ) − σζ (Mζ ) (ϑ̃, ω̃, ϵ̃, rh ) ∈ χh , χh = R̄3 × R3 × R3 ×Ω h , with Ω h being a com-
k2p ( M2v max{2φp , 1} + kv max{αp , βp }) +
Mv2 kp pact set, in order to define the hybrid system H = (C, F, D, G):
2kv Mp
+
g − Ka3 − σv (Mv ) − σζ (Mζ ) T(ϑ̃)ω̃
⎛ ⎞
σv (Mv ) ⎜J−1 (∆ (ω̃, ω̄) + τ + τ + d )⎟
(k2v + kv kζ )(αv + + kp αp ) + kζ αv kv
2
⎜ f ϑ ⎟
+ F(xh ) := ⎜ kϑ ω̃ (49a)
g − Ka3 − σv (Mv ) − σζ (Mζ )
⎜ ⎟
⎟
Rd [ωd ]×
⎜ ⎟
kζ Mζ−1 σv (Mv )2
2
+ kζ kv (αζ − Mζ ) + Ks
⎝ ⎠
+ Λω̇ B3
g − Ka3 − σv (Mv ) − σζ (Mζ )
In this way, bearing in mind (43), ∥ω̇d ∥ is bounded by: C(xh ) := {xh ∈ χh : ∥ϑ̃∥ ≤ 1 + δ} (49b)
√
∥ω̇d ∥ ≤ 3(Λü + Λu̇ Kω )2 + Kω̇2 = Λω̇ ■ G(xh ) := (Υ (ϑ̃), ω̃, ϵ̃, Rd , ωd ) (49c)
8
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
τθmax > e⊤
2 Je2 Λω̇ + 2
−1 ⊤
1 Je1 |Λω + ι
|e3 Je3 − e⊤ 2
(51b)
τψmax > e⊤
3 Je3 Λω̇ + 2
−1 ⊤
2 Je2 |Λω + ι
|e1 Je1 − e⊤ 2
(51c)
The switching between the original and the shadow sets is hys- set Āh be defined as follows:
teretic due to the inclusion of the parameter δ . In this way,
Āh = {xh ∈ χh : ϑ̃ = 0, ω̃ = 0, σ ϵ (kϵ ϵ̃) = dϑ } (54)
assuming the measurements are corrupted by an upper-bounded
noise, by setting the hysteresis parameter δ to a greater value, To demonstrate that Vh is positive-definite with respect to the set
noise-induced chattering is avoided. Thus, the switching becomes Āh , the functions V̄h and Vh∗ are evaluated separately. It is straight-
robust to measurement noise. In Fig. 3, the switching algorithm forward to verify that V̄h is positive-definite with respect to its
and the hysteresis region are illustrated. origin. Following from the symmetry property of the saturation
Before analyzing the stability of the hybrid system H, the function, the properties of the absolute value function, and the
relevant property of this closed-loop system stated in Lemma 6 condition Mϵ > dMϑ , Vh∗ is lower-bounded by
is worth noticing.
3 ∫
∑ kϵ |e⊤
i
ϵ̃| ∫ | e⊤ d |
i ϑ
Lemma 6. The hybrid system H is well-posed. Vh∗ ≥ σϵ (µ) dµ + σϵ−1 (µ) dµ − kϵ |e⊤
i ϵ̃||ei dϑ |
⊤
i=1 0 0
Proof. The flow set C is closed and the map Υ : R̄3 ↦ → R̄3 (55)
is continuous on D. Consequently, since the inverse image of a
closed set under a continuous mapping is closed, the jump set and, since σϵ is a non-decreasing function, similar to the approach
D is closed. The flow map F is a single-valued mapping. The followed in Casau et al. (2015), directly apply the Young’s in-
mapping ωd ↦ → Λω̇ B3 is convex and bounded. Furthermore, it equality to prove that Vh∗ }is positive-definite with respect to the
set ϵ̃ ∈ R3 : σ ϵ (kϵ ϵ̃) = dϑ . Therefore, the function Vh is positive-
{
is independent of xh , the graph is closed and, therefore, outer
semicontinuous. The remaining functions that F encompasses definite on C ∪ D with respect to the set Āh . The time derivative
of Vh is characterized by V̇h = −kϵ kϑ ω̃ σ ω (kω ω̃), from which one
⊤
are continuous on C, thus, are outer semicontinuous and locally
bounded, and correspond to differential equations, which, accord- concludes that V̇h ≤ 0 ∀ xh ∈ C. The evolution of Vh during
ing to Goebel et al. (2012, Assumption 6.5), can be identified jumps is expressed through
with a hybrid system satisfying hybrid basic conditions. Since the 3 ∫ kϑ e⊤ Υ (ϑ̃)
i
single-valued mapping Υ is continuous and Υ (ϑ̃) yields ∥ϑ̃∥ ≤
∑
Vh (G (xh )) − Vh (xh ) = σϑ (µ) dµ (56)
(1 + δ )−1 for xh ∈ D, and ω̃, ϵ̃, and rh remain constant during i=1 kϑ e⊤
i
ϑ̃
jumps, D × G(D) is closed and G(D) is bounded. Hence, G is outer
semicontinuous (Goebel et al., 2012, Lemma 5.10) and locally Given (3) and the fact that, for xh ∈ D, the condition ∥ϑ̃∥2 ≥
bounded (Goebel et al., 2012, Definition 5.14) relative to D. Hence, (1 + δ )2 holds, the latter equation simplifies into
since the hybrid basic conditions stated in Goebel et al. (2012, 3 ∫ kϑ e⊤ ϑ̃
∑ i
Assumption 6.5) are met, it follows from Goebel et al. (2012, Vh (G (xh )) − Vh (xh ) ≤ − σϑ (µ) dµ (57)
Theorem 6.30) that H is well-posed. ■ −kϑ e⊤ ϑ̃
i=1 i (1+δ )2
In Theorem 7, the global asymptotic stability result for the set Note that the jump always occurs from the exterior of the sphere
Ah := {xh ∈ χh : ϑ̃ = 0, ω̃ = 0} is demonstrated. of radius 1 + δ to its interior. In addition, the saturation function
σ ϑ is odd and nondecreasing. Thereby, it follows that Vh (G (xh ))−
Theorem 7. Let the conditions expressed in Assumption 2 hold for Vh (xh ) < 0 ∀ xh1 ∈ D. Thus, Vh is monotonically decreasing
all t ≥ 0 and let Mϵ > dMϑ . The solutions to H are complete and along flows and strictly decreasing during jumps, which implies
bounded, and the compact set Ah is globally asymptotically stable. that any solution xh (t , j) to H remains in U for all (t , j) ∈ dom xh .
9
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
Fig. 5. Simulation responses attained during trajectory tracking with the control
designed. From left to right, top to bottom: (a) x, (b) y, (c) z, and (d) Euler angles.
Fig. 9. Position and MRP error norms in the experimental test. From left to
right: (a) ∥p̃∥, (b) ∥ϑ̃∥.
Fig. 11. Saturation functions in the experimental test. From left to right, top
to bottom: (a) σ p (kp p̃), (b) σ ϑ (kϑ ϑ̃), (c) σ v (kv (ṽ + σ p (kp p̃))), (d) σ ω (kω ω̃), (e)
σ ζ (kζ ζ ), and (f) σ ϵ (kϵ ϵ̃).
Fig. 12. Thrust and Moments resulting from the control law implemented during
the experimental validation. From left to right, top to bottom: (a) Thrust; (b)
Moments.
Fig. 10. Position and yaw in the experimental test. From left to right, top to during the first seconds. Notwithstanding, throughout and after
bottom: (a) x, (b) y, (c) z, and (d) yaw.
this initial phase, the strategy exhibited a stable response and
converged to the desired trajectory, evidencing the capacity of
the saturated control solution. The resulting actuation, in terms
The position and MRP error norms are displayed in Fig. 9. The
of thrust and moments, is displayed in Fig. 12. Compared to
error norms converge to values close to zero, demonstrating the
tracking goal achievement. In particular, the position and MRP the simulation test, in the experimental validation, the actuation
error norms, after the first 10 s, are kept under 0.08 m and 0.03, has a higher fluctuation. Having considered a lumped parame-
respectively. To embark on a more detailed analysis, the position ter model to characterize the rotor (see Martins et al. (2021a)),
and yaw angle experimental responses are exhibited in Fig. 10. in conjunction with other high-order effects neglected in the
The control methodology converges to the desired position tra- simulation model and measurement noise, is a possible cause
jectory within the first 10 s of the experiment. After this period of this behavior. The quadrotor exhibited a stable response and
of convergence, the position experimental responses, component- converged to the desired trajectory while keeping the actuators
wise, do not deviate more than 7 cm from the given trajectory.
within their limits of operation, evidencing the tracking capacity
Regarding the yaw angle response, the strategy converges in less
of the saturated strategy. Thus, the results validate the control
than 5 s and does not exceed a tracking error of 2 deg afterward.
The evolution of the saturation function terms are depicted architecture and are consistent with the theoretical bounds and
in Fig. 11, with the respective saturation levels also plotted. The asymptotic stability properties discussed in Sections 4 and 5. A
significant initial position error, p̃(0) = (2.5, −0.9, −0.5) [m], video of the experimental validation is available at https://www.
led to most of these terms operating in the saturated region youtube.com/watch?v=Bc7VuAG7Wyk.
13
L. Martins, C. Cardeira and P. Oliveira Automatica 162 (2024) 111521
Luís Martins received the M.Sc. degree in mechani- He made several post-docs and sabbatical leaves, namely in IRIT and LAAS
cal engineering from Instituto Superior Técnico (IST), in Toulouse France, CERN in Geneva, Switzerland, and Schneider-Electric in
Lisbon, Portugal, in 2019. He is a Ph.D. student in Seligenstadt, Germany.
Mechanical Engineering at IST and his ongoing disserta-
tion project explores the design, stability analysis, and
validation of nonlinear control strategies for unmanned
aerial vehicles. His research interests include hybrid Paulo Oliveira received the Ph.D. degree in Electrical
systems, control theory, and autonomous vehicles. and Computer Engineering, in 2002, respectively, and
the Habilitation in Mechanical Engineering in 2016, all
from Instituto Superior Técnico (IST), Lisbon, Portugal.
Since 2020, he holds a joint position as Full Professor
Carlos Cardeira was born in Quinjenje, Angola, and in the Mechanical Engineering and Electrotechnical and
received the engineering and master of science degrees, Computer Engineering Departments of IST, is the Vice-
in electrical engineering from Instituto Superior Técnico president for the Research Affairs at the Associated
in Lisbon — Portugal, in 1986 and 1991, respectively. Laboratory for Energy, Transports, and Aeronautics,
He received the Ph.D. degree in electrical engineering and the Coordinator on Aerospace Engineering Doc-
and computer science from the Institut National Poly- toral Program, at IST. His research interests are on
technique de Lorraine in Nancy — France, in 1994. He Autonomous Robotic Vehicles with a focus on Mechatronic Systems Integration,
is a member of the Center of Intelligent Systems of Sensor Fusion, GPS and Positioning Systems, and Guidance, Navigation and
the IDMEC research laboratory and teaches at Instituto Control Systems (GNC). He is author or coauthor of more than 100 journal papers
Superior Técnico in Lisbon courses in Mechatronics (90% in first quartile,) and 180 conference communications and participated in
Systems, Industrial Automation, and Informatics areas. more than 40 European and Portuguese research projects, over the last 30 years.
15