Automatica 125 (2021) 109383

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Brief paper

Dynamic output stabilization of control systems: An unobservable

kinematic drone model✩
∗
Alain Ajami a , , Jean-Paul Gauthier b , Ludovic Sacchelli c
a
Université Saint Joseph de Beyrouth, Ecole Supérieure d’Ingénieurs de Beyrouth, ESIB, Lebanon
b
Université de Toulon, LIS, UMR CNRS 7020, Campus de la Garde, 83041 Toulon Cedex, France
c
Université Claude Bernard Lyon 1, LAGEPP, UMR CNRS 5007, 69100 Villeurbanne, France

article info a b s t r a c t

Article history: The problem of dynamic output stabilization is a very general and important problem in control theory.
Received 7 August 2019 This problem is completely solved in the case where the system under consideration is uniformly
Received in revised form 14 September 2020 observable. However, usually, nonlinear systems do not share this property: in general, systems are
Accepted 11 November 2020
observable or not depending upon the control as a function of time. In this general situation, very
Available online 26 December 2020
little is known about dynamic output stabilization.
Keywords: In this paper, we solve the problem for a classical academic kinematic model for drones whose
Control systems observability properties are especially bad.
Dynamic output stabilization © 2020 Elsevier Ltd. All rights reserved.
Asymptotic stability

1. Introduction In this paper, we present a case study where a strategy for dy-
namic output time-invariant feedback stabilization is built around
Dynamic output stabilization of a dynamical system is a classi- a well-studied model for drone dynamics. The aim of this paper is
cal problem from control theory. If the stabilization is achievable not reduced to proving that this system is stabilizable. Rather, it
with a state feedback law but only an output of the system illustrates that observers can be built to be convergent even in the
is known, a natural idea is to apply this feedback to an esti- presence of observability singularities in the system, and without
mation of the state provided by an observer. In the case of
prior knowledge of the feedback law. In Lagache, Serres, and
non-linear systems, this strategy was proved to be effective under
Gauthier (2017), another example of the same kind of problem
assumption of observability for all inputs, known as uniform
is treated with similar methodology, and the reader is invited to
observability (Esfandiari & Khalil, 1992; Gauthier & Kupka, 2001;
Jouan & Gauthier, 1996; Khalil & Esfandiari, 1993; Teel & Praly, consult Brivadis, Gauthier, Sacchelli, and Serres (2019) where the
1994, 1995). In full generality, however, uniform observability is general matter is discussed.
generically not satisfied (Gauthier & Kupka, 2001), including for Our example is the following academic kinematic model of
important classes such as state affine systems (bilinear dynamics a fixed wings drone (or UAV), flying at constant altitude, with
with linear observation). There may exist input controls that constant linear velocity:
make the system unobservable and working around them is a
⎨ ẋ = cos θ,
⎧
challenging task.
There have been attempts at building strategies for stabi- ẏ = sin θ, (1)
lization of poorly observable systems (Combes, Malrait, Martin, θ̇ = u, −umax ≤ u ≤ umax .
⎩
Rouchon, et al., 2016; Coron, 1994; Shim & Teel, 2003). These
approaches, however, rely on time-varying feedback. A funda- This system is a variation on the revered Dubins model (Boscain
mental and difficult problem is posed by the construction of & Piccoli, 2003; Dubins, 1957; Soueres & Laumond, 1996) where
time-invariant strategies relying only on an estimate of the state. (x, y)-trajectories on the plane have a minimum possible radius
of curvature 1/umax . It has been extensively studied for the mod-
✩ This work is partially supported by ANR, France Project SRGI ANR-15-CE40- eling of vehicles and fixed wings drones, especially in regard to
0018. The material in this paper was not presented at any conference. This paper trajectory optimality (Ajami, Balmat, Gauthier, & Maillot, 2013;
was recommended for publication in revised form by Associate Editor Daniele Balluchi, Bicchi, Piccoli, & Soueres, 2000; Chitsaz & LaValle, 2007;
Astolfi under the direction of Editor Daniel Liberzon. Lagache, Serres, & Andrieu, 2015; Maillot, Boscain, Gauthier, &
∗ Corresponding author.
Serres, 2014; Owen, Beard, & McLain, 2015).
E-mail addresses: [email protected] (A. Ajami),
[email protected] (J.-P. Gauthier), [email protected] Endowed with the only information given by x2 + y2 , the
(L. Sacchelli). square of the distance to the origin, we ask ‘‘is it possible to

https://doi.org/10.1016/j.automatica.2020.109383
0005-1098/© 2020 Elsevier Ltd. All rights reserved.
A. Ajami, J.-P. Gauthier and L. Sacchelli Automatica 125 (2021) 109383

This leads to the following reduction of the model. We define

the target set T by
T = {(x, y, θ ) | x = r sin θ, y = −r cos θ}. (2)
This set is traveled by the system under the input u = umax .
Rather than considering stabilization to T , a classical moving
frame allows to reduce the dimension of this dynamical system
and collapse T to a single point in the plane. We set
cos θ sin θ
( ) ( )( )
x̃ x
= . (3)
ỹ − sin θ cos θ y
Fig. 1. Symmetries in the measured output cause some straight trajectories to
be indistinguishable form each other. The plain and the dashed trajectory result In these new UAV-based coordinates (x̃, ỹ, θ ), system (1) can be
in the same measurement over time.
rewritten as
{
x̃˙ = u ỹ + 1,
(4)
stabilize this system on a circular trajectory of minimal radius ỹ˙ = −u x̃.
1/umax around the origin?’’
With full information, this poses no issue. However, the dis- For a non-zero u ∈ [−umax , umax ], system (4) possesses a single
tance output is especially poor in this context. Indeed, under the equilibrium (0, −1/u). In particular for u = umax and u =
input u = 0, trajectories are straight and distance measurements −umax , they have equilibria (0, −r) and (0, r). They correspond
are indistinguishable under rotational and reflection symmetry to the target set T being browsed counter-clockwise and clock-
in the plane (see Fig. 1). The reflection symmetry can imply (for wise respectively. If u is changed for −u, the two equilibria are
exchanged so we can indifferently consider one among the two
instance) a switch from θ to −θ that is not solvable by feedback
equilibria positions for stabilization. In these new coordinates, θ
design. Given a straight trajectory and the corresponding output,
does not play a role anymore. It can be preserved as an integrator
it is not possible to know if the aircraft should steer left (u > 0)
of the control, but stabilization toward T becomes a matter of
or right (u < 0) to the target.
stabilization to a point in the plane.
Under these very poor observability constraints, classical out-
Consider for systems (1), (4), the following ‘‘minimum informa-
put stabilization theorems cannot be applied. Nevertheless, we tion output’’, i.e. the square distance to the origin:
are able to prove the following statement.
ρ 2 = x2 + y2 = x̃2 + ỹ2 .
Theorem 1. For any smooth feedback stabilizing at the target For t ∈ [0, T ], if (x(t), y(t), θ (t)) is a trajectory of (1) for an
trajectory there exists a Luenberger-type observer for system (1), arbitrary input control u(t), it is clear that
for which the coupled closed loop state-observer system is asymp-
totically stable at the target, with an arbitrarily large basin of (x(t) cos θ0 − y(t) sin θ0 , x(t) sin θ0 + y(t) cos θ0 , θ (t) + θ0 )
attraction. is a trajectory of (1) with same input and same output.
This rotational symmetry implies that system (1) is not locally
Remark 1. Knowledge of the feedback law and the desired basin weakly observable in the sense of Hermann and Krener (1977):
of attraction may appear to be necessary for the specific choice of close to any (x, y, θ ), there is a continuum of points that are
the Luenberger-type observer. We show in the following sections indistinguishable to (x, y, θ ) by the observations, whatever the
that the size of the basin of attraction is only dependent on the input u(·).
size of a positive tuning parameter. Any bounded subdomain of Thanks to the system reduction, this is not the case for sys-
the plane can be covered by the basin of attraction of the coupled tem (4) (in fact it is the quotient of system (1) by the weak indis-
system if the tuning parameter is chosen large enough. Further- tinguishability relation from Hermann & Krener, 1977). However,
more, for any choice of the parameter, local stability is satisfied. system (4) is still not observable for all inputs: for the constant
Finally, and most importantly, no assumptions on the relationship control u ≡ 0, knowledge of the observation ρ 2 allows recon-
between the feedback law and the observability singularity at struction x̃ but ỹ can be reconstructed up to sign only. (This
u = 0 need to be made. Regularity and stabilizability are the only corresponds to the situation shown in Fig. 1.)
necessary assumptions.
Remark 2. Besides Hermann and Krener (1977), one can check
The paper is organized as follows: In Section 2, we detail the Sussmann (1974, 1976, 1978) for the general theory of quotient-
problem and a state-affine reduction. A precise statement of our ing through unobservability. The reader can refer to Brivadis et al.
result is given and discussed. In Section 3, we present the proof (2019) for a brief discussion of observability singularities in the
of this result. In Section 4, we show some simulations and give context of state-affine systems.
some closing remarks and perspectives.
The observation space of system (4) is finite-dimensional.
In the following, X ′ denotes the transpose of any matrix or
Following Fliess and Kupka (1983), it can be embedded into a
vector X .
state-affine system. Here, we simply set z = (z1 , z2 , z3 ), z1 =
x̃2 + ỹ2 , z2 = x̃, z3 = ỹ, and denoting the output by s, we get
2. State affine formulation the bilinear system with linear observation
ż = Az + uBz + b,
{
2.1. State affine embedding of the problem (5)
s = Cz , u ∈ [−umax , umax ]
The issue at hand is to stabilize system (1) at the origin. But 0 2 0 0 0 0 0
( ) ( ) ( )
what does it mean for a drone with constant velocity? In fact, it with A = 0 0 0 , B = 0 0 1 , b = 1 and
is required that it reaches a limit motion of turning around the 0) 0 0 0 −1 0 0
target achieving a circle of minimal radius r = 1/umax .
(
C = 1 0 0 .
2
A. Ajami, J.-P. Gauthier and L. Sacchelli Automatica 125 (2021) 109383

2.2. Main result 3. Proof of the main result

Since observation in the new state-affine system is linear, As we explained in the previous section, our observer is a
we introduce an observer ẑ with a linear correction term in its standard Luenberger-type observer for system (5), given in (6)
dynamics above.
Given the geometric constraints for (7), the coupled system
ẑ˙ = Aẑ + uBẑ + b − K (C ẑ − s). (6) satisfies in error-estimate coordinates:

⎨ ϵ̇ = (A + u(ẑ)B − KC )ϵ,
⎧
The choice of K and in general the design of the observer is open. ⎪
Here we consider for system (5) a ‘‘Luenberger-type’’ observer, ẑ˙2 = ẑ3 u(ẑ) + 1 − 2C ϵ, (8)
that is, with constant correction term K . For our purposes, we ⎪
⎩˙
make the following choice for K : for some arbitrary α > 0, ẑ3 = −ẑ2 u(ẑ),

K′ = α 0 . where ϵ is the estimation error, ϵ = ẑ − z, and u(ẑ) = u(ẑ2 , ẑ3 ) is

( )
2
a stabilizing feedback law for the system
A smooth stabilizing feedback at the target for (4) is a smooth {
map u : R2 → [−umax , umax ] such that the vector field (u(x, y)y + ẑ˙2 = ẑ3 u(ẑ) + 1,
(9)
1)∂x − u(x, y)x∂y admits a globally asymptotically stable equilib- ẑ˙3 = −ẑ2 u(ẑ).
rium at (0, −r).
Stabilization of (1) can then be achieved by proving semi- The question, in the remaining of this paper, is the stability of
global stability of the coupled closed-loop system this system at P ∗ ∈ R5 , where P ∗ is the point of coordinates {ϵ =
{ 0, ẑ2 = 0, ẑ3 = −r } corresponding to the control u = umax = 1/r.
ẑ˙ = Aẑ + u(ẑ) Bẑ + b − KC (ẑ − z), There are three steps to the proof of the (semi) global asymp-
(7) totic stability of this coupled system (8): proof of local asymptotic
ż = Az + u(ẑ) Bz + b,
stability, proof that bounded trajectories go to the target, proof
in which the state z is not arbitrary, but living inside the (invari- that all trajectories starting in a given compact set are bounded.
ant) manifold Z = {z | z1 = z22 + z32 }, and u(ẑ) = u(ẑ2 , ẑ3 ).
In particular, solutions to (7) evolve in R3 × Z , and the target 3.1. Local asymptotic stability
corresponds to the equilibrium (z ∗ , z ∗ ) with z ∗ = (r 2 , −r , 0).
We follow a classical scheme of proof. At the target point {ϵ =
Theorem 2. For any smooth stabilizing feedback at the target for 0, ẑ2 = 0, ẑ3 = −r } the linearized system is lower triangular.
system (4), for any bounded set B in R3 × Z , there exists α0 such Since KC = A′ + α e11 , the linearization relative to the ϵ -part
that for all α > α0 , system (7) is asymptotically stable at (z ∗ , z ∗ ) of the system is given by ϵ̇ = (A − A′ + umax B − α e11 )ϵ where e11
with basin of attraction containing B. denotes the 3 × 3 matrix with coefficient in position (1, 1) set
to 1, and others to 0. Then ∥ϵ∥2 is a Lyapunov function for the
2.3. Discussion of the main result sub-system, as
1 d
Before moving on to the proof of Theorem 2, let us discuss
∥ϵ∥2 = ϵ ′ Aϵ − ϵ ′ KC ϵ = −αϵ12 . (10)
2 dt
some implications of this result.
Since the pair (C , (A − A′ + umax B − α e11 )) is observable, this
First of all, this theorem answers the initial problem of stabi-
implies that 0 is an asymptotically stable equilibrium by LaSalle’s
lizing this kinematic drone model with distance information. It is
theorem. In particular, this implies that the eigenvalues relative
well understood (see, for instance, Mazenc, Praly, & Dayawansa, to the ϵ -part all have negative real part.
1994) that global feedback stabilization coupled with strong ob- Notice that the ẑ-diagonal block of the linearization of sys-
servability does not in general imply the possibility of global tem (8) (i.e. forgetting about ϵ ) coincides with the linearization of
dynamical output feedback stabilization. Only semi-global stabi- the system (9). Since u is a feedback law stabilizing (9) at (0, −r),
lization should be expected. its linearized can only have eigenvalues of non-positive real part.
The model is simple enough that one could design alterna- Furthermore, the asymptotic stability of (9) implies the exis-
tive approaches to this same question quite effectively. How- tence of a center manifold for (9) at (0, −r) that we denote C
ever, we are interested in the theoretical challenge of building (possibly empty if both eigenvalues of the linearized system have
a closed loop system for dynamic output stabilization under strictly negative real part). If C is nonempty, C is an invariant
these very constraining observability conditions. Here, we were manifold inside the invariant manifold {ϵ = 0} for the coupled
able to solve this problem without touching on the question system (8). Since all other eigenvalues have strictly negative real
of feedback design, except for regularity assumptions. Smooth part, C is then also a (stable) center manifold for the full coupled
stabilizing feedback laws were exhibited in Maillot et al. (2014). system (8). Hence we conclude at the asymptotic stability of the
The choice of the observer is free. Kalman-like observers are system at P ∗ .
well adapted to time-dependent bilinear systems (Anderson &
Moore, 2012; Bucy & Joseph, 2005; Trélat, 2005), and the speed 3.2. Bounded trajectories converge to the target
of convergence can be tuned, which is useful for proving a sep-
aration principle. However strong observability assumptions are First, along a bounded trajectory (ϵ (t), ẑ(t)) of system (8),
required to efficiently evaluate this convergence (see, for in- C ϵ (t) tends to zero. Indeed (10) holds. Therefore, C ϵ (t) = ϵ1 (t) is
stance, Gauthier & Kupka, 2001, Chapter 6, Section 2). Proving a L2 function over R+. Moreover, C ϵ (t) has bounded derivative
any general result of dynamic output stabilization for bilinear sys- since ϵ˙1 = −αϵ1 + 2ϵ2 and we are considering a bounded
tems that are not strongly observable remains an open problem. trajectory. A L2 function with bounded derivative tends to zero.
Alternative strategies built on fast observers such as sliding mode Looking at the ẑ-equation in (8), we see that in the ω-limit
are also of interest but face the same difficulties. Interestingly, a set Ω of the trajectory (ϵ (t), ẑ2 (t), ẑ3 (t)), the (ẑ2 , ẑ3 ) part of the
Luenberger-type observer is enough here. This points toward the system follows (9) again. This is the equation of the feedback
possible introduction of linear matrix inequalities approaches. system, which is globally asymptotically stable by assumption.
3
A. Ajami, J.-P. Gauthier and L. Sacchelli Automatica 125 (2021) 109383

Hence, by the general fact of invariance and closure of the ω-limit

set, Ω contains at least one trajectory such that ẑ2 ≡ 0, ẑ3 ≡ r.
Now, plugging u(ẑ) = −umax in the equation of ϵ , we see that
C ϵ ≡ 0 (preserved along trajectories in Ω ) can only be achieved
if ϵ ≡ 0 by observability of the system for −umax . Thus (0, (0, r))
belongs to Ω and, therefore, the trajectory (ϵ (t), ẑ(t)) enters in
finite time in the basin of attraction of P ∗ .

3.3. All semi-trajectories are bounded

As shown in Section 3.2, we have (10). This implies that ϵ is

bounded and C ϵ (t) = ϵ1 (t) tends to zero. But, ϵ̇1 = −αϵ1 + 2ϵ2 .
Hence
∫ t
ϵ1 (t) = e −α t
ϵ1 (0) + 2 e−α (t −s) ϵ2 (s)ds
0

and
2∥ϵ (0)∥
|ϵ1 (t)| ≤ e−αt |ϵ1 (0)| + (1 − e−α t ).
α
Therefore, we have the following.

Lemma 3. For all ψ > 0, τ > 0 and η > 0 there exists

ατ > 0 such that any semi-trajectory (ϵ, ẑ) of (8) with α > ατ
and ∥ϵ (0)∥ < ψ has |ϵ1 (t)| < η for all t > τ .
Let V (x̃, ỹ) be a strict proper Lyapunov function for the feed-
back system (4). Such a Lyapunov function can be obtained by
applying inverse Lyapunov’s theorems (see, for instance, Massera,
1949, 1956). Fig. 2. Simulations of the output feedback stabilization strategy, in both original
Let K be any compact subset of R5 (the state space of (8)) and and moving frame coordinates. The left column corresponds to α = 30 and the
let Π : R5 → R2 be the projection on the two last components right column to α = 0.5. The plain curve corresponds to the state of the system
(ẑ2 , ẑ3 ) (that are the estimates of (x̃, ỹ) in (4)). Let k ∈ N be a large while the dashed curve corresponds to the observer.

enough integer such that Π (K ) ⊂ Dk , where we denote the level

sets of V by
which we excluded. Therefore, (ẑ2 , ẑ3 ) remains in Dk+1 for ever.
Dδ = {(x̃, ỹ) | V (x̃, ỹ) ≤ δ}. We already know that ϵ (t) is bounded. Hence, the full trajectory
Let ψ > 0 be such that K ⊂ [−ψ, ψ]3 × Dk . Notice that the is bounded.
vector field (u(x, y)y + 1 − 2ϵ1 )∂x − u(x, y)x∂y , giving the velocity This ends the proof of Theorem 2.
of (ẑ2 , ẑ3 ) in (8), is uniformly bounded with respect to |ϵ1 | ≤ ψ
4. Conclusion and perspectives
on Dk+1 . Denoting this upper bound with R > 0, we set
1 We show two simulations in which we used the smooth sta-
τ= dist(Dk , Dk+1 ) > 0.
bilizing feedback control law from Maillot et al. (2014, Theorem
R+1
2.2) relative to system (4). The first simulation is with large α
Then the velocity of (ẑ2 , ẑ3 ) cannot allow to travel the distance
(α = 30), the second with small α (α = 0.5), see Fig. 2. Both
dist(Dk , Dk+1 ) in time τ , so that if ζ is a semi-trajectory of (8)
are taken with initial conditions (x0 , y0 , θ0 ) = (3, 5, π /4) and
starting in K , then Π (ζ (t)) remains in the interior of Dk+1 for all ẑ0 = (−7, 5, 4). On the top line of the figure are represented the
t ∈ [0, τ ]. This fact is independent on the choice of α since the trajectories of the drone (1) in the (x, y)-plane. On the bottom are
bound R is uniform and ϵ1 is decreasing. represented the trajectories of the reduced system (4) together
Denoting f (x̃, ỹ) = (u(x̃, ỹ)ỹ + 1)∂x̃ − u(x̃, ỹ)x̃∂ỹ and Lf the with the corresponding observer state estimate. The strong inob-
Lie-derivative with respect to the vector field f , let servability value u = 0 is actually crossed, at inflexion points of
⏐Lf V ⏐ > 0. the trajectory.
⏐ ⏐
m= inf
Dk+1 \Dk To finish, we would like to point out the challenge presented
As a consequence of Lemma 3, we can choose α > 0 such that by practical output stabilization of control systems in the un-
any semi-trajectory ζ of (8) with ζ (0) ∈ K , satisfies observable case. Here we have treated a case where the target
point is an observable point, and showed that after state-affine
⏐ ∂V ⏐
⏐ ⏐
⏐ < m, embedding, a Luenberger-type observer was converging reliably
2|ϵ1 (t)| sup ⏐⏐ ∀t > τ .
Dk+1 \Dk ∂ x̃ enough to allow a separation principle to take place. We refer
⏐
to Lagache et al. (2017) for a case where the target control makes
This implies that if (ẑ2 , ẑ3 ) ∈ Dk+1 \ Dk at t > τ , the system unobservable. It is particularly challenging to consider,
d ∂V for unobservable bilinear or bilinearizable systems, the coupling
V (ẑ2 , ẑ3 ) = Lf V (ẑ2 , ẑ3 ) + 2ϵ1 (t)
(ẑ2 , ẑ3 ) < 0. of a stabilizing feedback law with a Kalman-type observer. Such
dt ∂ x̃ a result would be very important.
However, if there exists t > 0 such that Π (ζ (t)) ∈ / Dk+1 , this For what regards this particular kinematic drone model, since
implies the existence of a time t ′ ∈ (τ , t) such that Π (ζ (t ′ )) ∈ straight trajectories can be time optimal, we consider the
Dk+1 and coupling of the observability problem with the minimum time
d optimal synthesis a particularly interesting question worthy of
V (ẑ2 , ẑ3 )|t =t ′ > 0, further research.
dt
4
A. Ajami, J.-P. Gauthier and L. Sacchelli Automatica 125 (2021) 109383

Acknowledgment Mazenc, F., Praly, L., & Dayawansa, W. (1994). Global stabilization by output
feedback: examples and counterexamples. Systems & Control Letters, 23(2),
119–125.
The authors would like to thank U. Serres for the many fruitful
Owen, M., Beard, R. W., & McLain, T. W. (2015). Implementing Dubins’ air-
discussions that led to the present paper. plane paths on fixed-wing UAVs. In Handbook of unmanned aerial vehicles
(pp. 1677–1701). Springer.
References Shim, H., & Teel, A. (2003). Asymptotic controllability and observability im-
ply semiglobal practical asymptotic stabilizability by sampled-data output
Ajami, A., Balmat, J., Gauthier, J.-P., & Maillot, T. (2013). Path planning and feedback. Automatica, 39(3), 441–454.
Ground Control Station simulator for UAV. In 2013 IEEE aerospace conference Soueres, P., & Laumond, J.-P. (1996). Shortest paths synthesis for a car-like robot.
(pp. 1–13). IEEE. IEEE Transactions on Automatic Control, 41(5), 672–688.
Anderson, B. D., & Moore, J. B. (2012). Optimal filtering. Courier Corporation. Sussmann, H. J. (1974). On quotients of manifolds: A generalization of the closed
Balluchi, A., Bicchi, A., Piccoli, B., & Soueres, P. (2000). Stability and robustness subgroup theorem. American Mathematical Society. Bulletin, 80(3), 573–576.
of optimal synthesis for route tracking by dubins’ vehicles. In Proceedings of Sussmann, H. J. (1976). Existence and uniqueness of minimal realizations of
the 39th IEEE conference on decision and control cat. no. 00CH37187 (vol. 1) nonlinear systems. Mathematical Systems Theory, 10(1), 263–284.
(pp. 581–586). IEEE. Sussmann, H. J. (1978). Single-input observability of continuous-time systems.
Boscain, U., & Piccoli, B. (2003). Optimal syntheses for control systems on 2-D Mathematical Systems Theory, 12(1), 371–393.
manifolds (vol. 43). Springer Science & Business Media. Teel, A., & Praly, L. (1994). Global stabilizability and observability imply semi-
Brivadis, L., Gauthier, J.-P., Sacchelli, L., & Serres, U. (2019). Avoiding observability global stabilizability by output feedback. Systems & Control Letters, 22(5),
singularities in output feedback bilinear systems. arXiv:1907.02311. 313–325.
Bucy, R. S., & Joseph, P. D. (2005). Filtering for stochastic processes with applications Teel, A., & Praly, L. (1995). Tools for semiglobal stabilization by partial state and
to guidance (vol. 326). American Mathematical Soc. output feedback. SIAM Journal on Control and Optimization, 33(5), 1443–1488.
Chitsaz, H., & LaValle, S. M. (2007). Time-optimal paths for a Dubins airplane. Trélat, E. (2005). Contrôle optimal: théorie & applications. Vuibert Paris.
In 2007 46th IEEE conference on decision and control (pp. 2379–2384). IEEE.
Combes, P., Malrait, F., Martin, P., Rouchon, P., et al. (2016). Adding virtual
measurements by signal injection. In 2016 American control conference Alain Ajami is an assistant professor at Saint Joseph
(pp. 999–1005). IEEE. University of Beirut, faculty of engineering (ESIB),
Coron, J.-M. (1994). On the stabilization of controllable and observable systems Lebanon. He completed his Master’s degree in Engi-
by an output feedback law. Mathematics of Control, Signals, and Systems, 7(3), neering (Electrical and control systems) in 2006. He got
187–216. Ph.D. degree in 2013 from University of Toulon, France.
Dubins, L. (1957). On curves of minimal length with a constraint on average His research interests include control theory, inverse
curvature, and with prescribed initial and terminal positions and tangents. optimal control problem and paths planning.
American Journal of Mathematics, 79(3), 497.
Esfandiari, F., & Khalil, H. K. (1992). Output feedback stabilization of fully
linearizable systems. International Journal of Control, 56(5), 1007–1037.
Fliess, M., & Kupka, I. (1983). A finiteness criterion for nonlinear input–
output differential systems. SIAM Journal on Control and Optimization, 21(5), Jean-Paul Gauthier is a professor at UTLN, University
721–728. of Toulon, France, at laboratory LIS, UMR CNRS 7020.
Gauthier, J.-P., & Kupka, I. (2001). Deterministic observation theory and He graduated in maths and computer sciences from
applications. Cambridge University Press. Institut National Polytechnique de Grenoble in 1975.
Hermann, R., & Krener, A. (1977). Nonlinear controllability and observability. He got Ph.D. degree in 1978, and state doctorate degree
IEEE Transactions on Automatic Control, 22(5), 728–740. in 1982. He joined CNRS (National Centre of Scientific
Jouan, P., & Gauthier, J.-P. (1996). Finite singularities of nonlinear systems. Research) in 1978. He was awarded several medals for
Output stabilization, observability, and observers. Journal of Dynamical and his research work including Medal of Institut Univer-
Control systems, 2(2), 255–288. sitaire de France, and Medal of Institut National des
Khalil, H., & Esfandiari, F. (1993). Semiglobal stabilization of a class of nonlinear Sciences Appliquées. He is also honorary member of
systems using output feedback. IEEE Transactions on Automatic Control, 38(9), Institut Universitaire de France. His research interests
1412–1415. lie in non-linear control theory.
Lagache, M.-A., Serres, U., & Andrieu, V. (2015). Time minimum synthesis for a
kinematic drone model. In 2015 54th IEEE conference on decision and control
(pp. 4067–4072). IEEE. Ludovic Sacchelli graduated from Ecole Normale
Lagache, M.-A., Serres, U., & Gauthier, J.-P. (2017). Exact output stabilization Supérieure de Cachan and received his master’s degree
at unobservable points: Analysis via an example. In 2017 IEEE 56th Annual in analysis of PDEs from Paris-Sud University in 2015.
conference on decision and control (pp. 6744–6749). IEEE. He received his Ph.D. in applied mathematics from
Maillot, T., Boscain, U., Gauthier, J.-P., & Serres, U. (2014). Lyapunov And Ecole Polytechnique in 2018. He spent the following
minimum-time path planning for drones. Journal of Dynamical and Control year as a postdoctoral fellow in the Electrical Engineer-
Systems, 21(1), 47–80. ing Department of University of Toulon, then became a
Massera, J. (1949). On Liapounoff’s conditions of stability. Annals of Mathematics, visiting assistant professor in the Mathematics Depart-
50(3), 705. ment of Lehigh University, in Bethlehem, Pennsylvania.
Massera, J. L. (1956). Contributions to stability theory. Annals of Mathematics, Since October 2020, Sacchelli is a postdoctoral re-
64(1), 182. searcher at LAGEPP, Lyon University. His research
interests lie in non-linear control theory, observation and sub-Riemannian
geometry.