www.ietdl.

org

Published in IET Control Theory and Applications

Received on 18th February 2008
Revised on 25th February 2009
doi: 10.1049/iet-cta.2008.0057

ISSN 1751-8644

Coupled linear parameter varying and

flatness-based approach for space re-entry
vehicles guidance
M. Zerar1 F. Cazaurang2 A. Zolghadri2
1
CreSTIC-URCA lab./Reims University, Moulin de la Housse, 51687 Reims, France
2
IMS lab./Bordeaux University, 351 cours de la libération, 33405 Talence, France
E-mail: [email protected]

Abstract: An linear parameter varying guidance method for the hypersonic phase of a space re-entry vehicle
is presented. The suggested guidance scheme, relying on flatness approach, is applied to the non-linear
model of the European Atmospheric Re-entry Demonstrator. It is shown that the overall guidance scheme
achieves robust stability and performance, even in the presence of entry point kinematics dispersions.
The design problem is formulated and solved using a finite set of linear matrix inequalities.
Finally, Monte Carlo simulation results are presented to demonstrate the effectiveness of the suggested
approach.

Nomenclature
LPV linear parameter varying c azimuth angle
LTV linear time varying X Downrange position
LMI linear matrix inequality a angle-of-attack
LFP linear fractional presentation M mach number
LFT linear fractional transformation mv vehicle mass
LQP linear quadratic problem g gravity acceleration
MIMO multiple input multiple output mT Earth gravitational constant
ARD atmospheric re-entry demonstrator m bank angle
TPS thermal protection systems L (a , M ) lift force
RCS reaction control systems D (a , M ) drag force
GNC guidance navigation and control L/D lift-to-drag ratio (gliding coefficient)
AoA angle-of-attack CL lift coefficient
R radial distance CD drag coefficient
Re mean earth radius S vehicle reference area
u longitude r atmospheric density
f latitude r0 atmospheric density at sea level
V relative velocity href scale height of atmospheric density model
g flight path angle z (k)(t) kth time derivative of z(t)

IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092 1081
doi: 10.1049/iet-cta.2008.0057 & The Institution of Engineering and Technology 2009
www.ietdl.org

1 Introduction and problem necessary for LPV guidance control are briefly reviewed in
Section 2. In Section 3, the ARD is presented and the
setting flatness of the ARD longitudinal dynamics is
Atmospheric re-entry mission presents challenges in several demonstrated. An LPV model is then build and used for
domains of engineering and science, namely in aero- designing an LPV controller to ensure robust trajectory
thermodynamics, thermal protection systems, reaction tracking. Finally, few simulation results are presented and
control systems, guidance, navigation and control system discussed in the last section.
design [1]. During the hypersonic re-entry, the vehicle is
subject to both mechanical constraints (as heat flux load 2 Trajectory tracking for flat
factor, dynamic pressure) and path constraint in order to
reach a predefined targeted point typically defined at Mach 2 systems
gate while avoiding any trajectory skipping on upper 2.1 Brief review of differential flatness
atmospheric layers. Typically a guidance unit consists of a
path planner designed to meet the aforementioned Differential flatness theory was first introduced by Fliess et al.
constraints, and a tracking controller tuned to maintain the [13] in a differential algebraic context. Flatness approach is
spacecraft on this trajectory in spite of plant perturbations focused on the determination of a set of particular outputs
and external disturbances. named flatness outputs such that NDI based on these
outputs does not possess any non-linear unobservable
subsystem. Therefore the stability analysis of the set NDI
Since the early 60s, several guidance schemes have been
block and non-linear model is reduced to check the root
proposed in the open literature to solve the hypersonic re-
locus of the linearised model [14].
entry issue. Among these guidance laws, the most referred
one to is the US’ Apollo/Space Shuttle entry guidance
A non-linear system is flat if there exists a set of variables
concept presented in [2, 3]. This last one proceeds by an
(equal in number to the number of inputs) such that all states
uncoupling of the in-plane motion, for the downrange
and inputs can be expressed in terms of those outputs and a
control and the out-of-plane motion, for the cross range
finite number of their time derivatives. Let us consider a non-
control, the management of the aerodynamic forces being
linear deterministic system of the general form
obtained by the vertical L/D modulation.
x_ ¼ f ðx, uÞ (1)
Several published works are available in the open literature
which use non-linear dynamic inversion (NDI) techniques to where x is the n-dimensional state vector, u is the
solve the guidance problem (see for instance [4, 5]). More m-dimensional input vector, f (.) is a non-linear function,
recently, in the frame of optimal control applied to and it is assumed that m  n.
atmospheric reentry, Neckel et al. [6] used non-linear
inversion to map the optimisation problem into a lower Definition 1: The non-linear system (1) is differentially flat
dimensional space, following ideas from [7] (see also [8, if there exists a set of m variables z that are differentially
9]). The coupled non-linear guidance equations are not flat independent, called flat outputs, of the form
but partly invertible.
 T  
z ¼ z1 , . . . , zm ¼ F x, u, u_ , . . . , u(a) (2)
Following the ideas proposed in [10], where a guidance
methodology is proposed to derive an LQP-like trajectory
tracking law, this paper proposes an original approach to such that
design a guidance scheme throughout the hypersonic 8  
phase, using both flatness and LPV tools. As will be < x ¼ Cx z, z_ , . . . , z(r1)
showed later, Monte Carlo simulations suggest that the   (3)
: u ¼ C z, z_ , . . . , z(r)
design method achieves specified level of robustness and u
performances. Considering an uncoupled model as in [3],
the longitudinal non-linear model is flat and the behaviour where Cx and Cu are smooth functions, z(k) i denoting the kth
of this model along the trajectory is computed into a order time derivative of the ith components of z(t), and the
fictitious LPV model. To obtain this LPV model, the multi-index r ¼ (r1 , . . . , rm ) containing the characteristic
dynamics of the trajectory tracking error is computed by numbers associated with the flat outputs such that
using coordinate transformation of the flat outputs, as well
(  )
as a first-order Taylor series expansion around the reference @z(k)
  i
trajectories. Then, techniques for LPV control design [11, ri ¼ min k [ N , 9j [ f1, . . . , mg =0 ,
 @uj
12] to ensure the specified control objectives.
i ¼ 1, . . . , m (4)
The remainder of this paper is organised as follows. The Pm
flatness theory and its application to LPV modelling with i¼1 ri ¼ n

1082 IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-cta.2008.0057
 www.ietdl.org

Thus, flat outputs z(t) are a minimal parametrisation of all proposition provides a tool to characterise the gap model
trajectories x(t), u(t) satisfying (1). which is used to build the LPV model.

Remark 1: The use of this definition to check candidate flat Proposition 3: Let us denote the gap vector e11 , . . . ,
outputs constitutes an informal way to find flat outputs. e1r1 , . . . , em1 , . . . , emrm Þ corresponding to the state of the
Some recent advances on formal characterisation of flat non-linear error model, around the desired trajectories
outputs based on the necessary and sufficient conditions (xd , ud ), induced by initial variations on the state and defined by
proposed in [15] can be found in [16, 17].
( j1) ( j1)
8i ¼ 1, . . . , m, ei, j ¼ zi  zid , j ¼ 1, . . . , ri
Remark 2: Flatness can also be defined as Lie – Bäcklund (5)
equivalence [18] between a non-linear system and a trivial (j)
system. A property of differential flatness is that z are the where zid is the jth time derivative of the desired flat output zid
(j)
states of the trivial system. The m-components of z and zi is the jth time derivative of the actual true flat output zi .
are differentially independent. Furthermore a Z-space
 dimension nz can be set with the coordinates Z ¼
of Then, the non-linear state model of the gap is given by
z, z_ , . . . , z(r) where r [ N. 8
< ei,_ j ¼ ei, jþ1 , j ¼ 1, . . . , ri  1
>
8i ¼ 1, . . . , m ei,_r ¼ ai,r (zd þ e, u d þ du)  z_ id (ri ) (6)
2.2 Trajectory tracking >
: dz ¼ e i i
i i,1
The trajectory tracking for flat systems can be viewed as a
three-step process in which reference trajectories, generated where ai,ri is a smooth function defined by
by a path planner (not necessary based on flatness), allow
for recovering flat outputs reference profiles. Then without r
ai,ri ¼ Lf i(x,u) zi (7)
having to solve differential equations, the inputs reference
profiles can be easily determined with a low computational r
Lf i(x,u) zi corresponding to the rith Lie derivative of zi along
burden by using an NDI block (i.e. the feed-forward loop).
the vector field f (x, u) and u ¼ (u, u_ , . . . , u(a) , . . . ).
Then, to circumvent exogenous disturbances and model
uncertainties, a feedback loop is needed as shown in Fig. 1.
Proof: Assume that the desired trajectory of the non-linear
Here, the control action u consists of a corrective input term flat system, whose output variables must be tracked, is
du based on flat output gap dz and the reference feed-forward given by zd . By using (3), the feasible desired feed-forward
control law ud generated by the dynamic inversion block. state xd and the associated nominal control input trajectory
ud can be computed as follows
Looking at the above two degree of freedom scheme first  
(r ) (r )
introduced in [19], it appears that to design the feedback xd (t) ¼ Cx z1d , . . . , z1 1 , . . . , zmd , . . . , z1 m
controller, an LPV model is needed between the flat output   (8)
(r þ1) (r þ1)
errors dz and the control inputs du. This model describes the ud (t) ¼ Cu z1d , . . . , z1 1 , . . . , zmd , . . . , z1 m
error due to the difference between the actual dynamics of the
disturbed system and the desired dynamics of the nominal
Now, consider the flat output transformation to a new set of
system corresponding to the reference trajectory (xd , ud ).
coordinates j given by
In order to determine this model, let us consider variations
of the states dx around the nominal states. The following 8 i ¼ 1, . . . , m, ji, j ¼ z(i j1) , j ¼ 1, . . . , ri (9)

Figure 1 Two-degree scheme for path planning and path tracking

IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092 1083
doi: 10.1049/iet-cta.2008.0057 & The Institution of Engineering and Technology 2009
www.ietdl.org

By using the previous coordinate transformation, the true canonical form

non-linear flat system (1) may be rewritten into the 0 1
0
following non-linear equivalent model (see, e.g. [20, 21] for de_1 1 A1,1 ... A1,m 0 de1 1
more details) B .. C B . .. .. C B. C
@. A ¼ B @ .. . . C A@ .. A
( de_m Am,1 ... Am,m dem
ji,_ j ¼ ji, jþ1 , j ¼ 1, . . . , ri  1
0 10
du1 1
8 i ¼ 1, . . . , m, B1,1 . . . B1,m
ji,_ri ¼ ai,ri (j, u )
B . .. .. C B. C
(10) þB
@ .. . . CA@ .. A (16)
  Bm,1 ... Bm,m dum
such that j1,1 , . . . , jm,rm denotes the new state of the non- with
linear system (10) and ai,ri is a smooth function defined by
0 1
0 di, j 0 ... 0
r
ai,ri (j, u ) ¼ Lf i (x, u)zi (x, u ) (11) B . .. .. .. .. C
B . C
B . C
Ai, j ¼B . . . .
C
B 0 ... ... 0 di, j C
The application of the desired feed-forward control law (8) to @ A
the non-linear equivalent model (10) allows us to obtain a ai, j,1 ... ... ... ai, j,ri
nominal model of the actual true flat system (1) which can  
@ai,ri jd þ e, u þ du 
be written as follows with ai, j,k ¼  (17)
@ej,k 
jd ,ud
(
j_ (i, j)d ¼ j(i, jþ1)d , j ¼ 1, . . . , ri  1
8 i ¼ 1, . . . , m, where dij represents the Kronecker’s symbol and
j_ (i,r )d ¼ ai,r (jd , u d )
i i
0 1
(12) 0 ... 0
B .. .. .. C
Bi, j ¼@ . . . A
with jd and ud being, respectively, the desired new state j and bi, j . . . bi, j
the nominal control input u. By considering the error  
( j) ( j)
ei, j ¼ ji, j  j(i, j)d ¼ zi  zid due to the difference @ai,ri jd þ e, u þ du 
with bi, j ¼  (18)
between the effective dynamics of the actual system (10) @duj 
jd ,ud
and the desired dynamic of the nominal system (12), a
non-linear error model is obtained as follows Both the reference flat outputs zd and the associated reference
0 1 0 1 control inputs ud trajectories are known in advance, so the
e_11 j_ 11  j_ (11)d elements ai, j,k and bi, j can be expressed exclusively as a
B. C B. C function of time. Consequently, the LTV plant (16) may
B .. C ¼ B .. C (13)
@ A @ A be viewed as an LPV model described by the following
e_mrm j_ mrm  j_ (mrm )d state-space equations

Now, let us consider the control input u needed to maintain d_e ¼ A(ai, j,k , bi, j )de þ B(ai, j,k , bi, j )du (19)
the system on a desired output profile
where de is the new state vector, du is the new control input
vector and (ai, j,k , bi, j ) is the measurable (time varying and
u ¼ ud þ du (14)
real) parameter vector.

Then, by combining expressions (10), (12) and (14), the non- There is more than one synthesis technique for designing an
linear error model (13) can be rewritten as follows LPV controller once the LPV model is known. The technique
used in this paper relies on the H1 control synthesis technique
8
> e_ ¼ ei, jþ1 , j ¼ 1,    , ri  1 and uses a linear fractional representation [22] of the LPV
>
< i, j system (see Fig. 2) to compute the controller.
8 i ¼ 1, . . . , m e_(i,ri )d ¼ a (j þ e, u þ du)  j_
i,ri d d (iri )d
>
>
: ¼ ai,ri (jd þ e, u d þ du) 
(r )
z_ id i
(15)

Based on the above result, an LPV model can be computed

by performing a first-order linearisation, along the reference
trajectory. This model can be rewritten in the following Figure 2 Upper LFT representation

1084 IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-cta.2008.0057
 www.ietdl.org

The design problem is formulated and solved by means of V

LMIs (see [11]). f_ ¼ cos g sin x
R
D(a, M)
In the following section, the proposed methodology is V_ ¼   g sin g
applied to the hypersonic re-entry guidance of a model of mv
  (20)
ARD. L(a, M, ) V g
g_ ¼ cos m þ  cos g
mv V R V
3 Application to a low L/D L(a, M) sin m V
c_ ¼ þ cos g sin c tan u
entry vehicle mv V cos g R
3.1 Description of the ARD
where R is the radial distance from the Earth centre to the
The ARD flown successfully in the late 1998 during the vehicle, u is the longitude, f is the latitude, V is the
experimental flight program developed by the European velocity, mv is the mass of the vehicle, g is the flight path
Space Agency. The ARD program’s main objective was the angle and c is the azimuth angle measured from the north
in-flight validation of key technical expertise and in a clockwise direction. Also, a simplified gravitational
technologies required to master orbital return, namely, the field such that g ¼ mT =R2 has been considered where mT
accuracy of the aerodynamics models, performance of the is the Earth gravitational constant. L(a, M) and D(a, M)
thermal protection system, performance and robustness of represent the lift and drag forces and are given, respectively,
GNC flight software. The ARD mission cycle consisted of by
four flight phases: launch phase, ballistic phase, hypersonic
phase (from 120 (km) high down to Mach 2 crossing) and
1
final descent. L(a, M) ¼ rV 2 SCL (a, M)
2
(21)
The ARD looks like a 70% scale Apollo capsule, with a 1
D(a, M) ¼ rV 2 SCD (a, M)
sphero-conical shape as shown in Fig. 3. Its maximum 2
diameter is 2.80 (m) for a 2.04 (m) length and the overall
mass is 2800 (kg) (see for instance [23]). where a is the AoA and M ¼ V =Vs is the Mach number (Vs
represents the sound velocity). In addition, S is the vehicle
3.2 Non-linear vehicle dynamics reference area, CL and CD are the lift and the drag
coefficients. In order to describe the properties of the
The non-linear dynamic equations of the ARD are borrowed
atmosphere, a simple model is used which is an
from Vinh et al. [24], Betts [25] and also used in [6]. In order
approximation of US standard atmosphere 1976 [26]. In
to ease the guidance design procedure, the flight dynamics of
this model, r is the R-dependent atmospheric density such
the re-entry vehicle is approximated by the rotational and
that
translational motion of a rigid body with respect to a
stationary spherical Earth. The motion of the ARD is  
defined by the following set of simplified equations R  Re
r ¼ r0 exp  (22)
obtained by neglecting Coriolis and Euler’s forces href

R_ ¼ V sin g
where r0 is the atmospheric density at sea level, Re is the
V cos g sin x mean equatorial radius and href is the reference altitude
u_ ¼
R cos u tuned for upper atmospheric layers.

The lift-to-drag (L/D) ratio is given by the following

expression

L CL (a, M)
¼ (23)
D CD (a, M)

During the entry, mv is assumed to be constant. Besides, a is

maintained by the control system near the natural trim AoA
of the capsule, which, for a given Mach number, is fixed by its
shape and its layout (centre of gravity location). Therefore the
Figure 3 Atmospheric re-entry demonstrator lift and drag coefficients, as well as the lift-to-drag ratio, can
a External view be assumed to be functions of M only. They are depicted in
b Size Figs. 4 and 5.

IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092 1085
doi: 10.1049/iet-cta.2008.0057 & The Institution of Engineering and Technology 2009
www.ietdl.org

Figure 4 Aerodynamic coefficients profiles

a Drag coefficient profile
b Lift coefficient profile

manifold criterion (introduced by Rouchon in [28]) and

showed that the coupled non-linear model, including
longitudinal and lateral dynamics, is not flat. Nevertheless,
based on the Harpold’s assumptions [3], it is possible to
consider longitudinal and lateral motions uncoupled and
then use the flatness property of the uncoupled model. The
induced conservativeness could be managed by adding an
additional constraint during control design.

3.3 Flatness property of the ARD

longitudinal model
The ARD longitudinal model is derived from the non-linear
equations (20) as follows. Under the assumption of non-
rotating and spherical Earth, let us introduce a new variable
X which describes the downrange position, that is, the
curvilinear coordinate referring to the geocentric Earth
Figure 5 Lift to Drag ratio frame along the ground track. Moreover, if the longitudinal
and lateral dynamics are considered to be uncoupled, the
longitudinal equations are described by the following set of
Since the ARD is essentially a scaled version of the Apollo differential equations
Command Module (ACM) borrowed from [27], Apollo
aerodynamics is used for this study with a scaled factor R_ ¼ V sin g
equal to 0.7.
X_ ¼ V cos g
It may be noticed that some vehicles are controlled by bank 2
1 rSV (24)
angle modulation only, whereas others use a combination of V_ ¼  CD (M)  g sin g
both AoA and bank angle. During the real ARD flight, 2 m
 
AOA modulation is not considered, excepted some limited 1 rSV 2 V g
g_ ¼ CL (M) cos m þ  cos g
predefined excursions when performing roll reversals to 2 m R V
overcome potential altitude offsets. In this paper, the cosine
of the bank angle cos m and the drag coefficient CD (M) The non-linear model (24) includes four states (R, X , V , g)
will be considered as control inputs. and two control inputs CD (M) and cos m. For this
longitudinal model, the candidate flat outputs are the radial
First, it is necessary to check the flatness property of the distance R and the downrange position X, denoted by z1
ARD non-linear model. In [6], the authors used the ruled and z2 , respectively, in the sequel. So, the system inputs

1086 IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-cta.2008.0057
 www.ietdl.org

(CD (M), cos m) and the system states (V, g) may be written as that is, the sign of the bank angle is reversed to stay
function of the candidate flat outputs z1 , z2 and their first- within a lateral re-entry corridor Dc defined with respect
and second-order time derivatives, as follows to V. The interested reader can refer to [3] for more
details.
R ¼ z1
X ¼ z2 The flatness property of the considered model can now be
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi used to generate the reference control inputs without having
V ¼ z21 þ z22 (25) to solve any differential
 equations. In order to compute the
reference states Vd , gd and the nominal control inputs
z_ 1 profiles CDd , cos md , the flat output reference trajectories
g ¼ arctan
z_ 2 Rd and Xd depicted in Fig. 6 are used. These trajectories
are generated by an industrial path planner in order to
Based on (24), the drag coefficient CD and the cosine of the lie within a predefined entry corridor as constrained by
bank angle cos m can be written as the equilibrium, heat flux, dynamic pressure and load
factor constraints [23]. Then, by using (26) and (27),
 
2mv V_  g sin g the desired trajectories are obtained as depicted in Figs. 7
CD ¼ and 8.
rV 2 S
    (26)
2mv V g In the next section, the robust trajectory tracking
cos m ¼ g_   cos g
rV (L=D)CD R V method proposed in this paper is applied in the presence
of plant perturbations and entry point kinematics
Moreover, V_ and g_ may be expressed as a function of flat dispersions.
outputs and a finite number of their time derivatives

z€1 z_1 þ z€2 z_2 3.4 LPV modelling and controller design
V_ ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
z21 þ z22 The goal of this section is to derive an equivalent
(27) parameterisation of the ARD longitudinal flat model as an
z€1 z_1  z€2 z_2
g_ ¼  2 ! LPV model. First, let us consider the following coordinate
z_ transformation j of the flat outputs
z_ 2 1 þ 2
2
z_1

T
T
j ¼ j11 , j12 , j21 , j21 ¼ z1 , z_ 1 , z1 , z_ 2
Thus, the control inputs can also be expressed as functions of
flat outputs and a finite number of their derivatives. ¼ ½R, V sin g, V cos gT (28)

Remark 4: In order to ensure lateral control authority, By using the previous diffeomorphism j, the non-linear flat
several roll reversals must be planned along the trajectory, model (24) can be rewritten under the following form

Figure 6 Reference trajectory profiles

a Radial distance Rd
b Downrange position Xd

IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092 1087
doi: 10.1049/iet-cta.2008.0057 & The Institution of Engineering and Technology 2009
www.ietdl.org

Figure 7 Reference state profiles

a Velocity profile Vd
b Flight path angle profile gd

Figure 8 Reference input profiles

a Drag profile CDd
b Bank angle profile (cos m)d

(see also [29]) Let us consider j11d , j12d , j21d , j21d as the desired profiles of
the state j11 , j12 , j21 , j21 . The dynamics of the gap model is
described as
j_ 11 ¼ j12
 e_11 ¼ e12
1
j_ 12 ¼  rSV 2 CD  g sin g sin g þ   
2m e_12 ¼ j_ 12  j_ 12d
   (30)
1 L V g
V  rSV 2 CD cos m þ  cos g cos g e_21 ¼ e22
2m D R v
e_22 ¼ j_ 22  j_ 22d
j_ 21 ¼ j22

_j ¼  1 rSV 2 C  g sin g cos g    The obtained model depending on the flat outputs is highly
22 D
2m non-linear. To obtain a linear model, a first-order
   linearisation of (30) around the desired trajectories jd , CDd
1 L V g
V  rSV 2 CD cos m þ  cos g sin g and cos md is performed. At this stage, it is useful to use a
2m D R v software that supports mathematical symbolical operations
(29) (for instance MAPLETM ). By using the notation

1088 IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-cta.2008.0057
 www.ietdl.org
0 1
introduced in Section 2, the LPV model can be described in A211 0 0 0 0 A411 0 0 0 0
state-space form as B 0 A211 0 0 0 0 A421 0 0 0C
B C
CD ¼ B C
@ 0 0 0 0 0 0 0 0 0 0A
0 1
e_11 0 1 0 0 A211 0 0 0 0 A441 0 0
B e_ C 0 1 0 0 0 0
B 12 C B A21 (t) A22 (t) 0 A24 (t) B21 (t) B22 (t) C  
B C B C 1 0 0 0
B e_21 C B 0 0 0 1 0 0 C and C2 ¼
B C B C 0 0 1 0
B e_ C ¼ B A41 (t) A42 (t) 0 A44 (t) B41 (t) B42 (t) C
B 22 C B C
B C B C
B C @ 1 0 0 0 0 0 A with Aij1 ¼ Aij max  Aij min =2, DDD ¼ D2D ¼ D22 ¼ 0, and
@ dz1 (t) A
0 0 1 0 0 0
dz2 (t) 0 1
0 1 0 0 0 B211 0 0 0 0 B411 0
e11 B0 0 0 0 0 0 0 0 0 0 C
B e C DD2 ¼ B
@0
C
B 12 C 0 0 0 0 0 0 0 0 0 A
B C 0 0 0 0 B221 0 0 0 0 B421
B e21 C
B C
B e C (31) (33)
B 22 C
B C
B C
@ du1 (t) A Based on the model computed by (31), (32) and (33), the
du2 (t) next stage is to design a guidance law in order to steer
the vehicle on a feasible trajectory that achieves the
specified target condition within the specified error.
where Aij (t) and Bij (t) are measured in real-time and These different objectives may be specified in terms of
correspond to the time varying parameters given in the weighting functions We (s) and Wu (s) on the sensitivity
Appendix. (S(s) transfer) and controls inputs (KS(s) transfer)
respectively (see Fig. 9). The weighting functions are
The above LPV model given by (16) can be rewritten as an tuned in order to remove the fast variations of the error
upper LFT such that dz(s) ¼ fu (G(s), D)du(s). See [30] for a signal dz(t), bound variations of the control signal du(t)
description of the method used. and also take into account some error model induced by
the uncoupled assumption.Within the predefined limits,
0 1 the retained profiles have been chosen based on a time-
A BD B2
B C domain analysis of Monte Carlo simulations in the
G(s) ¼ @ C DDD DD2 A and presence of initial kinetics dispersions for the considered
C2 D2D D22 hypersonic path.
0 1
dA21 (t)    0
B . Finally, using the control structure shown in Fig. 10, an
.. .. C
D(t) ¼ B
@ .. . . C A (32) eight-order gain-scheduled MIMO controller has been
synthesised by using the Matlab LMI Control Toolbox
0    dB42 (t)
[31]. The LPV controller K (D(t)) is given by the lower
LFT K-D(t) structure where D(t) matrix is the same matrix
   
   
with dAij (t) ¼ dBij (t) ¼ 1, and

0 1
0 1 0 0
BA C
B 210 A220 0 A240 C Aij max þ Aij min
A¼B C where Aij0 ¼
@ 0 0 0 1 A 2
A410 A420 0 A440

0 1
0 0 0 0 0 0 0 0 0 0
B1 1 1 1 1 0 0 0 0 0C
B C
BD ¼ B C and
@0 0 0 0 0 0 0 0 0 0A
0 0 0 0 0 1 1 1 1 1
 
0 B210 0 B410
B2 ¼
0 B220 0 B420 Figure 9 Weighting functions on S and KS transfers

IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092 1089
doi: 10.1049/iet-cta.2008.0057 & The Institution of Engineering and Technology 2009
www.ietdl.org

Figure 10 LPV controller design scheme

3.5 Guidance algorithm performances

assessment
In this section, Monte Carlo simulation results
corresponding to the newly designed hypersonic guidance
scheme are analysed. Initial re-entry point dispersions have
been considered to assess the efficiency of the guidance law.

More precisely, the initial radial distance value R(0) is

included into a 100 (m) dispersion circle around the
nominal value R(0) ¼ 6520 103 (m) and the downrange
is included in 50  103 (m) dispersion circle around the
nominal value X (0) ¼ 20  103 (m) (i.e. the uncertainties
taken into account are (e1(0) ¼ 100 (m), e2(0) ¼ 50  103
(m). The flat outputs trajectories associated to the Monte
Figure 11 LPV controller singular values Carlo simulation are shown in Fig. 12. The tracking error e
is depicted in Fig. 13. As it can be shown, the tracking
errors converge to zero asymptotically and the states profiles
used in the LPV model and defined in (32). Singular values trajectories track the desired trajectories with an error less
are depicted in Fig. 11 and corresponds approximately to a than 5% and a setling time of 120 (s). The goal of the
multivariable PID controller presented in [32]. LPV controller is to steer the ARD on the reference

Figure 12 Desired and actual state-space trajectories

a Radial distance profiles Rd and R
b Downrange profiles Xd and X

1090 IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-cta.2008.0057
 www.ietdl.org

Figure 13 Error trajectories

a Radial distance error
b Downrange error

trajectory in the presence of dispersions on initial kinematics [4] MEASE K.D., KREMER J.: ‘Shuttle entry guidance revisited
conditions. using nonlinear geometric methods’, AIAA J. Guid.,
Control Dyn., 1994, 6, (17), pp. 1350 – 1356
4 Conclusion [5] LU P.: ‘Entry guidance and trajectory control for
In this paper, a coupled LPV and flatness design scheme is reusable launch vehicle’, AIAA J. Guid., Control and Dyn.,
proposed to solve the hypersonic guidance problem for 1997, 1, (20), pp. 143– 149
ARD-like capsules. The proposed methodology is based on
LPV techniques and takes advantage of the flatness [6] NECKEL T., TALBOT C., PETIT N.: ‘Collocation and inversion for
properties of the longitudinal vehicle model. The design a reentry optimal control problem’. Proc. Fifth Int. Conf.
problem for the guidance was formulated as a fictitious Launcher Technology, ESA, Madrid, Spain, 2003
augmented LPV problem, making it possible to be solved
by available LPV techniques and tools. Finally, the [7] PETIT N., MILAM M., MURRAY R.: ‘Inversion based trajectory
suggested approach has been applied on the hypersonic optimization’. Proc. Fifth IFAC Symp. Nonlinear Control
guidance of the longitudinal model of ARD. Monte Carlo Systems Design, IFAC, St Petersburg, Russia, 2001,
simulation results are presented to demonstrate the pp. 2767 – 2774
potential of the overall approach. Further investigations are
needed to access the possible extension of this guidance [8] MURRAY R., HAUSER J., JADBABAIE A., MILAM M., PETIT N., DUNBAR
scheme for gliding re-entry vehicles, to deal with the W., FRANZ R.: ‘Online control customization via optimization-
Terminal Area Energy Management (TAEM) phase. based control’ in SAMAD T. , BALAS G. (EDS.): ‘Software-
enabled control, information technology for dynamical
systems’ (Wiley-Interscience, New York, 2003), Ch. 9,
5 Acknowledgments pp. 149– 174
The authors would like to thank Christophe Léonard from
Astrium Space Transportation and Vincent Morio at [9] CHAPLAIS F., PETIT N.: ‘Inversion in indirect optimal control
Bordeaux University, IMS Lab, for fruitful and helpful of multivariable systems’, ASAIM: Control, Optimisation and
discussions on this work. Calculus of Variations, 2008, 14, (2), pp. 294– 317

[10] CARSON J., EPSTEIN M., MACMYNOWSKI D., MURRAY R.: ‘Optimal
6 References nonlinear guidance with inner-loop feedback for
hypersonic re-entry’. Proc. 2006 American Control
[1] REGAN F.J., ANANDAKRISHNAN S.M.: ‘Dynamics of Conference, AIAA/CIRA, USA, 2006, p. 6
atmospheric re-entry’ (AIAA, New York, USA, 1993)
[11] APKARIAN P., GAHINET P.: ‘A convex characterization of gain-
[2] ARRINGTON J.P., JONES J.J.: ‘Shuttle performance: lessons scheduled h1 controllers’, IEEE Trans. Automatic Control,
learned’. Proc. NASA conf., NASA, Hampton, USA, 1983, 1995, 5, (40), pp. 853– 864
p. 2283
[12] BIANNIC J.M., APKARIAN P., GARRARD W.L.: ‘Parameter-varying
[3] HARPOLD J., GRAVES C.: ‘Shuttle entry guidance’, control of high performance aircraft’, AIAA J. Guid.,
J. Astronaut. Sci., 1979, 3, (27), pp. 239– 268 Control Dyn., 1998, 2, (20), pp. 225 – 231

IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092 1091
doi: 10.1049/iet-cta.2008.0057 & The Institution of Engineering and Technology 2009
www.ietdl.org

[13] FLIESS M., LÉVINE J., MARTIN P., ROUCHON P.: ‘Flatness and [25] BETTS J.T.: ‘Practical methods for optimal control using
defect of non-linear systems: introduction theory and nonlinear programming’ (SIAM Press, Philadelphia, USA, 2001)
examples’, Int. J. Control, 1995, 6, (61), pp. 1327 – 1361
[26] United States Committee on Extension to the
[14] SMITH R.P.: ‘Functional control law design using exact Standard Atmosphere: ‘U.S. standard atmosphere’,
non-linear dynamic inversion’. Proc. AIAA Guidance, National Oceanic and Atmospheric Administration,
Navigation and Control Conference, AIAA, Scottsdale, National Aeronautics and Space Administration, United
USA, 1994, pp. 481 – 489 States Air Force, Washington, D.C., 1976

[15] LEVINE J.: ‘On necessary and sufficient conditions for [27] HARPOLD J.C., GRAVES C.A.J.: ‘Re-entry targeting philosophy
differential flatness’, 2006, URL http://www.citebase.org/ and flight results from apollo 10 and 11’. Proc. AIAA Eigth
abstract?id¼oai:arXiv.org:math/0605405 Aerospace Sciences Meeting, vol. N70-28, AIAA,
New York, USA, 1970.
[16] MORIO V., CAZAURANG F. , ZOLGHADRI A.: ‘On the formal
characterization of reduced-order at outputs over an ore [28] ROUCHON P.: ‘Necessary condition and genericity of
algebra’. Proc. IEEE Multi conf. Systems and Control, IEEE, dynamic feedback linearization’, J. Math. Syst. Estim.
San Antonio, TX, USA, 2008, pp. 208 – 214 Control, 1994, 3, (5), pp. 345– 358

[17] ANTRITTER F., LÉVINE J.: ‘Towards a computer algebraic [29] ZERAR M., CAZAURANG F., ZOLGHADRI A.: ‘Robust tracking of
algorithm for at output determination’. ISSAC’08 Proc. nonlinear mimo uncertain at systems’. Proc. IEEE Int.
21st Int. Sym. Symbolic and Algebraic Computation, ACM, Conf. Systems Man and Cybernetics, ESA, The Hague, The
New York, NY, USA, 2008, pp. 7 – 14 Netherlands, 2004, pp. 536– 541

[18] FLIESS M., LÉVINE J., MARTIN P., ROUCHON P.: ‘A lie-backlund [30] STEINBUCH M., TERLOUW J.C., BOSGRA O.H. : ‘Robustness
approach to equivalence and flatness of nonlinear analysis for real and complex perturbations applied to an
systems’, IEEE Trans. Autom. Control, 1999, 5, (44), electro-mechanical system’. Proc. 1991 American Control
pp. 922– 937 Conf., IEEE, Boston, USA, 1991, pp. 556 – 561

[19] VAN NIEUWSTADT M., MURRAY R.M.: ‘Real time trajectory [31] GAHINET P., NEMIROVSKII A., LAUB A.J., CHILALI M.: ‘The LMI
generation for differentially flat systems’. Proc. 34th IEEE control toolbox for use with matlab’ (The Mathworks Inc.,
Control and Decision Conference, IEEE, New Orleans, USA, Massachusetts, USA, 1995)
1995, pp. 4224 – 4230
[32] ZHENG F., WANG Q.-G., LEE T.: ‘On the design of multivariable
[20] FLIESS M., LÉVINE J., MARTIN P. , ROUCHON P. : ‘Generalized pid controllers via lmi approach’, Automatica, 2002, 3, (38),
controller canonical forms for linear and nonlinear pp. 517– 526
dynamics’, IEEE Trans. Autom. Control, 1990, 9, (35),
pp. 994– 1001 7 Appendix
[21] HAGENMEYER V., DELALEAU E.: ‘Robustness analysis of exact Bounds of time varying parameters Aij (t) and Bij (t)
feedforward linearization based on differential flatness’,
Automatica, 2003, 11, (39), pp. 1941– 1946 0:0032 100  A21 (t)  þ1:4252 106

[22] GHAOUI L.E. , SCORLETTI G.: ‘Control of rational 0:0545 100  A22 (t)  3:3675 107
systems using linear fractional representations and linear þ0:0022 100  A24 (t)  þ0:0294 100
matrix inequalities’, Automatica, 1996, 9, (39),
pp. 1273 – 1284 3:0681 107  A41 (t)  þ0:0093 100

[23] PIGNIÉ J., CLAR P., FERREIRA E., BOUAZIZ L. , CAILLAUD J.: 0:0044 100  A42 (t)  þ0:0074 100
‘Navigation, guidance and control of the atmospheric re- 0:0566 100  A44 (t)  þ5:2794 105
entry demonstrator’. Proc. Third ESA Int. Conf. Spacecraft
Guidance, Navigation and Control Systems, ESA, þ7:3357 104  B21 (t)  17:2091 100
Noordwijk, The Netherlands, 1996, p. 149
þ8:1090 104  B22 (t)  20:8171 100
[24] VINH N.X. , BUSEMANN A., CULP R.D.: ‘Hypersonic and 49:9945 100  B41 (t)  0:0019 100
planetary entry flight mechanics’ (The University of
Michigan Press, Ann Arbor, USA, 1980) 0:0348 100  B41 (t)  þ3:4230 100

1092 IET Control Theory Appl., 2009, Vol. 3, Iss. 8, pp. 1081 – 1092
& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-cta.2008.0057