Vous êtes sur la page 1sur 15

Published in IET Control Theory and Applications

Received on 6th January 2009


Revised on 19th July 2009
doi: 10.1049/iet-cta.2009.0008
ISSN 1751-8644
Asymptotic generalised dynamic inversion
attitude control
A.H. Bajodah
Aeronautical Engineering Department, King Abdulaziz University, Jeddah 21589, Saudi Arabia
E-mail: abajodah@kau.edu.sa
Abstract: This study introduces a generalised dynamic inversion control methodology for asymptotic spacecraft
attitude trajectory tracking. An asymptotically stable second-order servo-constraint attitude deviation dynamics is
evaluated along spacecraft equations of motion, resulting in a linear relation in the control vector. A control law
that enforces the servo-constraint is derived by generalised inversion of the relation using the Greville formula.
The generalised inverse in the particular part of the control law is scaled by a decaying dynamic factor that
depends on desired attitude trajectories and body angular velocity components. The scaled generalised
inverse uniformly converges to the standard MoorePenrose generalised inverse, causing the particular part
to converge uniformly to its projection on the range space of the controls coefcient generalised inverse, and
driving spacecraft attitude variables to nullify attitude deviation. The auxiliary part of the control law acts on
the controls coefcient nullspace, and it provides the spacecraft internal stability with the aid of the null-
control vector. The null-control vector construction is made by means of novel semidenite nullprojection
control Lyapunov function and state-dependent null-projected Lyapunov equation. The generalised dynamic
inversion control signal is multiplied by an exponential factor during transient closed-loop response to
enhance the control signal in terms of magnitude and rate of change. Illustrating examples show efcacy of
the methodology.
1 Introduction
Non-linear dynamic inversion (NDI) is a transformation
from a non-linear system to an equivalent linear system,
performed by means of a change of variables and through
feedback. The theory of NDI was initially formalised by Su
[1] and Hunt et al. [2], and its rst reported application to
spacecraft attitude control problem is due to Dwyer [3].
The methodology is widely accepted among control system
practitioners because it substantially facilitates control
system design. Additionally, it preserves the non-linear
nature of plants dynamics and thus it avoids limitations of
linearising approximations.
Classical NDI is based on constructing inverse mapping of
the controlled plant and augmenting it within the feedback
control system. Therefore the linearising transformation
depends heavily on the nature of the plant, and it becomes
difcult or impossible as complexity of the plant increases.
For this reason it may become necessary to introduce
simplifying approximations to the plants mathematical
model in order to obtain the NDI linearising
transformation, which adversely affects closed-loop control
system stability and performance characteristics in real
implementations of the transformation. Additionally, NDI
in particular situations must be local in state space, as it is
the case for spacecraft attitude dynamics [3].
A paradigm shift was made to NDI by Paielli and Bach in
[4] in the context of spacecraft attitude control. Their
approach aims to impose a prescribed dynamics on the
errors of spacecraft attitude variables from their desired
trajectory values. Rather than inverting the mathematical
model of the spacecraft, the desired attitude error dynamics
is inverted for the control variables that realise the
dynamics. The transformation is global and does not
involve deriving inverse equations of motion. It involves
simple mathematical inversions of terms that include
motion variables and control system design parameters, and
therefore it is easier and more systematic than its counterpart.
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 827
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
Nevertheless, a commonfeature betweenthe above-mentioned
NDI approaches is that the linearising transformation eliminates
non-linearities from the transformed closed-loop system
dynamics without distinguishing between types of non-
linearities. For instance, a non-linearity may cause the spacecraft
at a particular time instant to accelerate in a manner that is in
favour of the control objective, for example, in performing a
desired manoeuvre. Yet a needless control effort is made to
eliminate that non-linearity, and an additional control effort is
made to satisfy the control objective. This can be extremely
disadvantageous as large control signals may cause actuator
saturation and control systems failure.
It is therefore desirable to come up with a dynamic
inversion control design methodology that provides a global
linearising transformation, gets around the difculty of
plants mathematical model inversion and requires less
control effort to perform the inversion by avoiding blind
cancelation of dynamical systems non-linearity. These
features are offered by generalised non-linear dynamic
inversion (GNDI) control. Some basic elements of GNDI
were introduced by the author in [5, 6], together with
particular GNDI control designs. Every design exhibits
different characteristics in terms of closed-loop system
stability, performance and control signal behaviour.
The GNDI methodologies add the exibility of non-square
inversion to the simplicity of NDI by observing that although
the inverse mathematical model of a plant is unique, the
inverse dynamics problem is a problem with non-unique
solution, that is, there exist innite sets of values for the state
and control variables that satisfy a specic dynamics at a
specic time instant. Therefore the original philosophy of
dynamic inversion is quite restrictive, and there must exist
innite generalised inversion control laws that realise a
servo-constraint dynamics, that is, the differential equation
in systems state variables that has its steady-state solution
satises the control design objective.
A GNDI spacecraft control design begins by dening a norm
measure function of attitude error fromdesired attitude trajectory.
An asymptotically stable second-order linear differential equation
in the norm function is prescribed, resembling the desired
servo-constraint dynamics. The differential equation is then
transformed to a relation that is linear in the control vector by
differentiating the norm measure function along the trajectories
dened by solution of the spacecrafts state-space mathematical
model. The Greville formula [7] is utilised thereafter to invert
this relation for the control law required to realise desired stable
linear servo-constraint dynamics.
The Greville generalised inversion formula exhibits useful
geometrical features of generalised inversion. It consists of
auxiliary and particular parts, residing in the nullspace of
the inverted matrix and the complementary orthogonal
range space of its transpose, respectively. The particular
part involves the standard MoorePenrose generalised
inverse (MPGI) [8, 9], and the auxiliary part involves a free
null-vector that is projected onto nullspace of the inverted
matrix by means of a nullprojection matrix.
The Greville formula is capable of modelling solution
non-uniqueness to problems where requirements can be
satised in more than one course of action. For that reason, the
formula had remarkable contributions towards advancements
in science and engineering. In the arena of robotics, it has been
extensively used in analysis and design of kinematically
redundant manipulators [10]. Utilisation of the formula in the
eld of analytical dynamics was made by deriving the
UdwadiaKalaba equations of motion for constrained
dynamical systems [11]. Other applications include the
evolving subject of pointwise optimal control in the sense
of Gauss principle of least constraint [12], see for example,
[1315].
However, a fundamental shortcoming of the Greville
formula for matrices containing dynamic elements is MPGI
singularity. This problem is well known in applications of
the formula, and it has been thoroughly investigated in
the subject of inverse kinematics, for example, [16]. The
reason for MPGI singularity is that a matrix with
continuous function elements has discontinuous MPGI
function elements. These discontinuities occur whenever the
inverted matrix changes rank. Moreover, these discontinuous
elements approach innite values at discontinuities.
Accordingly, the corresponding solutions provided by the
Greville formula must also be discontinuous and unbounded.
The MPGI singularity forms an obstacle in the way of
utilising the Greville formula in engineering solutions.
Several remedies for the problem of generalised inversion
instability due to MPGI singularity have been offered in
the literature of robotics and control moment gyroscopic
devices, in what has become known as the singularity
avoidance problem. Remedies are either nullspace
parametrisation-based, made by proper choices of the null-
vector in the auxiliary part of the Greville formula, for
example, [1719] or approximation-based, made by
modifying the denition of the generalised inverse itself in
the particular part of the formula, for example, [2022].
A few solutions to the generalised inversion instability
problem have been provided in the context of GNDI
control. One solution is made by deactivating the particular
part of the Greville formula-based control law in the
vicinity of singularity, resulting in discontinuous control
laws [23]. Another solution is presented in [5], made by
modifying the denition of MPGI by means of a damping
factor, resulting in uniformly ultimately bounded attitude
trajectory tracking and a trade-off between generalised
inversion stability and closed-loop system performance.
This paper introduces a novel concept of generalised
inversion by which the Greville formula is modied for
guaranteed generalised inversion stability and asymptotic
tracking. The concept is based on replacing the MPGI
828 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
matrix in the Greville formula by a growth-controlled
dynamically scaled generalised inverse (DSGI) matrix, such
that the DSGI matrix elements converge uniformly to the
standard MPGI matrix elements, resulting in uniform
asymptotic convergence of spacecraft attitude parameters to
their desired values in time.
The procedure begins by dening a reference angular
velocity vector that depends on desired attitude trajectory
vector function and its time derivative. The reference angular
velocity vector has the property that its convergence to
spacecrafts body angular velocity vector implies spacecrafts
attitude vector convergence to desired attitude vector.
The DSGI is constructed by adding a dynamic scaling
factor to the controls coefcients squared norm in the
denominators of MPGIs elements. The dynamic scaling
factor is the pth integer power of the vector p norm of error
between spacecraft body angular velocity and reference
angular velocity. The null-control vector in the auxiliary
part of the control law is designed to nullify the dynamic
scaling factor such that the DSGI asymptotically recovers
the structure of the MPGI. This causes the particular part
of the control law to converge to its projection on the range
space of the controls coefcients MPGI, which drives the
attitude variables to satisfy desired servo-constraint stable
dynamics, resulting in global asymptotic attitude trajectory
tracking.
The GNDI paradigm makes it possible to merge dynamic
inversion with other control system design methodologies to
enhance control system design features. This is achieved
through construction of the null-control vector that appears
explicitly in the auxiliary part of the control law. The null-
control vector provides by its afne parametrisation of
controls coefcients nullspace a convenient way to stabilise
internal dynamics of the closed-loop control system
without affecting servo-constraint realisation.
In particular, Lyapunov control design can be augmented
with GNDI to reduce control effort required by dynamic
inversion. Lyapunov control design has a successful history
in spacecraft control, and is well known to consume less
energy than dynamic inversion. This fact is veried in the
context of GNDI in [6], and it makes merging the two
methodologies quite promising for control engineering
practice.
This paper utilises some attractive geometrical features of
generalised inversion with a novel type of positive
semidenite control Lyapunov functions [24, 25] for null-
control vector design. The control Lyapunov function
involves the controls coefcient nullprojection matrix, and
applying Lyapunov direct method [26] yields a controls
coefcient null-projected Lyapunov equation. The equation
is solved to obtain a simple control law for global
asymptotic stability of internal spacecraft dynamics.
Despite the excellent performance of GNDI control design,
heavy controls load at initial closed-loop control time causes
relatively big initial control signal magnitude compared to
control signal magnitudes at later stages of closed-loop
control times. Moreover, for the control system to enforce
desired servo-constraint dynamics, a rapid decrease of
control signal magnitude follows. This causes undesirable
high-frequency behaviour of spacecraft angular velocity and
may excite un-modelled structural modes, which adversely
affects spacecraft closed-loop dynamics. In order to avoid the
possible corrupting of spacecraft functionality, the GNDI
control signal is multiplied by an exponential factor during
transient closed-loop control system response, resulting in
substantial enhancement of control signal behaviour in terms
of initial magnitude and rate of change.
The contribution of this article is two-fold. First, the
GNDI methodology is modied for global asymptotic
attitude trajectory tracking and globally asymptotically
stable internal dynamics. Second, the control design is
enhanced to reduce control signal magnitude and rate
during transient closed-loop control system response.
2 Spacecraft mathematical model
The spacecraft mathematical model is given by the
following system of kinematical and dynamical differential
equations
r = G(r)v, r(0) = r
0
(1)
v = J
1
v

J v +t, v(0) = v
0
(2)
where r [ R
31
is the spacecraft vector of modied Rodrigues
attitude parameters (MRPs) [27], v [ R
31
is the vector of
spacecraft angular velocity components in its body reference
frame, J [ R
33
is the spacecrafts symmetric body
moments of inertia matrix and t := J
1
u [ R
31
is the
vector of scaled control torques, where u [ R
31
contains
the applied gas jet actuator torque components about the
spacecrafts principal axes. The cross product matrix x

which corresponds to a vector x [ R


31
is skew symmetric
of the form
x

=
0 x
3
x
2
x
3
0 x
1
x
2
x
1
0

and the matrix valued function G(r) : R


31
R
33
is nite and invertible for any value of r [ R
31
, and is
given by
G(r) =
1
2
1 r
T
r
2
I
33
r

+rr
T

(3)
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 829
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
3 Servo-constraint attitude
deviation dynamics
Let r
r
(t) [ R
31
be a prescribed desired spacecraft attitude
vector such that r
r
(t) is twice continuously differentiable
in t. The spacecraft attitude error vector from r
r
(t) is
dened as
r(r, t) := r(t) r
r
(t) (4)
Consequently, the scalar attitude deviation norm measure
function f: R
31
[0, 1) [0, 1) is dened to be the
squared norm of r
f(r, t) = r(r, t)
2
(5)
Therefore a servo-constraint on the attitude dynamics that
represent the control design objective is given by
f(r, t) ; 0 (6)
The rst two time derivatives of f along the spacecraft
trajectories given by the solutions of (1) and (2) are

f =
f
r
G(r)v +
f
t
(7)
= 2 r
T
(r, t)[G(r)v r
r
(t)] (8)
and

f = 2[G(r)v r
r
(t)]
T
[G(r)v r
r
(t)]
+2 r
T
(r, t)[

G(r, v)v +G(r)[ J
1
v

J v +t] r
r
(t)]
(9)
where

G(r, v) is the time derivative of G(r) obtained by
differentiating the individual elements of G(r) along the
kinematical subsystem given by (1). The desired
dynamics of f that leads to asymptotic realisation
of the servo-constraint given by (6) is described to
be stable second-order in the general functional form
given by

f = L(f,

f, t) (10)
where L is continuous in its arguments. With f,

f and

f given by (5), (8) and (9), it is possible to write (10) in


the pointwise-linear form
A(r, t)t = B(r, v, t) (11)
where the vector valued function A(r, t) : R
31

[0, 1) R
13
is given by
A(r, t) = 2 r
T
(r, t)G(r) (12)
and the scalar valued function B(r, v, t) : R
31
R
31

[0, 1) R is given by
B(r, v, t) = 2[G(r)v r
r
(t)]
T
[G(r)v r
r
(t)]
2 r
T
(r, t)[

G(r, v)v +G(r)J
1
v

J v r
r
(t)]
+L(f(r, t),

f(r, v, t), t)
(13)
The row vector function A(r, t) is the controls coefcient
of the attitude deviation norm measure dynamics given by
(10) along the spacecraft trajectories, and the scalar
function B(r, v, t) is the corresponding controls load.
3.1 Linear attitude deviation dynamics
A special choice of L(f,

f, t) is
L(f,

f, t) = c
1

f c
2
f (14)
where c
1
and c
2
are positive scalars. With this choice of
L(f,

f, t), the stable attitude deviation servo-constraint
dynamics given by (10) becomes linear in the form

f +c
1

f +c
2
f = 0 (15)
The corresponding controls load B(r, v, t) given by (13)
becomes
B(r, v, t) =2[G(r)v r
r
(t)]
T
[G(r)v r
r
(t)]
2 r
T
(r, t)[

G(r, v)v+G(r)J
1
v

J v r
r
(t)]
2c
1
r
T
(r, t)[G(r)v r
r
(t)] c
2
r(r, t)
2
(16)
4 Realisability of attitude
deviation dynamics
Realisability of attitude deviation dynamics is a pointwise
assessment of the control system for the ability to enforce a
servo-constraint on the controlled spacecraft.
Denition 1 (Realisability of attitude deviation
servo-constraint dynamics): For a given desired
spacecraft attitude vector r
r
(t), the linear attitude deviation
norm measure dynamics given by (10) is said to be
realisable by spacecraft equations of motion (1) and (2) at
specic values of r and t if there exists a control vector t
that solves (11) for these values of r and t. If this is true
for all r and t such that r(r, t) =0
31
, then the linear
attitude deviation norm measure dynamics is said to be
globally realisable by the spacecraft equations of motion.
Realisability of a prescribed attitude dynamics judges on
the existence of control vector values that enforce that
dynamics for every attitude state and at every time instant.
830 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
This distinguishes the notion from the equivalent notion of
controllability, which evaluates the ability of driving a state
from one point to another in the state space. The algebraic
form of (11) is a substitute for the differential form of (10),
and realisability of the dynamics given by (10) by spacecraft
equations of motion (1) and (2) for a given desired
spacecraft attitude vector r
r
(t) at specic values of r and t
is equivalent to the existence of a control vector t that
solves (11) for these values of r and t.
Proposition 1 (Global realisability of servo-constraint
dynamics): The attitude deviation servo-constraint
dynamics given by (10) is globally realisable by spacecraft
equations of motion (1) and (2).
Proof: The existence of a vector t that solves (11) at specic
values r
w
, v
w
and t
w
is equivalent to the fact that
B(r
w
, v
w
, t
w
) is in the range space of A(r
w
, t
w
). This is
possible for any value that B(r
w
, v
w
, t
w
) may take, provided
that not all elements of A(r
w
, t
w
) vanish, for which the
equation is said to be consistent. Since G(r) is of full rank for
all r [ R
31
, the expressiongivenby (12) for A(r, t) implies that
A(r, t) = 0
13
r(r, t) = 0
31
(17)
which proves realisability of (10) for all r and t such that
r(r, t) =0
31
. A
5 Reference angular velocity
and acceleration
Invertibility of the matrix G(r) in (1) makes it possible to
solve explicitly for the vector v, which takes the form
v = G
1
(r) r (18)
Therefore a vector of reference angular velocity v
r
(t) is
obtained by substituting the desired vector of attitude
variables r
r
(t) and its time derivative r
r
(t) in place of r and
r, respectively, in (18), such that
v
r
(t) = G
1
(r
r
(t)) r
r
(t) (19)
A vector of reference angular acceleration v
r
(t) at
t = t(r, v
r
, t) is obtained from (2) by substituting v
r
(t) in
place of v such that
v
r
(t) = J
1
v

r
(t)J v
r
(t) +t(r, v
r
, t) (20)
6 Generalised dynamic inversion
attitude control
The MPGI-based Greville formula is used now to obtain a
preliminary form of GNDI spacecraft attitude control laws.
Proposition 2 (Linearly parameterised attitude
control laws): The innite set of all control laws that
globally realise the attitude deviation servo-constraint
dynamics given by (10) by the spacecraft equations of
motion is parameterised by an arbitrarily chosen null-
control vector y [ R
31
as
t = A
+
(r, t)B(r, v, t) +P(r, t)y (21)
where A
+
stands for the MPGI of the controls coefcient
(abbreviated as CCGI), and is given by
A
+
(r, t) =
A
T
(r, t)
A(r, t)A
T
(r, t)
, A(r, t) =0
13
0
31
, A(r, t) = 0
13

(22)
and P(r, t) [ R
33
is the corresponding controls coefcient
nullprojector (CCNP), given by
P(r, t) = I
33
A
+
(r, t)A(r, t) (23)
Proof: Multiplying both sides of (21) by A(r, t) recovers the
algebraic system given by (11). Therefore t enforces the
attitude deviation servo-constraint dynamics given by (10)
for all A(r, t) =0
13
. A
The controls coefcient nullprojector P(r, t) projects the
null-control vector y onto the nullspace of the controls
coefcient A(r, t). Therefore the choice of y does not
affect realisability of the linear attitude deviation norm
measure dynamics given by (10). Nevertheless, the choice
of y substantially affects transient state response and
spacecraft internal stability, that is, stability of the closed-
loop dynamical subsystem
v = J
1
v

J v +A
+
(r, t)B(r, v, t) +P(r, t)y (24)
obtained by substituting (21) into (2) [28].
7 Perturbed controls coefcient
nullprojector
Denition 2 (Perturbed controls coefcient
nullprojector): The perturbed CCNP

P(r, d, t) is given by

P(r, d, t) := I
33
h(d)A
+
(r, t)A(r, t) (25)
where h(d) : R
11
R
11
is any continuous function such
that
h(d) = 1 if and only if d = 0
7.1 Properties of perturbed controls
coefcient nullprojector
The rst and second properties below are proven in [5]. The
third property is veried by direct evaluation of P(r, t) and

P(r, d, t) expressions given by (23) and (25).


1.

P(r, d, t) is of full rank for all d =0.
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 831
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
2.

P
1
(r, d, t)P(r, t) = P(r, t)

P
1
(r, d, t) = P(r, t).
3. P(r, t)

P(r, d, t) =

P(r, d, t)P(r, t) = P(r, t).


8 Generalised inverse instability
The expression given by (12) for the controls coefcient
implies that if the dynamics given by (10) is globally
realisable by the spacecraft equations of motion, then
lim
f0
A(r, t) = 0
13
(26)
Accordingly, the expression of A
+
(r, t) given by (22) implies
that for any initial condition r
0
=r
r
(0), state trajectories of
the closed-loop control system given by (1) and (24) must
evolve such that
lim
f0
A
+
(r, t) = 1
31
(27)
That is, A
+
(r, t) must go unbounded as the spacecraft
dynamics approaches steady state. This is a source of
instability for the closed-loop system because it causes the
control law expression given by (21) to become unbounded.
One solution to this problem is made by switching the
value of the CCGI according to (22) to A
+
(r, t) = 0
31
when the controls coefcient A(r, t) approaches singularity,
which implies deactivating the particular part of the control
law as the closed-loop system reaches steady state, leading to
a discontinuous control law [23].
Alternatively, a solution is made by replacing the MPGI in
(21) by a damped generalised inverse [5], resulting in
uniformly ultimately bounded trajectory tracking errors, and
a trade-off between generalised inversion stability and
steady-state tracking performance. A solution to this
problem that avoids control law discontinuity and provides
asymptotic attitude tracking is made by replacing the
MPGI in (21) by the growth-controlled DSGI introduced
in the next section.
9 Dynamically scaled generalised
inverse
The notion of dynamically scaled generalised inversion is
crucial for internally stable asymptotic generalised inverse
attitude control.
Denition 3 (DSGI): The DSGI A
+
s
(r, v, t) : R
31

R
31
[0, 1) R
31
is given by
A
+
s
(r, v, t) =
A
T
(r, t)
A(r, t)A
T
(r, t) +v v
r
(t)
p
p
(28)
where the positive integer p is the generalised inversion
dynamic scaling index, and
p
is the vector p norm.
9.1 Properties of DSGI
The following properties can be veried by direct evaluation
of the CCGI A
+
(r, t) given by (22) and its dynamic scaling
A
+
s
(r, v, t) given by (28).
1. A
+
s
(r, v, t)A(r, t)A
+
(r, t) = A
+
s
(r, v, t).
2. A
+
(r, t)A(r, t)A
+
s
(r, v, t) = A
+
s
(r, v, t).
3. (A
+
s
(r, v, t)A(r, t))
T
= A
+
s
(r, v, t)A(r, t).
4. lim
vv
r
(t)
p
0
A
+
s
(r, v, t) = A
+
(r, t).
10 DSGI control
The DSGI control law is obtained by replacing the CCGI in
the particular part of the expression given by (21) by the
DSGI as
t
s
= A
+
s
(r, v, t)B(r, v, t) +P(r, t)y (29)
resulting in the following spacecraft closed-loop system
equations
r = G(r)v, r(0) = r
0
(30)
v = J
1
v

J v +A
+
s
(r, v, t)B(r, v, t)
+P(r, t)y, v(0) = v
0
(31)
Proposition 3 (Bounded trajectory tracking error): If
the null-control vector y in the control law expression given
by (29) is chosen such that the angular velocity vector v of
the closed-loop system given by (30) and (31) satises
v v
r
(t)
p
, 1, t 0 (32)
then the resulting closed-loop attitude trajectory error vector
r(r, t) is bounded.
Proof: Since the matrix G(r) has nite elements for any
attitude vector r, then it is evident from the expression of
the controls coefcient A(r, t) given by (12) that A(r, t) is
bounded if and only if r(r, t) is bounded. Therefore
assuming on the contrary that there exists a null-control
vector y that causes the closed-loop angular velocity vector
v to satisfy (32) such that
lim
t1
r(r, t) = 1 (33)
then it follows that
lim
t1
A(r, t) = 1 (34)
which implies from (28) and (32) that
lim
t1
A
+
s
(r, v, t) = A
+
(r, t) (35)
832 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
It accordingly follows from the expression of t
s
given by
(29) that
lim
t1
t
s
= t (36)
where t is given by (21), causing the closed-loop system
trajectories to asymptotically satisfy the stable servo-
constraint dynamics given by (15), and resulting in
lim
t1
f = 0 (37)
which contradicts (33). Therefore the control law t
s
given by
(29) must yield
r(r, t) , 1 (38)
and must yield bounded elements of A(r, t). A
Proposition 4 (Asymptotic attitude trajectory
tracking): If the null-control vector y in the control law
expression given by (29) is chosen such that the angular
velocity vector v of the closed-loop system given by (30)
and (31) satises (32) and such that
lim
t1
v = v
r
(t) (39)
then the attitude vector r asymptotically converges to the
desired attitude vector r
r
(t).
Proof: Let f
s
be a norm measure function of the
attitude deviation obtained by applying the control law
given by (29) to the spacecraft equations of motion (1)
and (2), and let

f
s
,

f
s
be its rst two time derivatives.
Therefore
f
s
:= f
s
(r, t) = f(r, t) (40)

f
s
:=

f
s
(r, v, t) =

f(r, v, t) (41)

f
s
:=

f
s
(r, v, t
s
, t) =

f(r, v, t, t) +A(r, t)t


s
A(r, t)t (42)
where t and t
s
are given by (21) and (29), respectively.
Adding c
1

f
s
+c
2
f
s
to both sides of (42) yields

f
s
+c
1

f
s
+c
2
f
s
=

f+c
1

f+c
2
f+A(r, t)t
s
A(r, t)t (43)
= A(r, t)[t
s
t] (44)
Therefore the boundedness of A(r, t) inferred from
Proposition 3 in addition to satisfaction of (39) imply that
lim
t1
[

f
s
+c
1

f
s
+c
2
f
s
] = lim
t1
[A(r, t)[t
s
t]] = 0 (45)
resulting in
lim
t1
f
s
= 0 (46)
and therefore
lim
t1
r = r
r
(t) (47)
for all r
0
[ R
3
. The same conclusion is obtained by
multiplying both sides of (29) by A(r, t), resulting in
A(r, t)t
s
= A(r, t)A
+
s
(r, v, t)B(r, v, t) (48)
where
A(r, t)A
+
s
(r, v, t) =
A(r, t)A
T
(r, t)
A(r, t)A
T
(r, t) +vv
r
(t)
p
p
(49)
Therefore
0 , A(r, t)A
+
s
(r, v, t) 1 (50)
and
lim
vv
r
(t)
A(r, t)A
+
s
(r, v, t) = 1 (51)
Dividing (48) by A(r, t)A
+
s
(r, v, t) yields
A(r, t)

t = B(r, v, t) (52)
where A(r, t) and B(r, v, t) are the same controls coefcient
and controls load in (11), and

t =
t
s
A(r, t)A
+
s
(r, v, t)
(53)
Furthermore, (51) implies that
lim
vv
r
(t)

t = lim
vv
r
(t)
t
s
= t (54)
Therefore

t in the algebraic system given by (52)
asymptotically converges to t, recovering the algebraic
system given by (11), and resulting in asymptotic
convergence of f
s
(t) to f
s
= f = 0, and r to r
r
(t). A
Proposition 4 states that employing the DSGI A
+
s
(r, v, t)
in the attitude control law yields the same attitude
convergence property that is obtained by employing the
CCGI A
+
(r, t), provided that the conditions given by (32)
and (39) are satised. A design of the null-control vector y
is made in the next section to guarantee global satisfaction
of the conditions given by (32) and (39).
Remark 1: It is well known that topological obstruction of
the attitude rotation matrix precludes the existence of globally
stable equilibria for the attitude dynamics [29]. Therefore
although the servo-constraint attitude deviation dynamics
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 833
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
given by (10) is globally realisable, there exists no null-control
that renders the spacecraft attitude dynamics globally stable.
In particular, if r
r
(t) ; 0
31
then for any null-control
vector y there exists an attitude vector r
0
such that the
closed-loop system given by (30) and (31) is unstable in the
sense of Lyapunov.
11 Nullprojection Lyapunov
control design
A Lyapunov-based design of null-control vector y is
introduced in this section to enforce spacecraft internal
stability. Let y be chosen as
y = K(v v
r
(t)) (55)
where v
r
(t) is given by (19), and K [ R
33
is a matrix gain
that is to be determined. Hence, a class of control laws that
realise the attitude deviation norm measure dynamics given
by (10) is obtained by substituting this choice of y in (29)
such that
t
s
= A
+
s
(r, v, t)B(r, v, t) +P(r, t)K(v v
r
(t)) (56)
Consequently, a class of spacecraft closed-loop control
systems that realise the servo-constraint dynamics given by
(10) is obtained by substituting the control law given by
(56) in (2), and it takes the form
r = G(r)v, r(0) = r
0
(57)
v = J
1
v

J v +A
+
s
(r, v, t)B(r, v, t)
+P(r, t)K(v v
r
(t)), v(0) = v
0
(58)
The closed-loop reference angular acceleration vector v
r
(t) is
obtained by replacing v by v
r
(t) in (58), resulting in
v
r
(r, t) = J
1
v

r
(t)J v
r
(t) +A
+
s
(r, v
r
(t), t)B(r, v
r
(t), t)
+P(r, t)K(v
r
(t) v
r
(t))
(59)
= J
1
v

r
(t)J v
r
(t) +A
+
(r, t)B(r, v
r
(t), t) (60)
By introducing the angular velocity error variable
v(t) ; v(t) v
r
(t), the error dynamics

v is obtained from
(58) and (60) as

v = J
1
v

J v J
1
v

r
(t)J v
r
(t)
+A
+
s
(r, v, t)B(r, v, t)
A
+
(r, t)B(r, v
r
(t), t) +P(r, t)K v (61)
The matrix gain K is synthesised by utilising the positive-
semidenite control Lyapunov function
V(r, v, t) = v
T
P(r, t) v (62)
Evaluating the time derivative of V(r, v, t) along solution
trajectories of the error dynamics given by (61) yields

V(r, v, t) =2 v
T
P(r, t)[ J
1
v

J vJ
1
v

r
(t)J v
r
(t)
+A
+
s
(r, v, t)B(r, v, t) A
+
(r, t)B(r, v
r
(t), t)]
+2 v
T
P(r, t)K v+ v
T

P(r, v, t) v (63)
Skew symmetry of the cross product matrix [
.
]

, the
nullprojection property of P(r, t), and the second property
of A
+
s
(r, v, t) imply that the rst term in the above
equation is the zero matrix. Therefore the equation can be
written in the symmetrical form

V(r, v, t) = v
T
[P(r, t)K +K
T
P(r, t)
+

P(r, v, t)] v (64)
Because V(r, v, t) is only positive semidenite, it is
impossible to design a matrix gain K that renders

V(r, v, t) negative denite. Nevertheless, a matrix gain K


that renders

V(r, v, t) negative semidenite guarantees
global Lyapunov stability of v = 0
31
if it asymptotically
stabilises v = 0
31
over the invariant set of r, v, and t
values on which V(r, v, t) = 0. Moreover, the same gain
matrix globally asymptotically stabilises v = 0
31
if and
only if it asymptotically stabilises v = 0
31
over the largest
invariant set of r, v, and t values on which

V(r, v, t) = 0
[24].
Proposition 5: Let K = K(r, v, t) be a full-rank
normal matrix gain, that is, KK
T
= K
T
K for all t 0.
Then the equilibrium point v = 0
31
of the closed-loop
error dynamics given by (61) is asymptotically stable over
the invariant set of r, v, and t values on which
V(r, v, t) = 0.
Proof: Since the matrix P(r, t) is idempotent, the function
V(r, v, t) can be rewritten as
V(r, v, t) = v
T
P(r, t) v = v
T
P(r, t)P(r, t) v (65)
which implies that
V(r, v, t) = 0 P(r, t) v = 0
31
(66)
Therefore
V(r, v, t) = 0 v [ N(P(r, t)) (67)
where N() refers to matrix nullspace. Since the matrix
K(r, v, t) is normal and of full-rank, it preserves
matrix range space and nullspace under multiplication.
Accordingly,
N(P(r, t)) = N(P(r, t)K(r, v, t)) (68)
834 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
which implies from (66) that
V(r, v, t) = 0 P(r, t)K(r, v, t) v = 0
31
(69)
Therefore the last term in the closed-loop error dynamics
given by (61) is the zero vector, and the closed-loop error
dynamics becomes

v = J
1
v

J v J
1
v

r
(t)J v
r
(t)
+A
+
s
(r, v, t)B(r, v, t)
A
+
(r, t)B(r, v
r
(t), t) (70)
On the other hand, since [30, 11]
N(P(r, t)) = R(A
T
(r, t)) (71)
it follows from (67) that
V(r, v, t) = 0 v [ R(A
T
(r, t)) (72)
Accordingly, V(r, v, t) = 0 if and only if there exists a
continuous scalar function a(t), t 0, satisfying
0 , |a(t)| , 1 (73)
such that
v = a(t)A
T
(r, t) (74)
Therefore assuming that v goes unbounded, then A
T
(r, t)
also goes unbounded, both expressions of A
+
(r, t) and
A
+
s
(r, t) given by (22) and (28) must go to zero, and the
closed-loop error dynamics given by (70) approaches the
Lyapunov-stable uncontrolled dynamics

v = J
1
v

J v J
1
v

r
(t)J v
r
(t) (75)
implying boundedness of v, in contradiction with the
original argument. Therefore, the trajectory of v must
remain in a nite region, and it follows from the Poincare
Bendixon theorem [31] that the trajectory goes to the
equilibrium point v = 0
31
.
Theorem 1 (CCNP Lyapunov control design): Let the
controls coefcient nullspace-projected gain matrix be
P(r, t)K(r, v, t) = vec
1
{[

P(r, d
1
, t)

P(r, d
1
, t)]
1
vec[

P(r, v, t) +P(r, t)U(r, v, t)]}
(76)
where the operation is the kronecker sum of matrices,
vec and vec
21
are the matrix vectorising and inverse
vectorising operators [30, p. 251],

P(r, v, t) is obtained by
differentiating the elements of P(r, t) along attitude
trajectory solutions of the closed-loop kinematical
subsystem given by (57), d
1
=0 is an arbitrary scalar, and
U(r, v, t) : R
31
R
31
[0, 1) R
33
is orthogonal
and positive-denite. Then the equilibrium point v = 0
31
of the closed-loop error dynamics given by (61) is globally
asymptotically stable, and the attitude vector r of closed-
loop system equations (57) and (58) is globally
asymptotically convergent to r
r
(t).
Proof: The existence of a matrix gain K that renders the
expression of

V(r, v, t) given by (64) negative semidenite is
guaranteed because the range space of

P(r, v, t) is a subset
from the range space of P(r, t). This is shown by writing
P(r, t) = P(r, t)P(r, t)

P(r, v, t)
= 2P(r, t)

P(r, v, t) (77)
so that
R[

P(r, v, t)] = R[P(r, t)

P(r, v, t)] # R[P(r, t)] (78)
where R() refers to matrix range space. Negative
semideniteness of

V(r, v, t) is equivalent to the existence of
a positive-semidenite matrix function Q(r, v, t) : R
31

R
31
[0, 1) R
33
such that
P(r, t)K +KP(r, t) +

P(r, v, t) +Q(r, v, t) =0
33
(79)
Furthermore, if the above equation is consistent then every term
in the equation must map into the range space of P(r, t), which
implies that a polar decomposition of Q(r, v, t) is given by
Q(r, v, t) = P(r, t)U(r, v, t) (80)
where U(r, v, t) is orthogonal and positive denite.
Substituting (80) into (79) and using the third property of

P(r, d, t) yields

P(r, d
1
, t)P(r, t)K +KP(r, t)

P(r, d
1
, t)
+

P(r, v, t) +P(r, t)U(r, v, t) = 0
33
(81)
The unique solution of the above equation for the gain matrix
nullprojection P(r, t)K is given by [30]
P(r, t)K(r, v, t) = vec
1
{[I
33

P(r, d
1
, t)
+

P(r, d
1
, t) I
33
]
1
vec[

P(r, v, t) +P(r, t)U(r, v, t)]}
(82)
where denotes the kronecker product of matrices.
Equation (82) can be written in the compact form given
by (76). Solution uniqueness of (81) implies that the
symmetric matrix gain K(r, v, t) remains non-singular for
all t 0. Accordingly, K(r, v, t) satises the condition of
Proposition 5. Hence, in addition to rendering V(r, v, t)
negative semidenite, K(r, v, t) guarantees asymptotic
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 835
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
stability of vover the invariant set of r, vand t values on which
V(r, v, t) = 0, and Lyapunov stability of v = 0
31
follows
[24]. Since V(r, v, t) is radially unbounded with respect to
v, Lyapunov stability of v = 0
31
is global. Moreover, it is
noticed from the expression of

V(r, v, t) given by (63) and
from (78) that the largest invariant set of r, v and t on
which

V(r, v, t) = 0 is the same invariant set on which
V(r, v, t) = 0, implying global asymptotic stability of the
equilibrium point v = 0
31
[24]. Global asymptotic
convergence of the attitude vector r to the desired attitude
vector r
r
(t) follows from Proposition 4. A
Since the orthogonal positive-denite matrix function
U(r, v, t) is arbitrary, it follows from (77) that there is no
loss of generality in writing

P(r, v, t) +P(r, t)U(r, v, t) = P(r, t)Q (83)


where Q [ R
33
is an arbitrary orthogonal and positive-
denite constant matrix. Accordingly, the gain matrix
nullprojection P(r, t)K given by (82) can be rewritten as
PK(r, t) = vec
1
{[

P(r, d
1
, t)

P(r, d
1
, t)]
1
vec[P(r, t)Q]} (84)
Increasing the magnitude of Q causes faster convergence of v
to the origin, at the attendant cost of increasing the relative
magnitude of the auxiliary part in the control law.
12 Damped controls coefcient
nullprojector
Although the CCNP has bounded elements, dependency of
CCNP on the unbounded vector A
+
(r, t) may cause
undesirable behaviour of the auxiliary part in the control
law t
s
during steady-state tracking response of time-varying
trajectories. For this reason, a damped controls coefcient
nullprojector (DCCN) P
d
(r, e, t) is used in place of
P(r, t) in (56). The DCCN is dened as
P
d
(r, e, t) := I
33
A
+
d
(r, e, t)A(r, t) (85)
where e is a small positive number, and A
+
d
(r, e, t) is given by
A
+
d
(r, e, t) :=
A
T
(r, t)
A(r, t)A
T
(r, t) +e
(86)
Therefore
lim
f0
A
+
d
(r, e, t) = 0
31
(87)
and consequently
lim
f0
P
d
(r, e, t) = I
33
(88)
Hence, the DCCN maps the null-control vector to itself in
steady-state phase of response, during which the auxiliary
part of the control law converges to the null-control vector.
Independency of nullprojection on the attitude state of the
spacecraft substantially eliminates unnecessary abrupt
behaviour of the control vector.
13 Exponentially factored GNDI
control
The GNDI control design does not guarantee an acceptable
behaviour of the GNDI control signal. If the initial value of
controls load B(r
0
, v
0
, 0) is big, then the magnitude of initial
control signal must be big also. Moreover, continuation of
servo-constraint dynamics enforcement causes a rapid decay
of control signal by one or more orders of magnitude in a
very short time compared to transient response time. Such
control signal characteristics are extremely undesirable from
several technical aspects. For instance, if the control
actuators magnitude or rate limits are met, then the control
system fails to provide the required GNDI control signal. In
order to avoid unnecessarily big GNDI control signal
magnitude and rate of change, the GNDI control signal is
exponentially factored during early closed-loop control time as
t
e
= [1 exp(ht)]t
s
(89)
The control signal t
e
starts at zero values at t 0 and then
converges to t
s
given by (56), where h . 0 determines the
rate of convergence.
Increasing the magnitude of the parameter h causes more
suppression of the control signal in the very early stage of
closed-loop response, but it causes a rapid decay of the
exponential term, which limits the benet of the
exponentially factored design as the control load B(r, v, t)
may remain big after the exponential term vanishes. On the
other hand, excessive decrease in magnitude of the
parameter h causes the closed-loop control signal to remain
small for a long time, which adversely affect the tracking
performance of the control system by delaying steady-state
phase of response. A suitable compromise for the value of
h is in the order of 0.1.
To ensure continuity of the control signal, the
exponentially factored signal t
e
is applied for a sufciently
big time interval before the GNDI signal t
s
takes over.
14 Control system design
procedure
The GNDI control system design methodology for tracking
smooth attitude trajectories is summarised in the following
steps:
1. A desired spacecraft attitude trajectory r
r
(t) is prescribed,
where r
r
is at least twice differentiable in t. The desired
angular velocity vector v
r
(t) is given by (19), where G(r) is
given by (3).
836 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
2. The attitude deviation norm measure dynamics given by
(10) is specied such that the dynamics of f is
asymptotically stable. A special choice is given by (15),
where both c
1
and c
2
are strictly positive.
3. The expressions given by (12) and (13) for A(r, t) and
B(r, v, t) are obtained, where r(r, t) is given by (4).
4. The control law t
s
given by (56) is obtained, where
A
+
s
(r, v, t) is given by (28), the controls coefcient null-
projected gain matrix P(r, t)K is given by the expression of
(84), the perturbed CCNP

P(r, d, t) is given by (25) and
the constant matrix Q is arbitrary but orthogonal and
positive denite.
5. The exponentially factored control law t
e
given by (89) is
used in (2) until the exponential term vanishes, and the
control law t
s
is applied afterwards.
6. Equations (1) and (2) are integrated to obtain the closed-
loop trajectories of r(t) and v(t).
15 Numerical simulations
The rst manoeuvre considered is a rest-to-rest slew
manoeuvre, aiming to reorient a spacecraft at an initial
attitude given by r(0) to a different attitude given by r
r
(T),
where T is the duration of the manoeuvre. The spacecraft
has principal moments of inertia I
11
200 kg m
2
,
I
22
150 kg m
2
, I
33
175 kg m
2
and the control torque
actuators are mounted along principal axes. To insure
smooth transition to the new state, it is required that the
spacecraft attitude variables follow the MRPs trajectories
given by the following fth-order polynomial [32]
r
r
(t) = r
r
(0) + 10
t
T

3
15
t
T

4
+6
t
T

5

[r
r
(T) r
r
(0)] (90)
where r
r
(0) = [0 0 0]
T
, r
r
(T) = [0.70 0.80 1.20]
T
, and
T 180 s. To avoid oscillatory closed-loop state response
induced by underdamped servo-constraint dynamics, it is
preferred to choose values of c
1
and c
2
that yield
overdamped second-order servo-constraint dynamics.
Additionally, damping ratio of the servo-constraint
dynamics should be sufciently big to produce a relatively
short duration of the transient response, which can be
obtained by choosing c
1
to be of an order of magnitude
larger than c
2
. Values of second-order attitude deviation
dynamics constants are chosen to be c
1
= 1.0 and c
2
= 0.1,
resulting in an overdamped servo-constraint dynamics with
damping ratio equals to 1.58. The orthogonal positive-
denite Lyapunov matrix Q is selected to be the identity
matrix, and the function h(d) is taken to be
h(d) =
1
1 +d
(91)
where a design value d = 0.1 is chosen. With
r(0) = [0.25 0.20 0.10]
T
and a dynamic scaling index
p 2, Figs. 1 and 2 show r and v trajectories, respectively.
Fig. 3 shows time history of the GNDI control variables
t
s
1
and t
s
2
, and compares with time history of the
exponentially factored GNDI control variables t
e
1
and t
e
2
,
where h 0.1 is selected. The control vector t
s
takes
initial value t
s
(0) = [7.41 10
2
2.47 10
2
4.94
10
2
]
T
(not shown), and drops rapidly in magnitude to the
order of 10
25
in less than 10 s. The exponential factoring
lasts for the rst 40 s, and it has a minor effect on
spacecraft state response. However, it substantially improves
Figure 2 Rest-to-rest slew manoeuvre: angular velocity
components against t
Figure 1 Rest-to-rest slew manoeuvre: MRPs attitude
parameters against t
Figure 3 Rest-to-rest slew manoeuvre: control torques
against t
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 837
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
the control signal by reducing its magnitude and rate of
change. No discontinuities in the control variables at
t 40 s are tangible because t
e
converges to t
s
in shorter
time. Similar plots of t
s
3
and t
e
3
are obtained, but are not
shown.
The second manoeuvre considered is a trajectory tracking
manoeuvre. The desired attitude trajectory is dened by
reference attitude parameters
r
ri
(t) = cos 0.1t, i = 1, 2, 3 (92)
The spacecraft has principal moments of inertia
I
11
100 kg m
2
, I
22
60 kg m
2
, I
33
75 kg m
2
. For
control torque actuators mounted along principal axes,
dynamic scaling index p 4, d = 0.1, h(d) given by (91),
e = 10
5
, Q = I
33
, r(0) = [0.5 0.4 0.8]
T
and v(0) =
[0.2 0.7 0.4]
T
. Figs. 4 and 5 show time response of the
components of attitude and angular velocity vectors r and
v, respectively. Transient and asymptotic behaviours of
attitude and angular velocity variables indicate excellent
closed-loop control system performance. Fig. 6 shows time
history of attitude deviation scalar f for different values of
dynamic scaling factor p. It is observed that increasing p
has a favourable inuence on attitude deviation and
tracking performance. This is because the expression given
by (28) provides better approximations of the controls
coefcients MPGI for higher values of p as steady-state
response is approached, that is, as the controls coefcient
squared norm in the denominator of the expression
vanishes. However, it is noticed from Fig. 6 that no further
improvement of tracking performance is achieved by
increases p over 4, and that the value of f starts to oscillate
within the order of e. Accordingly, more tracking
performance enhancement requires decreasing e. Finally,
Fig. 7 shows time history of GNDI control variables t
s
1
and t
s
2
and exponentially factored GNDI control variables
t
e
1
and t
e
2
, where h = 0.1 is selected. The t
s
signal suffers
from a high initial magnitude and a rapid rate of decay.
The exponential factoring lasts for the rst 5 s, and it
substantially improves control signal by reducing its
magnitude and rate of change with no major effect on
spacecraft state response or tangible discontinuities in the
control vector components. Similar plots of t
s
3
and t
e
3
are
obtained, but is not shown.
16 Conclusion
Driven by the advantages that generalised dynamic inversion
provide over classical dynamic inversion, this paper builds
on the recently developed GNDI control paradigm by
improving the control law to yield asymptotic tracking of
desired smooth trajectories. The DSGI in the particular
Figure 4 Trajectory tracking manoeuvre: MRPs attitude
parameters against t
Figure 5 Trajectory tracking manoeuvre: angular velocity
components against t
Figure 6 Trajectory tracking manoeuvre: f against t,
e 10
25
Figure 7 Trajectory tracking manoeuvre: control torques
against t
838 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
part of the control law is capable of overcoming controls
coefcient generalised inversion singularity, and it
converges to the standard MPGI as the dynamic scaling
factor decays and closed-loop steady-state response
approaches. The null-control vector in the auxiliary part of
the control law is designed by means of novel semidenite
control Lyapunov function and nullprojected control
Lyapunov equation that utilise geometric features of the
GNDI control laws structure. The exponentially factored
GNDI control signal replaces the GNDI control signal in
early stage of the control time to improve control signal
quality. Perfect plant mathematical modelling and an ideal
control environment are assumed for the purpose of
constructing the null-control vector. The present
construction of the null-control vector can be used as a
base for designing adaptive and robust GNDI control laws
in the presence of input disturbances, measurement noises
and modelling uncertainties.
17 References
[1] SU R.: On the linear equivalents of nonlinear systems,
Syst. Control Lett., 1982, 2, pp. 4852
[2] HUNT L.R., SU R., MEYER G.: Global transformations of
nonlinear systems, IEEE Trans. Autom. Control, 1983, AC-
28, pp. 2431
[3] DWYER III T.: Exact nonlinear control of large angle
rotational maneuvers, IEEE Trans. Autom. Control, 1984,
29, (9), pp. 769774
[4] PAIELLI R.A., BACH R.E.: Attitude control with realization of
linear error dynamics, J. Guidance, Control, Dyn., 1993, 16,
(1), pp. 182189
[5] BAJODAH A.H.: Singularly perturbed feedback
linearization with linear attitude deviation dynamics
realization, Nonlinear Dyn., 2008, 53, (4), pp. 321343
[6] BAJODAH A.H.: Generalized dynamic inversion spacecraft
control design methodologies, IET Control Theory Appl.,
2009, 3, (2), pp. 239251
[7] GREVILLE T.N.E.: The pseudoinverse of a rectangular or
singular matrix and its applications to the solutions of
systems of linear equations, SIAM Rev., 1959, 1, (1),
pp. 3843
[8] MOORE E.H.: On the reciprocal of the general algebraic
matrix, Bull. Am. Math. Soc., 1920, 26, pp. 394395
[9] PENROSE R.: A generalized inverse for matrices, Proc.
Cambridge Philosophical Society, 1955, vol. 51, pp. 406413
[10] SICILIANO B., KHATIB O.: Springer handbook of robotics
(Springer, 2008)
[11] UDWADIA F.E., KALABA R.E.: Analytical dynamics, a new
approach (Cambridge University Press, New York, NY, 1996)
[12] GAUSS C.F.: Ueber ein neues algemeines grundgesetz
der mechanik, Zeit. Fuer reine angew. Math., 1829, 4,
pp. 232235
[13] DE SAPIO V., KHATIB O., DELP S.: Least action principles and
their application to constrained and task-level problems in
robotics and biomechanics, Multibody Syst. Dyn., 2008, 19,
(3), pp. 303322
[14] UDWADIA F.E.: Optimal tracking control of nonlinear
dynamical systems, Proc. R. Soc. Lond. A, 2008, 464,
pp. 23412363
[15] PETERS J., MISTRY M., UDWADIA F., NAKANISHI J., SCHAAL S.: A
unifying framework for robot control with redundant
DOFs, Auton. Robots, 2008, 24, (1), pp. 112
[16] BAKER D.R., WAMPLER II C.W.: On inverse kinematics of
redundant manipulators, Int. J. Robot. Res., 1988, 7, (2),
pp. 321
[17] LIEGEOIS A.: Automatic supervisory control of the
conguration and behavior of multi-body mechanisms,
IEEE Trans. Systems, Man, Cybernet., 1977, 7, (12),
pp. 868871
[18] MAYORGA R.V., JANABI-SHARIFI F., WONG A.K.C.: A fast approach
for the robust trajectory planning of redundant robot
manipulators, J. Robot. Syst., 1995, 12, (2), pp. 147161
[19] YOON H., TSIOTRAS P.: Singularity analysis of variable-
speed control moment gyros, J. Guidance, Control, Dyn.,
2004, 27, (3), pp. 374386
[20] NAKAMURA Y., HANAFUSA H.: Inverse kinematic solutions
with singularity robustness for robot manipulator control,
J. Dyn. Syst. Meas., Control, 1986, 10, (8), pp. 163171
[21] WAMPLER II C.W.: Manipulator inverse kinematic
solutions based on vector formulations and damped
least-squares methods, IEEE Trans. Syst., Man, Cybernet.,
1986, SMC-16, pp. 93101
[22] OH H.S., VADALI S.R.: Feedback control and steering laws
for spacecraft using single gimbal control moment gyros,
J. Astronaut. Sci., 1991, 39, (2), pp. 183203
[23] BAJODAH A.H.: Stabilization of underactuated
spacecraft dynamics via singularly perturbed feedback
linearization, J. King Abdulaziz Univ. Eng. Sci., 2006, 16,
(2), pp. 3553
[24] IQQIDR A., KALITINE B., OUTBIB R.: Semidenite Lyapunov
functions stability and stabilization, Math. Control, Signals
Syst., 1996, 9, (2), pp. 95106
IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840 839
doi: 10.1049/iet-cta.2009.0008 & The Institution of Engineering and Technology 2010
www.ietdl.org
[25] BENSOUBAYA M., FERFERA A., IQQIDR A.: Stabilization
of nonlinear systems by use of semidenite
Lyapunov functions, Appl. Math. Lett., 1999, 12, (7),
pp. 1117
[26] KHALIL H.K.: Nonlinear systems (Prentice-Hall, Inc.,
2002, 3rd edn.)
[27] SHUSTER M.D.: A survey of attitude representation,
J. Astronaut. Sci., 1993, 41, (4), pp. 439517
[28] BAJODAH A.H., HODGES D.H., CHEN Y.H.: Inverse dynamics of
servo-constraints based on the generalized inverse,
Nonlinear Dyn., 2005, 39, (12), pp. 179196
[29] BHAT S., BERNSTEIN D.: A topological obstruction to
continuous global stabilization of rotational motion and
the unwinding phenomenon, Syst. Control Lett., 2000, 39,
(1), pp. 6370
[30] BERNSTEIN D.: Matrix mathematics: theory, facts, and
formulas with application to linear system theory
(Princeton University Press, 2005)
[31] SLOTINE J.E., LI W.: Applied nonlinear control (Prentice-
Hall, 1991)
[32] MCINNES C.R.: Satellite attitude slew manoeuvres using
inverse control, Aeronaut. J., 1998, 102, pp. 259265
840 IET Control Theory Appl., 2010, Vol. 4, Iss. 5, pp. 827840
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cta.2009.0008
www.ietdl.org
Copyright of IET Control Theory & Applications is the property of Institution of Engineering & Technology
and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for individual use.

Vous aimerez peut-être aussi