Spacecraft Attitude Dynamics Bacconi

UNIVERSIT
`
A DEGLI STUDI DI FIRENZE
Dipartimento di Sistemi e Informatica
Dottorato di Ricerca in
Ingegneria Informatica e dellAutomazione
ING-INF/04
XVIII Ciclo
Spacecraft Attitude Dynamics
and Control
Fabio Bacconi
Ph.D. Coordinator
Prof. Edoardo Mosca
Advisors
Prof. Edoardo Mosca
Prof. Alessandro Casavola
Anno Accademico 20052006
Contents
Introduction 1
Prior Literature on Spacecraft Attitude Control . . . . . . . . . . . . . . 2
Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
I Geo-Stationary Spacecraft Attitude Control 7
1 Dynamics of the 3D Pendulum 9
1.1 Rigid Bodies Mathematical Models . . . . . . . . . . . . . . . . . . 10
1.2 The 3D Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Equilibria of the 3D Pendulum . . . . . . . . . . . . . . . . . . . . 17
1.4 Triaxial Attitude Control Testbed . . . . . . . . . . . . . . . . . . . 19
1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2 Special Cases of the 3D Rigid Pendulum 22
2.1 Axisymmetric 3D Rigid Pendulum . . . . . . . . . . . . . . . . . . 23
2.2 Spinning Axisymmetric 3D Rigid Pendulum . . . . . . . . . . . . . 26
2.3 Hanging Equilibrium of the Symmetric Spinning Top . . . . . . . . 28
2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 Stabilization of an Asymmetric 3D Rigid Pendulum 34
3.1 3D Rigid Body Control Torques . . . . . . . . . . . . . . . . . . . . 35
3.2 Asymptotic Stabilization of the Hanging Equilibrium . . . . . . . . 38
3.3 Experiments on Stabilization of the Hanging Equilibrium . . . . . . 43
CONTENTS i
3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4 Attitude Control of an Asymmetric 3D Rigid Pendulum 48
4.1 Local Asymptotic Stabilization of an Arbitrary Equilibrium . . . . . 49
4.2 Attitude Control of a Geo-Stationary Spacecraft . . . . . . . . . . . 52
4.3 Simulation on Stabilization of an Arbitrary Equilibrium . . . . . . . 56
4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5 Attitude Control of a 3D Axially Symmetric Rigid Pendulum 62
5.1 Global Coordinates Model of the 3D Axially Symmetric Pendulum . 64
5.2 Stabilization of the Hanging Equilibrium of the Symmetric Spinning
Top . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.3 Stabilization of the Inverted Equilibrium of the 3D Axially Symmetric
Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.4 Attitude Control of an Underactuated Geo-Stationary Spacecraft . . 75
5.5 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
II MEO and LEO Spacecraft Attitude Control 87
6 Medium and Low Earth Orbit Satellites 89
6.1 LEO Spacecraft Attitude Model . . . . . . . . . . . . . . . . . . . . 90
6.2 LEO Spacecraft Position Model . . . . . . . . . . . . . . . . . . . . 94
6.3 LEO Spacecraft Control Problem Formulation . . . . . . . . . . . . 95
6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7 Attitude Control of a Low Earth Orbit Satellite 100
7.1 Command Governor Approach for LEO Spacecraft Control . . . . . 101
7.2 Simulation of a LEO Spacecraft Reconguration Maneuver . . . . . 106
7.3 Simulation of a LEO Spacecraft Tracking Maneuver . . . . . . . . . 108
7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
8 Attitude Control of a LEO Satellite for Large Angle Maneuvers 116
8.1 LEO Spacecraft Nonlinear Attitude Model . . . . . . . . . . . . . . 117
8.2 Hybrid Command Governor for LEO spacecraft . . . . . . . . . . . 120
8.3 Simulations of a LEO spacecraft performing Large Angle Maneuvers 123
8.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Conclusion 128
Geo-stationary Spacecraft and the 3D Pendulum . . . . . . . . . . . . . 129
Constrained MEO and LEO Spacecraft . . . . . . . . . . . . . . . . . . . 131
Research Contribution and Future Developments . . . . . . . . . . . . . 132
Bibliography 134
CONTENTS iii
List of Figures
1.1 Inertial references frame I and body reference frame B . . . . 10
1.2 Sequence of Eulers angles to be applied to frame I for obtaining
frame B orientation . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 A schematic of a 3D Rigid Pendulum . . . . . . . . . . . . . . . . . 14
1.4 Triaxial Air Bearing Testbed. . . . . . . . . . . . . . . . . . . . . . 20
2.1 A schematic of a 3D Rigid Pendulum with one axis of symmetry . . 23
2.2 Classical representation of the 2D spherical pendulum . . . . . . . . 25
2.3 A schematic of a heavy spinning top not constrained by the table . 27
3.1 Fan thrusters scheme of the Triaxial Attitude Control Testbed. . . . 37
3.2 Experimental results for the evolution of the angular velocity of the
3D pendulum in the body frame. . . . . . . . . . . . . . . . . . . . 45
3.3 Experimental results for the evolution of the components of the di-
rection of gravity in the body frame. . . . . . . . . . . . . . . . . 45
3.4 Experimental result for the evolution of the angle between the re-
duced attitude vector (t) and the desired reduced attitude vector
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1 Evolution of the closed loop system (3.2), (4.1) and (4.2) towards
the unstable equilibrium. . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Evolution of the angular velocity of the 3D pendulum in the body
frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3 Evolution of the components of the direction of gravity in the
body frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 Evolution of the angle between the reduced attitude vector (t) and
the desired reduced attitude vector
0
. . . . . . . . . . . . . . . . . 58
4.5 Vertical angular momentum (solid line) and total energy (dash line)
of the 3D pendulum. . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.6 Motion of the vector between the pivot and the center of mass of
the 3D pendulum in the inertial frame. . . . . . . . . . . . . . . . . 60
0
for (0) = 0. . . . . . . . . . 60
5.1 Evolution of the angular velocity of the symmetric heavy top in the
body frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
body frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3 Vertical angular momentum (solid line) and total energy (dash line)
of the symmetric top. . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4 Closed-loop trajectory of the symmetric top in the inertial frame. . . 82
5.5 Evolution of the angular velocity of the 2D pendulum in the body
frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
body frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
0
for (0) = 0. . . . . . . . . . 84
5.8 Closed-loop trajectory of the 2D pendulum in the inertial frame. . . 84
6.1 Reference frame B motion with respect to an Earth centered ref-
erence frame ECI . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.2 Reference frame B with respect to an orbiting reference frame
O, both moving in ECI . . . . . . . . . . . . . . . . . . . . . 92
7.1 Command Governor scheme. . . . . . . . . . . . . . . . . . . . . . 101
7.2 Relative position components x(k), y(k), z(k) of the LEO satellite,
under the proposed LQ+CG control law (7.14). . . . . . . . . . . . 108
LIST OF FIGURES v
7.3 Eulers angles (k), (k), (k) of the LEO satellite, under the pro-
posed LQ+CG control law (7.14). . . . . . . . . . . . . . . . . . . 108
7.4 Input forces f
i
(t) generated by the proposed LQ+CG control law
(7.14). The dash lines represent the constraint boundaries. . . . . . 109
7.5 Input torques
i
(7.14). The dash lines represent the constraint boundaries. . . . . . 109
7.6 Relative positions x(k), y(k), z(k) of the LEO satellite under LQ
control law only. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.7 Eulers angles (k), (k), (k) of the LEO satellite under LQ control
law only. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.8 Input forces f
i
(t) required by the LQ control law. The dash lines
represent the constraint boundaries. . . . . . . . . . . . . . . . . . 111
7.9 Input torques
i
represent the constraint boundaries. . . . . . . . . . . . . . . . . . 111
7.10 Evolution of the relative position variables x, y and z of the LEO
satellite with the application of LQ+CG (tracking problem) for a
period of revolution. . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.11 Errors e
i
(t), i=1,2,3 in coordination accuracy for the position com-
ponents x(k), y(k) and z(k) related to the circumference tracking
maneuver under (7.14). The dash lines represent the constraint
boundaries ([e
i
[
max
= 0.1). . . . . . . . . . . . . . . . . . . . . . . 113
7.12 Values of (t) and (t) for the CG selection logic (7.14) correspond-
ing to the maneuver. . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.13 Trajectory in the (x, y) plane under (7.14) (left) in comparison with
reference signal (right). . . . . . . . . . . . . . . . . . . . . . . . . 114
7.14 Errors e
i
(t) in coordination accuracy for the position components x,
y and z related to the circumference tracking maneuver under LQ
only. The dash lines represent the constraint boundaries ([e
i
[
max
=
0.1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.1 Set of reference frames O
i
and leader orbiting reference frame
C. At any time instant, there exists at least one O
i
satisfying
the small angular displacement assumption with respect to B. . . 118
8.2 Switching logic for attitude reconguration. . . . . . . . . . . . . . 123
8.3 Reconguration for a LEO spacecraft, requiring a large angle maneuver.124
8.4 Relative position y(t) (desired reference in dash lines), control force
f
2
(t) and error e
2
(t) in coordination accuracy for y(t) (constraint
boundaries in dash lines) under HCG. The vertical dash lines indicate
instants of switching. . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.5 Eulers angle (t) (desired reference in dash lines), control torque
3
(t) and error e
3
(t) in coordination accuracy for (t) (constraint
boundaries in dash lines) under HCG. The vertical dash lines indicate
instants of switching. . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.6 Values of (t) and (t) for the HCG selection logic (7.14), (8.7). . 125
8.7 Relative position y(t), control force f
2
(t) and error e
2
(t) in coor-
dination accuracy for y(t) under LQ (constraint boundaries in dash
lines). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
LIST OF FIGURES vii
Introduction
Attitude control has been one of the most interesting and challenging problems
in aerospace systems, from the rst half of the twentieth century, to present
time. A modern spacecraft is a multibody system, composed of rigid bodies
connected together by exible rods. From a mechanical point of view and with
the aim of controlling its orientation, however, it can be modeled as a single rigid
body. As a consequence, a spacecraft can be thought as a mechanical system
with three translational and three rotational degrees of freedom. Moreover, if
the satellite orbits at an altitude of approximately 35.786 km above Earths
surface, it appears stationary. Thus, only rotational motion is observed from
Earth.
Behavior of space vehicles on Medium (MEO) and Low Earth orbits (LEO)
is quite dierent from the one at GEO altitudes. On a GEO orbit, dynamics
is described by attitude models. Moreover, gravity does not aect rotational
motion. At lower altitudes, on the contrary, also translational motion, and its
eects on rotational motion, have to be taken into account. Indeed, dierent
positions may correspond to dierent angular velocities, producing changes on
attitude.
The main goal of this dissertation is modeling and controlling the attitude of
spacecraft, orbiting around Earth at GEO, MEO and LEO altitudes. Although
on GEO orbits, gravity does not aect motion, we generalize the former case,
introducing a new mechanical system with three rotational degrees of freedom
in uniform gravity. Such a system, consists of a rigid body constrained to
INTRODUCTION 1
rotate about a pivot. We refer to it as 3D Pendulum. Furthermore, we study
attitude dynamics and control of rigid bodies traveling on MEO and LEO orbits.
To do that, we introduce a mathematical model describing both rotations and
translations.
Prior Literature on Spacecraft Attitude Control
The idea of using satellites for communication purposes, goes back to 1945,
when the science ction author Arthur C. Clarke, proposed a geo-stationary or-
bit where placing a mechanical system, suitably oriented with respect to Earth.
From that time, a huge literature has been produced on the several aspect
related to the motion of space vehicles, as orbital motion and maneuvers, per-
turbations, and attitude acquisition. Related theory has been summarized in
numerous text books on spacecraft motion and control. Two classical texts
are mainly referred in this dissertation. The rst one, has been published by
Kaplan [Kap76] in the seventies. It analyzes spacecraft dynamics. The second
one, written by Huges [Hug86] during the subsequent decade, is more focused
on attitude. From a more technological perspective, especially related to space
vehicles moving on MEO and LEO orbits and to their instrumentations, promi-
nent is the text recently published by Sidi [Sid00]. Astrodynamics, on the other
hand, are extensively described in [BMW71]. In particular, it illustrates in de-
tail the classical two body and three body problems. Dynamics on MEO and
LEO orbits are strictly connected to this theory.
In this thesis, we study attitude control problems. Beyond the peculiar prop-
erties of space systems, listed above, we focus our attention on the mechanical
properties of rigid bodies. From this point of view, related theory is especially
connected to classical mechanics and geometry. Therefore, attitude models de-
scribed by Goldstein et al. in [GPS80] and geometric properties presented by
Marseden et al. in [MR99] and [MRS00] are fundamental. The literature on
rotating mechanical systems is wide. One and two degrees of freedom rotating
rigid bodies have been analyzed in several works, giving rise to pendulum mod-
els. They constitute fundamental background for studying more sophisticated
structures, as three degrees of freedom rotating rigid bodies. A comprehensive
survey on pendulum literature has been presented by Furuta in [Fur03]. Related
references cover the evolution of corresponding theory. Prominent is the work
made by Shirieav et al. in the nineties. [SEL99], [SPLE00] and [SEL04] propose
control techniques for 2D pendulum, using conserved quantities of the motion
and passivity-based results. We use a similar approach in this dissertation. In
particular, we refer to a 3D rotating rigid body and we use reduction theory
for modeling the system. Then, we use input-output passivity, for controlling
attitude. Reduction theory is described in [MRS00]. Its main aspects have
been developed by Bloch et al., for controlling mechanical systems [BLM00],
[BLM01]. Connections to the 2D pendulum and to the spinning heavy top,
have been illustrated by Simo et al. in [SLRM92], by Wan et al., [WCB95],
[WCB95], and by related references. Passivity theory has been extensively de-
scribed by Ortega et al. in [OSR98].
Passivity control, lies in the more general context of energy-based control. In
this thesis we make extensive use of energy-based control and Lyapunov func-
tions. They are particularly suitable in the presence of mechanical systems.
Ideas on suitable Lyapunov functions for rotating 3D rigid bodies have been
provided by Burkov. In [Bur95], the author proposes a Lyapunov function for
controlling the attitude of a spacecraft, with three control torques. On the other
hand, control techniques based on the use of only two input torques, have been
summarized by Doumtchenko and Tsiotras in [DT00]. Tsiotras et al. have pro-
vided prominent work on modeling and controlling underactuated spacecraft.
See [CLT00], [WTCB94] and related references.
Recent theory in spacecraft attitude dynamics and control, has been focusing
its attention on small satellites placed on MEO and LEO orbits. For Earth ob-
serving missions and interferometric systems, these structures are more suitable
than GEO spacecraft. If large instrumentation is required, the use of distrib-
uted systems, shared between vehicles ying in formation, has been suggested.
A survey on this new theory has been illustrated by Esper et al. in [ENS
+
03].
Suitable models and control techniques have been analyzed for such systems.
Due to their small dimension, constraints play a key role on their dynamics.
On this subject, Veres et al. present a Model Based Predictive Control (MPC)
technique for a eet of two LEO nano-satellites, in [VGRM01]. In this disser-
INTRODUCTION 3
tation we propose an attitude control technique also based on MPC. It consists
on a suitably modied version of the Command Governor (CG) approach, de-
veloped by Bemporad et al. [BCM97] and subsequently adopted in [CMA00]
and [ABC
+
03].
Dissertation Outline
We treat attitude models and control for rigid spacecraft. Since, dynamics of
rigid bodies on GEO orbits are dierent from dynamics on LEO orbits, this
dissertation can be broadly divided into two parts. In the rst part, we face the
problem of modeling and controlling attitude of geo-stationary satellites. Apart
of the presence of gravity, we notice immediately that their motion can be re-
produced on Earth, pivoting a rigid body on a xed point. Thus, in Chapter 1
we introduce the new concept of 3D Pendulum. It consists exactly of a three
rotational degrees of freedom rigid body, whose translations are obstructed by
the pivot. Its equilibria and their stability properties are described as well.
Actually, the 3D pendulum model generalizes GEO spacecraft attitude models.
Moreover, it generalizes 2D pendulum, 1D pendulum and spinning top mod-
els. These aspects are developed in Chapter 2. Assumptions to be placed on
the 3D pendulum model, for obtaining the mathematical description of these
classical systems, are analyzed. Furthermore, advantages on using the 3D pen-
dulum model, for controlling the equilibria of 2D pendulum and spinning top,
in comparison to prior techniques, are motivated. A new mechanical system,
following directly from the 3D pendulum model, is illustrated in Section 2.3.
It represents a heavy top, spinning below its pivot. This system has not been
studied in previous literature.
Subsequently, we face the problem of stabilizing the natural equilibria of a 3D
pendulum. They consist of the two congurations where direction of gravity is
vertical, with respect to the rigid body. In Chapter 3, we propose passivity-
based control laws for stabilizing these congurations in the general case, i.e.
when the 3D pendulum is asymmetric. Techniques for stabilizing arbitrary ori-
entations of the 3D pendulum are illustrated in Chapter 4. They represent
attitude controllers that can be directly used to stabilize any attitude of geo-
stationary spacecraft.
The case when the 3D pendulum is axi-symmetric, is illustrated in Chapter 5.
In this case, the control laws we propose, actually stabilize the 2D pendulum
and the spinning top. In particular, in Section 5.3 we show that our approach
guarantees the use of a continuous state-feedback for swinging up the spherical
pendulum. It is a prominent result, if compared to previous literature, where
this problem is solved by a swing-up controller, singular at the desired equilib-
rium, a local stabilizing controller, and a technique for switching between the
two.
Attitude control of MEO and LEO spacecraft is faced in the second part of this
dissertation. In Chapter 6 we describe the main properties of space vehicles
orbiting around Earth at constant velocity, under a leader-following approach.
The subsequent model, describing both rotational and translational motion, is
presented. Model Based Predictive Control techniques, under the assumption of
small angular displacements between the satellite and a suitably dened orbital
leader reference frame, are illustrated in Chapter 7. They adopt a Command
Governor approach, for taking the presence of saturation constraints into ac-
count. Indeed, satellites moving on low orbits are usually small. Thus, their
thrusters cannot provide large torques. Moreover, we introduce suitable state-
accuracy constraints, usually present in real missions. The importance of con-
sidering constraints is the main dierence between space systems on high orbits,
as geo-stationary spacecraft, and space systems on low orbits. In this disser-
tation this dierence is the watershed between the rst and the second part.
Finally, in Chapter 8, we remove the small angular displacements assumption.
In order to restore results illustrated in the previous chapter, we address our
control technique in a hybrid framework, proposing a Hybrid Command Gover-
nor controller. It consists of a bank of controller as those presented in Chapter
7, each one designed with respect to a reference attitude, and of a technique for
switching between them.
INTRODUCTION 5
Part I
Geo-Stationary Spacecraft
Attitude Control
7
Chapter
1
Dynamics of the 3D Pendulum
Three degrees of freedom (3D) rigid bodies have been extensively studied in me-
chanics and their dynamics equations are well known in literature. They provide
the fundamental background for modeling and controlling modern space and un-
derwater vehicles. Simply, a rigid body is a system of particles whose relative
distances are xed with time.
In this chapter we introduce a particular kind of 3D rigid body: the 3D pendu-
lum. Its dynamics is strictly related to the one of spacecraft in geo-stationary
(GEO) orbits. Pendulum models have provided a rich source of examples in
nonlinear dynamics and, in recent years, in nonlinear control. Several dierent
models have been developed to analyze the properties of the uncontrolled planar
1D pendulum and spherical 2D pendulum. However, their extension to a three
degrees of freedom pendulum, supported by a xed frictionless pivot, has never
surprisingly been performed. Indeed, very few publications view the pendulum
as a rigid body.
We rst remind the main concepts describing 3D rigid body attitude models.
This provides appropriate background for the subsequent introduction of the
3D pendulum. Further, we present both the complete mathematical model of
this system and a simplied version of it, particularly suitable for control pur-
9
poses. We analyze the stability properties of the 3D pendulum, as well. Finally,
we introduce a laboratory 3D pendulum developed at the Attitude Dynamics
and Control Laboratory, Department of Aerospace Engineering, University of
Michigan: the Triaxial Attitude Control Testbed (TACT). It will be useful in
the following chapter to verify the models and their stability properties and to
assess the performance of control laws proposed in the rst part of this disser-
tation.
1.1 Rigid Bodies Mathematical Models
Equations of motion of a 3D rigid body can be determined using standard ar-
guments [GPS80], [KK73]. They are composed by the dynamics and kinematics
equations describing the rate of change of the orientation of an inertial frame
I = X, Y, Z with respect to a body-xed three versors right-oriented set
B = X
, Y
, Z
.
Figure 1.1. Inertial references frame {I} and body reference frame {B}
The dynamics equations concern the eects of forces on the motion of the body.
They can be simply obtained representing the time evolution of the momentum
P of the body-xed reference frame in terms of its angular velocity. Briey, the
time derivative of the vector P in B can be written as
dP
dt
I
=
dP
dt
B
+ P (1.1)
where is the angular velocity of B in I. Now,
dP
dt
B
= J
d
dt
= J
with J denoting the constant inertia matrix in the body-xed reference frame,
and
dP
dt
I
=
d
dt
(r mv) = m(
dr
dt
v) +m(r
dv
dt
) = r
d(mv)
dt
= (r F) =
with m representing the total mass of the rigid body, r the position of its center
of mass in I and a possibly applied torque. Thus,
J = J + (1.2)
Equation (1.2) is referred to as the Eulers equation and represents the dynamics
of the rigid body in the general case. It is usually a simplifying assumption to
select B aligned with the principals axes of the body, or equivalently, the mo-
mentum P parallel to the angular velocity . In this case, all the extra-diagonal
terms of J are null and the system reduces to three scalar equations. Details
can be found in [GPS80], [Kap76] and [Sid00].
Kinematics equations complete the mathematical model of the 3D rigid body.
They describe the geometry of the motion apart from its causes. Many dierent
kinematics models have been proposed in mechanics literature and aerospace
control. An extensive overview of these models with focus on their main dier-
ences can be found in [Hug86]. This text has also strongly inspired the connec-
tions between space vehicles control and the 3D pendulum control described in
the following chapters. In this dissertation we make use of two of these models.
The rst one, namely the Poissons model, aims at describing the orientation of
the inertial frame I in B, simply expressing its versors in terms of the B
coordinates X
, Y
and Z
. It is useful to control the attitude of satellites in

GEO orbits. For this reason, it will be used hereafter to describe the 3D pen-
dulum. The second model uses the Eulers angles equations and it is suitable to
control the attitude of space vehicles in medium (MEO) and low (LEO) orbits.
It will be widely described in the second part of this dissertation.
The kinematics model arises from the representation of I in B by a suitable
rotation matrix R
I = RB (1.3)
11
It follows from the fact that the conguration manifold of a rotating body lies
in
1
SO(3) R
3
, in the general case [MR99], being SO(3) the group of the
orthogonal matrices with determinant equal to 1:
SO(3) = R (3) : RR
T
= I, det(R) = 1
As described in detail in [Hug86] and in [Hal02], equation (1.3) is not a proper
matrix product, but a notation commonly referred to as the vectrix represen-
tation. The elements of R represent the cosines directors. Indeed, the ori-
entation of any axes of I can be expressed as a linear combination of the
B versors: X = R
11
X
+ R
12
Y
+ R
13
Z
, Y = R
21
X
+ R
22
Y
+ R
23
Z
and Z = R
31
X
+ R
32
Y
+ R
33
Z
where X
= [1, 0, 0]
T
, Y
= [0, 1, 0]
T
and
Z
= [0, 0, 1]
T
in B. Now, applying equation (1.1) to the rst axis of I,
gives
X = X = X (1.4)
and analogous for Y and Z. Here is the skew-symmetric matric denoting the
cross products between two vectors, such that a b = ab. It corresponds to
a =
_
_
0 a
3
a
2
a
3
0 a
1
a
2
a
1
0
_
_
(1.5)
Hence, rewriting (1.4) in terms of (1.3), we have directly
R = R (1.6)
that is the kinematics equation of the rigid body.
The Poissons representation is originated by writing the above expression for
any component. However, this representation is redundant because of the six
orthogonality constraints. In fact SO(3) is a sub-variety of dimension 3 of R
9
[MR99]. The, three variables, corresponding to the three degrees of freedom
of the rigid body are enough for describing its orientation in space. Several
mathematical models have been developed in order to characterize R using a
lower number of variables. The most common is the Eulers angles kinematic
model. It describes R in terms of the angles corresponding to three subsequent
1
The properties of SO(3) aecting control actions on rigid bodies will be discussed in the
following chapters. More details on the topology of SO(3) can be found in [MR99] and [Hat01].
rotations to apply to the inertial frame, to become coincident with the body
frame. In this work we consider three successive elementary rotations dened
by the yaw angle , the pitch angle and the roll angle . First, impressing o
Figure 1.2. Sequence of Eulers angles to be applied to frame {I} for obtaining
frame {B} orientation
rotation of angle to the Z axis of I, we have a new reference frame I
:
I
= R
I =
_
_
cos() sin() 0
sin() cos() 0
0 0 1
_
_
I
Subsequently, we can impress to this new reference frame a second rotation of
pitch angle giving another frame I
and then a third rotation of roll angle

to I
, obtaining B = R
I. Thus, R in (1.3) corresponds to

R = R
T
R
T
R
T
=
_
_
cc sc +css ss csc
sc cc sss cs +ssc
s cs cc
_
_
(1.7)
where cx stands for cos(x) and sx stands for sin(x). Equation (1.6) can then
be written in terms of Eulers angles taking the rst derivative of (1.7) and
extracting

,

and

. It produces three rst order dierential equations
=
y
sinsec +
z
cos sec (1.8)
=
y
cos
z
sin (1.9)
=
x
+
y
sintan +
z
cos tan (1.10)
known as the Eulers angles representation. Note that these equations are valid
as long as the pitch angle avoid the singularities at /2.
Model (1.2), (1.6) and some variations will be used in the rst part of this thesis
in order to illustrate the properties of the 3D pendulum and its connections to
13
previous literature. It is particularly suitable for controlling the attitude of
GEO spacecraft. On the contrary, model (1.2), (1.8)-(1.10) is more convenient
for attitude control of MEO and LEO satellites. It will be mostly used in the
second part of this thesis.
1.2 The 3D Pendulum
A pendulum is a rigid body supported by a xes, frictionless pivot, acted on by
gravitational forces and disturbance and control forces or moments. In the most
general case, we consider a pendulum without axes of symmetry. A schematic
of a rigid pendulum is shown in Figure 1.3. The supporting pivot allows three
Figure 1.3. A schematic of a 3D Rigid Pendulum
degrees of rotational freedom to the pendulum. Uniform constant gravity is
assumed. It represents the element in equation (1.2). Thus, the pendulum is
described by at least three spacial coordinates and three rotational coordinates,
suggesting the terminology 3D pendulum.
Two coordinate frames, are of interest. The inertial reference frame I has
its origin at the pivot, the rst two coordinate axes lie in the horizontal plane
and the third coordinate axis is vertical in the direction of gravity. The origin
of the body frame B is also located at the pivot. In this frame the inertia
matrix J is constant and can be simply computed from the moment of inertia
of a translated coordinate frame whose origin is located at the center of mass of
the pendulum, using the parallel axis theorem [HRW04]. The body-xed vector
from the pivot to the center of mass of the pendulum is denoted by . The
symbol g denotes the constant acceleration due to gravity. This is the data on
which the equations of motion are based. The dynamics are given by the Eulers
equation (1.2) that includes the moment due to gravity:
J = J +mg R
T
e
3
(1.11)
where m is the total mass of the rigid body and e
3
is the third unit vector in
the inertial coordinate frame, namely e
3
= (0, 0, 1)
T
. Equation (1.11) and the
rotational kinematics equation (1.6) dene the full dynamics of a rigid pendulum
in SO(3) R
3
.
It is important to point out that in GEO aerospace applications the dynamics
are not aected by gravity terms [Kap76]. Hence, these systems are special
cases of the above model. In particular, this is the case when the center of mass
of the rigid body is located at the pivot, that is = 0. In the context of the
rigid pendulum, we refer to this situation as the balanced case. It allows at
viewing the geo-stationary spacecraft attitude control problem as a simplied
version of the 3D pendulum attitude control problem. Therefore, the focus on
what follows, will be on the more general unbalanced case, viz. ,= 0.
There are two conserved quantities for the rigid pendulum. The rst one is
the total energy, which is the sum of the rotational kinetic energy and the
gravitational potential energy. In addition, there is a rotational symmetry of the
equations of motion corresponding to the group of rotations about the vertical
line through the pivot. These two results are summarized as follows.
Proposition 1.1 The total energy
E = 0.5
T
J mg
T
R
T
e
3
and the component of the angular momentum vector about the vertical axis
through the pivot
h =
T
JR
T
e
3
are constant along motions of the rigid pendulum.
The proof follows by showing that the time derivative of the total energy and the
time derivative of the angular momentum component are identically zero. This
can be shown using the dynamics equations (1.11) and the kinematics equation
(1.6).
15
Thanks to the presence of a conserved quantity along the motion, it is possible
to obtain a lower dimensional reduced model for the pendulum. In practice,
the function h can be seen as a constraint imposing the system to lie in a lower
dimensional manifold. It allows at expressing one of the variables of the motion
with respect to the others. A comprehensive explanation of the constants of the
motion for a dynamic system and their eects on the conguration manifold can
be found in [MR99]. Here, we introduce a model reduction by exploiting the
fact that the dynamics and kinematics can be written in terms of the reduced
attitude vector = R
T
e
3
, which is the unit vector that expresses the gravity
direction in the body-xed frame B.
Proposition 1.2 The dynamics of the 3D pendulum given by (1.11) and (1.6)
induce a ow on S
2
R
3
given by the reduced dynamics
J = J +mg (1.12)
and the reduced kinematics
= (1.13)
The proof follows from the denition of the reduced attitude vector and
demonstration that || = 1.
Equations (1.12) and (1.13) are in non-canonical form, but they are very useful
for studying the reduced dynamics of the rigid pendulum. Thus, they are used
in this dissertation, in place of the full dynamical model. It is worth to point
out that the reduced model describes the orientation of the 3D rigid pendulum
modulo a rotation along its third axis Z
. However, this is not a limitation in

spacecraft control problems, where usually, the task is of controlling the direc-
tion of an axis, e.g. represented by an antenna, instead of the complete attitude.
Finally, equations (1.12) and (1.13) can also be expressed in terms of the Eulers
angles noticing that
(, ) =
_
_
sin
sincos
cos cos
_
_
(1.14)
In fact, equations (1.9)-(1.10) and (1.13) using (1.14) are exactly an expression
of the reduced dynamics of motion of the rigid pendulum on S
2
R
3
expressed
in terms of the Eulers angles. Further, the kinematics can be shown to be
independent from the yaw angle [MR99].
1.3 Equilibria of the 3D Pendulum
To further understand the dynamics of the 3D pendulum, we study its equilibria.
To simplify the analysis it is convenient to choose the body-xed frame so that
its third axis Z
is aligned with the vector from the pivot to the center of mass.
Thus, in B, = (0, 0, l)
T
, where l > 0. Consequently, the gravity terms that
appear in equation (1.11) can be considerably simplied.
The equilibrium solutions for the 3D pendulum satisfy = 0, so that the
attitude of the pendulum is constant. Substituting = 0 in equation (1.11), we
obtain
R
T
e
3
= 0
The above equation implies that the 3D pendulum is in equilibrium only if
the center of mass vector is collinear with the gravity vector R
T
e
3
in the
body-xed coordinate frame. This implies that
R
T
e
e
3
= e
3
where = 1. For = 1, the center of mass vector points in the direction
of the gravity vector. Similarly, for = 1, the center of mass vector points
in the direction opposite to the gravity vector. Now, let R
e
denote an attitude
rotation that solves equation (1.11) and dene
e
= R
T
e
e
3
. Then, any attitude
in the conguration manifold given by
R SO(3) : R = R
e
Q(
),

arbitrary
denes an equilibrium attitude corresponding to = 0 where
Q(
) =
_
_
cos

sin

0
sin

cos

0
0 0 1
_
_
Hence, if R
e
denes an equilibrium attitude for the 3D pendulum, then a rota-
tion of the pendulum about the gravity vector by an arbitrary angle is also an
17
equilibrium. Consequently, in SO(3) there are two disjoint equilibrium mani-
folds of the 3D pendulum. We dene the manifold corresponding to = 1 in the
above description as the hanging equilibrium manifold, since the center of mass
is below the pivot for each attitude in the manifold. The manifold correspond-
ing to = 1 in the above description is dened as the inverted equilibrium
manifold, since the center of mass is above the pivot for each attitude in the
manifold.
Next, we study the equilibria of the reduced equations (1.12) and (1.13). As
before, the equilibria of the system satises = 0. Substituting in (1.12), we
get
= 0.
Thus, we obtain two isolated equilibrium solutions of the reduced equations in
S
2
R
3
given by = e
3
, = 0, and by = e
3
, = 0. Alternatively, the
two equilibrium solutions can be expressed using Eulers angles as = 0, = 0,
= 0 and = 0, = , = 0. For each of these representations of the
equilibrium solutions, the rst corresponds to the hanging equilibrium where
the center of mass is below the pivot and the second corresponds to the inverted
equilibrium where the center of mass is above the pivot.
The stability of the two isolated equilibrium solutions of the reduced equations
of motion is readily assessed. The stability of the two families of equilibrium
manifolds of the full equations of motions follows. The stability analysis can
then be based on the reduced model.
Proposition 1.3 The hanging equilibrium of the reduced dynamics of the 3D
pendulum described by equations (1.12) and (1.13) is locally stable in the sense
of Lyapunov.
Proof. The hanging equilibrium corresponds to the conguration
0
= 0 and
0
= (0, 0, 1)
T
. Consider the Lyapunov function
V (, ) =
1
2
T
J +mg(l
T
)
V (
0
,
0
) = 0 and V (, ) > 0 elsewhere. The derivative of V along the
trajectories is
V =
T
J mg
T

=
T
(J +mg ) mg
T
( )
=
T
mg mg
T
= mg(
T

T
) 0
where the last equality follows from the mixed vector product property [HRW04].
Hence, V is a continuously dierentiable function, positive denite and such
that V (
0
,
0
) = 0 and
2

V (, ) 0 (, ). Thus, the hanging equilibrium
is locally Lyapunov stable [Kha01].
Proposition 1.4 The inverted equilibrium of the reduced dynamics of the 3D
pendulum described by equations (1.12) and (1.13) is locally unstable.
Proof. The hanging equilibrium corresponds to the conguration
0
= 0 and
0
= (0, 0, 1)
T
. Consider the Lyapunov function
V (, ) =
1
2
T
J +mg(l +
T
)
V (
0
,
0
) = 0 and V (, ) > 0 elsewhere. The derivative of V along the
trajectories is
V =
T
J +mg
T

=
T
(J +mg ) +mg
T
( )
=
T
mg +mg
T
= 2mg
T
( )
where the last equality follows from the mixed vector product property [HRW04].
V is not denite. Hence, V is a continuously dierentiable function such that

V (
0
,
0
) = 0 and V (, ) > 0,

V (, ) > 0 for some points (, ) arbitrarily
close to (
0
,
0
). Thus, by Chetaevs theorem [Kha01], the inverted equilibrium
is unstable.
As a consequence of the two propositions above, one can conclude that the
hanging equilibrium manifold of the full dynamics of the 3D pendulum is stable
in the sense of Lyapunov and that the inverted equilibrium manifold of the full
dynamics of the 3D pendulum is unstable in the sense of Lyapunov.
1.4 Triaxial Attitude Control Testbed
The test-bed used for experiments in the rst part of this dissertation is based
on the spherical air bearing manufactured by Space Electronics, Inc., Berlin,
2
Notice that the result

V 0 could be expected, since V denes the total energy of the
system, that is a constant of the motion, as described in section 1.2.
19
CT, depicted in Figure 1.4. It is available at the Attitude Dynamics and Con-
trol Laboratory, Department of Aerospace Engineering, University of Michigan.
We refer to this systems as the Triaxial Attitude Control Testbed (TACT).
Figure 1.4. Triaxial Air Bearing Testbed.
An aluminum sphere of diameter 11 inch oats on a thin lm of air that exits
holes located in the surface of the cup. Air at high constant pressure is supplied
to the cup by means of a hose that passes through the center of the vertical
support. Once the main components are mounted, additional masses can be
added to modify the nal mass distribution. Hence, the center of mass can
be balanced at the rotational center (no eects from the gravity), in order to
simulate a geo-stationary spacecraft, or in any other position, representing a 3D
unbalanced rigid pendulum. The spherical air bearing guarantees low friction
and three-dimensional motion with unrestricted roll and yaw and a maximum
45 degrees pitch angle.
A comprehensive description of the hardware related to the TACT has been pro-
vided by [BMB01]. Mathematical models for the TACT have been introduced in
[CSMB01] and [CSM03]. There, a description equivalent to (1.11) and (1.6), is
obtained by means of the Lagrangian equations. This geometrical approach to
mechanical systems, well known in the literature, is widely described in [MR99]
and [MRS00].
As illustrated in Figure 1.4, a one-piece shaft passes through the center of the
sphere and extends between a pair of circular mounting plates. Extension shafts
connect the circular mounting plates to larger square mounting plates. Circular
plates are used for supporting hardware instrumentation, while square plates
are used for supporting fan thrusters, batteries and additional weights. In-
deed, weights can be attached to the TACT, thanks to access holes cut into the
plates. This allows at modifying the inertia matrix J of the TACT, before per-
forming experiments. Once the entire system is set up, the procedure described
in [CBA
+
06] can be adopted, to identify the inertia matrix of the system.
1.5 Conclusions
In this chapter we have introduced the new concept of 3D rigid pendulum.
It represents a three rotational degrees of freedom rigid body, supported by a
frictionless pivot. We have presented the 3D pendulum model in the general
unbalanced case. The special balanced case, corresponding the absence of grav-
ity terms in the system, encompasses the geo-stationary spacecraft dynamics
model. Therefore, we have addressed the problem of controlling the attitude
of GEO satellites, as a simplied version of the 3D pendulum attitude control
problem.
21
Chapter
2
Special Cases of the 3D Rigid
Pendulum
The published literature on pendulum models is very large, since they are use-
ful for both pedagogical and research reasons. Indeed, they represent physical
mechanisms that can be viewed as simplied versions of mechanical systems
that arise in robotics and spacecraft. Standard pendulum models are dened
by a single rotational degree of freedom, referred to as a planar rigid pendulum,
or two rotational degrees of freedom, referred to as a spherical rigid pendulum.
Control problems for planar and spherical pendulum models have been studied
in many references from the seventies [MNF76], to the last decade. [BS96],
[GD99], [AF00] and [Ang01] are only a few examples of the huge literature on
this subject. A comprehensive survey on this literature is presented in [Fur03].
Prominent is the work made by Shirieav et al. in the last decade: [SEL99],
[SPLE00] and [SEL04]. In these references, some geometrical concepts are ap-
plied to a spherical pendulum, suggesting passivity-based control laws [OSR98].
In this chapter we show that the classical mathematical models of the 1D pen-
dulum, the 2D pendulum and the heavy spinning top, can be simply viewed as
special cases of the 3D pendulum model described in Chapter 1. We rst intro-
duce assumptions to be placed on model (1.12), (1.13) for obtaining a spherical
pendulum model and then a planar pendulum model. Further, we show that,
with simple assumptions, the 3D pendulum model can describe the heavy top
dynamics. The corresponding dynamics seem not to have been previously stud-
ied. Finally, we study the stability properties of this new spinning top model,
illustrating connections with the classical heavy top problem, widely discussed
in [GPS80].
2.1 Axisymmetric 3D Rigid Pendulum
We select the body-xed reference frame B such that its versors X
, Y
and
Z
are aligned with the principal axes of the 3D rigid body. This renders the
extra-diagonal terms of the inertia matrix J null. In this case, the diagonal
terms of J are referred to as the principal moments of inertia. Hence, model
(1.12) using (1.14), reduces to three scalar equations, as described in Chapter
1.
Moreover, we consider the axisymmetric case, where two of the principal mo-
ments of inertia of the pendulum are identical and the pivot is located on the
axis of symmetry of the pendulum. This is an usual assumption while facing
Figure 2.1. A schematic of a 3D Rigid Pendulum with one axis of symmetry
attitude control problems for underactuated geo-stationary spacecraft [DT00].
In particular, we assume that the body-xed axes are selected so that J =
diag(J
t
, J
t
, J
a
) and = le
3
where l is a positive scalar constant. Consequently,
23
equation (1.12) using (1.14), can be written in scalar form, as
J
t

x
= (J
t
J
a
)
y
z
mgl cos sin
J
t

y
= (J
a
J
t
)
z
x
mgl sin
J
a

z
= 0
(2.1)
From the last equation in (2.1) we see that the component of the angular velocity
vector about its axis of symmetry is constant. This means that a constant
z
denes an invariant manifold for the pendulum dynamics. The resulting reduced
equations of motion, for constant
z
, can be represented by equations (2.1) and
(1.9)-(1.10)
J
t

x
= (J
t
J
a
)
y
z
mgl cos sin
J
t

y
= (J
a
J
t
)
z
x
mgl sin
=
y
cos
z
sin
=
x
+
y
sintan +
z
cos tan
(2.2)
Remember that the yaw motion of the pendulum is described by the kinematic
equation (1.8). Hence, the evolution of can be deduced from the evolution of
the other variables of motion.
Now, the special case
z
0 leads to an axisymmetric rigid body whom atti-
tude dynamics are described by two angles and two angular velocities, namely,
a 2D spherical pendulum.
Proposition 2.1 Assume the rigid pendulum has a single axis of symmetry and
the pivot is located on the axis of symmetry of the pendulum. The equations of
motion of the rigid pendulum dene an induced ow on the manifold S
2
R
2
corresponding to
z
= 0, given by the equations
J
t

x
= mgl cos sin
J
t

y
= mgl sin
=
y
cos
=
x
+
y
sintan
(2.3)
These equations are said to represent a 2D spherical pendulum. They exhibit
singularities for = n/2.
It is worth to point out that the above representation is not the one commonly
used for describing the dynamics of the spherical pendulum. Indeed, in litera-
ture the 2D pendulum is not introduced as a rotating rigid body. Usually, the
mathematical description of the system follows from the denition of the two
angles q
1
and q
2
in I depicted in Figure 2.2 [SEL04]. There, q
1
is the angle
Figure 2.2. Classical representation of the 2D spherical pendulum
between the vertical and the axis of symmetry of the pendulum and q
2
is the
angle representing the displacement of the center of mass of the pendulum with
respect to axis X of frame I. Then, the model is obtained by writing the po-
sition of the center of mass in I coordinates and taking the second derivative
or, more simply, by writing the Lagrangian equations of the system
d
dt
(
L
q
i
)
L
q
i
= 0 (2.4)
Here, the Lagrangian
1
L is
L = T V =
1
2
ml
2
[ q
2
1
+ (sin
2
q
1
) q
2
2
]
where T and V are, respectively, the kinetic and the potential energy. Thus,
the model of the 2D spherical pendulum is
q
1
= sin q
1
(cos q
1
) q
2
2

g
l
sinq
1
q
2
= 2
cos q
1
sinq
1
q
1
q
2
(2.5)
This is the description commonly used for representing the dynamics of the 2D
spherical pendulum. The fundamental dierence with respect to model (2.3)
1
Lagrangian approach is widely used in mechanical systems modeling. An extensive de-
scription of this technique can be found in [GPS80], [HRW04] and many other references
discussing rational mechanics.
25
is that now the singularities are in q
1
= n, i.e. in the upward position.
This leads to several disadvantages while designing control laws for swinging
up the spherical pendulum, that is a basic problem in the literature related to
attitude control. For instance, in [SEL99] and [SEL04] the authors use model
(2.5) and they switch to a dierent model in the close proximity of the upward
position. It produces a discontinuous representation of the system with many
related diculties. As it will be seen in the following chapters, it could simply
be avoided, by selecting the more suitable description of the mechanical system
(2.3), arising from the interpretation of the 2D spherical pendulum as a rotating
rigid body.
Now, beyond the assumption above of
z
= 0, consider the case where
x
= 0
and = 0. It is straightforward to show that this leads to an invariant dynamics
of the axisymmetric rigid pendulum, described as follows:
the pivot is located on the axis of symmetry of the pendulum. The equations of
motion of the rigid pendulum dene an induced ow on the manifold S
1
R
1
corresponding to
x
= 0,
z
= 0, and = 0, given by the equations
J
t

y
= mgl sin
=
y
(2.6)
These equations are said to represent a 1D planar pendulum.
Thus, for an axially symmetric pendulum with the pivot located on the axis
of symmetry, the well known 2D spherical pendulum and 1D planar pendulum,
can be viewed as special cases of the 3D rotating rigid pendulum. On the other
hand, if the pendulum is asymmetric, then its dynamics are general, in the
sense that neither the spherical pendulum dynamics, nor the planar pendulum
dynamics, arise as special cases.
2.2 Spinning Axisymmetric 3D Rigid Pendulum
The 2D spherical pendulum has been shown to originate from the 3D axisym-
metric rigid pendulum, under the assumption of
z
0. The induced dynamics
corresponding to a nonzero value of
z
is fundamentally dierent from this case.
These dynamics seem not to have been previously studied.
Thus, consider again an axisymmetric pendulum with the pivot located on
its axis of symmetry. Its dynamics are described by model (2.2). Assume
z
= k ,= 0. Since the third equation in (2.1) holds true, we have the following
result:
the pivot is located on the axis of symmetry of the pendulum. For any constant
k ,= 0, the equations of motion of the rigid pendulum dene an induced ow on
the manifold S
2
R
2
corresponding to
z
= k, given by the equations
J
t

x
= (J
t
J
a
)k
y
mgl cos sin
J
t

y
= (J
a
J
t
)k
x
mgl sin
=
y
cos k sin
=
x
+
y
sintan +k cos tan
(2.7)
These equations represent the dynamics of a symmetric rigid body spinning
steadily around its axis of symmetry under the eects of gravity, namely, an
heavy spinning top [GPS80]. In particular, dierently from previous literature,
they extend the concept of heavy spinning top to the space under the table.
It is depicted in Figure 2.3.
Figure 2.3. A schematic of a heavy spinning top not constrained by the table
Thus, the motion of the symmetric top can be seen as a special case of the 3D
rigid pendulum dynamics.
Problems related to the symmetric spinning top have been extensively studied in
27
mechanics. A wide discussion on the properties of the motion, on the conditions
for precession and on the stability of the upward conguration, can be found in
[GPS80]. Several papers have analyzed the properties of its dynamics as well,
with particular focus on stabilizing the sleeping motion. On this subject, see
[WTCB94], [WCB95] and related references. The sleeping motion consists of
the situation where the top stands in the vertical conguration, steady rotating.
Furthermore, the dynamics of the heavy top represent basic background in the
literature of astrodynamics, for instance, to explain the Earth wobble and the
precession of equinox [BMW71].
Surprisingly, none of the published works have extended the motion of the heavy
top to the case where there is no table, i.e. when the top is also allowed to move
under the horizontal plane. From this point of view, current theory can be
considered incomplete. Further, proposed control laws for the vertical upward
equilibrium, dened global as in [WCB95], are actually not global. This is
due to the presence of another equilibrium for the system, represented by the
downward vertical conguration. In the language introduced for the 3D rigid
pendulum, this conguration consists of the hanging equilibrium of the reduced
dynamics (1.12) and (1.13). On the other hand, the upward equilibrium cong-
uration, consists of the inverted equilibrium.
In order to complete the description of the dynamics of the spinning top in
the new context of the 3D rigid pendulum, we study the stability of its hang-
ing equilibrium. An almost global attitude control law for the system, will be
proposed in the following chapters.
2.3 Hanging Equilibrium of the Symmetric Spinning
Top
For any constant velocity
z
= k, system (2.7) has two equilibria on S
2
R
2
.
They are given by the two states where
x
=
y
= 0 and angles and cor-
responding to the upward and the downward conguration. The rst one is
described by = 0, = or = , = 0 and it is referred to as the inverted
equilibrium. The second one is described by = 0, = 0 or = , =
and it is referred to as the hanging equilibrium. As introduced above, only
the inverted equilibrium has been studied in classical mechanics, restricting the
analysis to the space above the horizontal plane: (/2, 3/2). However,
model (2.7) describes the motion on the entire space S
2
R
2
, provided that
angle does not cross the horizontal: ,= /2 and ,= 3/2. Thus, we study
the properties of the second equilibrium, located below the horizontal plane. It
aims at completing theory related to the heavy symmetric top.
In order to analyze the stability of the hanging equilibrium, we make use of a
dierent model with respect to (2.7), representing a 3-2-1 formulation of the
Eulers angles. The reason of this selections is motivated by the fact that in
model (1.8)-(1.10), the sleeping motion does not correspond to an isolated equi-
librium. Indeed, when the rst rotation is around axis Z, the second and the
third Eulers angles are not distinguishable. It could be expected noticing that
each equilibrium is described by two dierent congurations.
In this context, the problem can be overcame by using a 2-1-3 Eulers angles
formulation, as the one previously presented in [WCB95]. There, the inertial
frame I attitude is described by three angles corresponding to three subse-
quent rotations, where the rst one is around Y , the second one is around the
rst axis of the new frame and the last one is around the third axis of the ob-
tained frame. In [WCB95] an angle is introduced, corresponding to the rst
rotation about Y . Then, the new frame is rotated of angle about the rst
axis. Resulting reference frame is referred to as F. Finally, a positive rotation
of the F-frame of angle about its third axis, results in the body-xed frame.
Thus, F precesses, but does not spin with the top. The angular velocity of the
3D rigid body can then be expressed in F, giving
= [
,

cos ,

sin]
T
(2.8)
It allows at writing the kinetic and potential energy in F coordinates and then
at following the Lagrangian approach (2.4). The resulting model is
J
t
+J
t

2
cos sin +p

cos mgl sin cos = 0
J
t

cos
2
2J
t

cos sin p

cos mgl cos sin = 0
(2.9)
29
where p
is the constant generalized momentum corresponding to :

p
=
L
= J
t
(

sin)
The rational under description (2.9), depending only on two of the three Eulers
angles, is that does not appear in the Lagrangian [WCB95], so it is an ignorable
variable. As a consequence, the Lagrangian can be replaced by the Routhian R,
dened by R = Lp

. Reduction theory, useful in the presence of symmetries
in mechanical systems, is extensively described in [MRS00].
Now, letting
x
1
= x
2
=

x
3
= x
4
=

b =
p
J
t
c =
2mgl
J
t
(2.10)
we obtain a state space representation for model (2.9):
x
1
= x
2
x
2
=
x
2
4
2
sin(2x
1
) bx
4
cos x
1
+
c
2
sinx
1
cos x
3
x
3
= x
4
x
4
= (2x
2
x
4
sinx
1
+bx
2
+
c
2
sinx
3
)
1
cos x
1
(2.11)
Notice again the presence of singularities, here for = n/2, n Z. As ex-
pected, the equilibria of the system correspond now to isolated points repre-
sented by x = [i, 0, j, 0]
T
. This results in the inverted equilibrium, i.e. the
upward vertical sleeping motion, when i j is even. Otherwise, it corresponds
to the hanging equilibrium.
In [WCB95] a linearized analysis around the inverted equilibriumx
i
= [0, 0, 0, 0]
T
is performed, to show that it is unstable for b
2
< c. It means that the rotation
of the top needs to be fast enough, to guarantee a vertical conguration. An
energy based equivalent proof is provided in [GPS80]. Here, we use a similar ap-
proach, dening a suitable Lyapunov function. Indeed, it is easy to show that
the linearized model around x
h
does not provide information on its stability,
being the eigenvalues complex conjugates.
Thus, we introduce a Lyapunov function based on the conserved quantities of the
motion embedded in (2.11): total energy and vertical momentum. This way to
proceed, previously suggested by [SLRM92], is referred to as the Energy-Casimir
method. It consists of selecting a Lyapunov function composed by Casimir func-
tionals, i.e. dierences between conserved quantities and their minimum values.
Now, the total energy of (2.9) is given by
E =
1
2
J
t
(

2
+

2
cos
2
) +
1
2
J
a
(

sin)
2
+mgl cos cos
Disregarding the kinetic energy related to the rotation around the spinning axis
and dividing by J
t
we obtain a constant of the motion:
h
c1
(x) =
1
2
(x
2
2
+x
2
4
cos
2
x
1
) +
c
2
cos x
1
cos x
3
(2.12)
expressed in the state space variables dened above. Further, the vertical mo-
mentum is given by h =
T
J with dened by (2.8) and as introduced in
Chapter 1, but expressed in F coordinates:
=
_
_
1 0 0
0 cos sin
0 sin cos
_
_
_
_
cos 0 sin
0 1 0
sin 0 cos
_
_
_
_
0
0
1
_
_
=
_
_
sin
sin cos
cos cos
_
_
Hence,
h = J
t

sin +J
t

cos cos sin +J
3
(

sin) cos cos =
= J
t

sin +J
t

cos cos sin +J
t
b cos cos
Dividing by J
t
and using the state space representation, we have a second con-
stant of the motion:
h
c2
(x) = x
2
sinx
3
+x
4
sinx
1
cos x
1
cos x
3
+b cos x
1
cos x
3
(2.13)
Therefore, a suitable Energy-Casimir based Lyapunov function for the system
is
V (x) = [h
c1
(x) h
c1
(x
h
)] +
b
2
[h
c2
(x) h
c2
(x
h
)]
Consequently, stability of the hanging equilibrium can be studied.
Proposition 2.4 For any constant spinning velocity
z
= k ,= 0, the hanging
equilibrium of the axisymmetric rigid pendulum (2.7), is locally stable in the
sense of Lyapunov.
31
Proof. Consider description (2.11) of the system and the following Energy-
Casimir based Lyapunov function
V (x) = [h
c1
(x) h
c1
(x
h
)] +
b
2
[h
c2
(x) h
c2
(x
h
)]
=
1
2
(x
2
2
+x
2
4
cos
2
x
1
) +
c
2
cos x
1
cos x
3
+
c
2
+
b
2
(x
2
sinx
3
+x
4
sinx
1
cos x
1
cos x
3
+b cos x
1
cos x
3
+b)
(2.14)
where b and c are dened by (2.10).
V (x
h
) = 0 and

V (x) 0 x. Indeed, (2.14) is composed by two constants of
the motion. Moreover, consider the Hessian matrix of V (x) evaluated in x
h
:
V (x
h
) =
_
_
c+b
2
2
0 0
b
2
0 1
b
2
0
0
b
2
c+b
2
2
0
b
2
0 0 1
_
_
(2.15)
Its eigenvalues are
1,2
=
b
4
+ 2b
2
c +c
2
4c + 4
4
+
b
2
+c + 2
4
3,4
= +
b
4
+ 2b
2
c +c
2
4c + 4
4
+
b
2
+c + 2
4
Simple manipulations show that
1,2
can be negative only if b
2
+ 2c < 0, but
this is clearly impossible, since c = 2mgl/J
t
> 0. Moreover,
3,4
are always
positive. Thus, all the eigenvalues of the Hessian matrix evaluated in x
h
are
positive, i.e.

V (x
h
) is positive denite. Therefore, the hanging equilibrium is
Lyapunov stable [Kha01].
As a consequence of the above proposition, the inverted equilibrium cannot be
globally stabilized in S
2
R
2
.
Finally, in [GPS80] it is shown that the heavy top is in steady precession when
p
2
4J
t
mgl cos cos
Thus, since any attitude of the top below the horizontal plane is such that
cos cos < 0, there the precession is always steady, no matter is the angular
velocity. These dynamics are fundamentally dierent from those of the spherical
pendulum. They open new interesting theoretical problems that seem not to be
previously studied, as the research of possibly existing periodic trajectories.
2.4 Conclusions
In Chapter 1 we have introduced a 3D rigid pendulum mathematical model
encompassing geo-stationary spacecraft descriptions. In this chapter we have
shown that this model actually embodies many other classical problems. Indeed,
the 2D spherical pendulum, the 1D planar pendulum and the symmetric heavy
top have been shown to represent special cases of model (1.12)-(1.13). Thus,
the study of the 3D pendulum includes the study of these classical problems.
Moreover, we have extended the classical description of the heavy top to the
space below the horizontal plane, giving rise to novel dynamics. Finally, we
have analyzed the equilibria of this new inverted top illustrating connections
with current literature.
33
Chapter
3
Stabilization of an Asymmetric
3D Rigid Pendulum
In the previous chapters we have introduced several dierent classications of
an uncontrolled 3D rigid pendulum supported by a frictionless pivot, both in
the asymmetric and axisymmetric case. Related theory is basic in spacecraft
dynamics and control problems, since an asymmetric balanced 3D rigid pen-
dulum, actually represents a generic geo-stationary satellite [Hug86], [Kap76].
Moreover, assumptions on the symmetry of the rigid body, have been taken in
the literature, while facing attitude control problems for underactuated space-
craft. Indeed, a rigid body with only two control inputs satises the fundamental
Broketts condition [BMS83] for continuous feedback stabilization, only if it is
symmetric. Consequent model coincide with the axisymmetric balanced 3D
rigid pendulum developed in Chapter 2.
In this chapter we focus the attention on an asymmetric 3D rigid pendulum
and we study the stabilization of its reduced model. For stabilizing the reduced
model of the 3D pendulum we mean stabilizing its hanging equilibrium, i.e.
pointing the vector downwards.
First, we make control actuation assumptions for the system, reminding briey
some alternatives to our selection. Then, we present a simple controller consist-
ing of angular velocity feedback. The related control law is based on damping
injection and it is shown to have a dense domain of attraction. Damping in-
jection, as well as passivity based control [OSR98], is particularly suitable for
mechanical systems, as the TACT presented in Chapter 1. Afterwards, we show
experimental results on the TACT, conrming the eectiveness of the proposed
controller.
3.1 3D Rigid Body Control Torques
So far, we have considered the mathematical model of an uncontrolled 3D rigid
body supported by a frictionless pivot, making assumptions on its geometrical
structure and analyzing its dynamics. In order to consider control inputs to the
system, the dynamics equation have to be modied for including moments. In
particular, (1.11) can be modied as
J = J +mg R
T
e
3
+M (3.1)
where M represents the vector control moment on the 3D rigid pendulum, ex-
pressed in the body-xed coordinate frame. The specic form of the control
moment depends on the actuation assumptions, i.e. on the hardware selected
for providing torques to the system. Several dierent actuation structures have
been studied in the literature. One important example of control actuation as-
sumption, generalizing the development in [AF00], is the pivot acceleration in
the control input. In this case the control moment is
M = m R
T
u
where u denotes the pivot acceleration, expressed in the inertial coordinate
frame. Such a control actuation assumption typically breaks the symmetry of
the uncontrolled rigid pendulum dynamics. Consequently, the angular momen-
tum component about the vertical axis through the pivot is not conserved and
it is not possible to express the controlled dynamics in terms of the reduced
attitude .
Another control actuation assumption, generalizing the development in [CSMB01]
35
and [CM02], is that proof mass actuators are rigidly mounted on the 3D pen-
dulum. Although the model presented in these references include actuation
dynamics, it is possible to give a simple expression for the control moments as
M = mgu R
T
e
3
ignoring the dynamics of the proof actuators. Here, the control input u denotes
the incremental displacement vector due to the proof mass motion, so that the
position vector of the center of mass of the total body is +u, expressed in the
body-xed coordinate frame. In this case, the mass m denotes the total mass of
the 3D pendulum and the proof mass actuators. It is easy to show that this type
of control actuation does not invalidate conservation of the angular momentum
component about the vertical axis through the pivot. Hence, the resulting
controlled dynamics can still be expressed in terms of the reduced attitude vector
. However, such a kind of control actuation is useful only with unbalanced
systems, since it makes use of the gravity vector. Thus, it is not suitable for
geo-stationary spacecraft control problems. Other actuation schemes, based on
reaction wheels, magnetic torques, precession and spin stabilization have been
proposed. This argument is a constant subject of study.
In this dissertation, besides these several actuation systems, we consider the
case where M is directly provided by thrusters, as the gas jet actuators usually
mounted on geo-stationary satellites. This way to proceed is inspired by the
control test-bed presented in Chapter 1. There, four fan thrusters are mounted
on one of the two external plates, as illustrated in Figure 1.4. Combining their
forces, these thrusters are able to provide roll, pitch and you torques. Dening
as F
i
, where i = 1, , 4, the forces provided by each fan,
roll
,
pitch
and
yaw
can be calculated. Their structure is depicted in Figure 3.1, where is the angle
between the fans position and the center of mass of the rigid body and r is the
distance between the center of mass of the TACT and each fan thruster. With
simple calculations it is possible to show that the torques applied to the system
by the fans are
roll
= (rF
1
rF
2
+rF
3
rF
4
) sin
pitch
= (rF
3
+rF
4
) cos
yaw
= (rF
1
+rF
2
) cos
Y
Z
X
F
1
r
F
2
F
3
F
4
CM
q
q
Figure 3.1. Fan thrusters scheme of the Triaxial Attitude Control Testbed.
In the case of the TACT, forces F
i
are generated applying suitable voltages to
the fan thrusters. The relations between torques and voltages are described
in [CBA
+
06]. They allows at expressing control inputs to the TACT in terms
of voltages to be applied to the fans. Results obtained in this reference are
promising for continuing fruitful research on the control test-bed.
Beyond the relations between F
i
and the torques in the roll, pitch and yaw
directions, we assume here that inputs to the 3D rigid body are directly the
angular momentum corresponding to
roll
,
pitch
and
yaw
. Hence, the dynamics
model extending (1.11) is simply
J = J +mg R
T
e
3
+u
Moreover, R
T
e
3
can be replaced by , since the angular momentum component
about the vertical axis through the pivot is conserved along the trajectories
of rigid body. Therefore, the control model for the fully actuated 3D rigid
pendulum on which we focus our attention is
J = J +mg +u
=
(3.2)
where R
3
is the angular velocity of the body, S
2
is the reduced at-
titude vector representing the direction of gravity in B and u R
3
is the
angular momentum input vector, and where all the variables are expressed in
B coordinates.
37
3.2 Asymptotic Stabilization of the Hanging Equilib-
rium
In this section, we develop a simple controller that makes the hanging equi-
librium of the asymmetric 3D pendulum asymptotically stable. The controller
is based on the observation that the control model given by equation (3.2) is
input-output passive if the angular velocity is taken as the output. Passivity
based control is a natural way to proceed in the presence of mechanical systems,
as well as in the presence of electrical systems [OSR98]. Briey, a system with
state x, input u and output y is dissipative with respect to a supply function
s(u(t), y(t)) if there exists a function H such that
H(x(T)) H(x(0)) +
_
T
0
s(u(t), y(t))dt (3.3)
Such a function is referred to as the storage function. When the supply rate
is s(u, y) = u
T
y, the system is said input-output passive. In this case, it is
immediate to notice that
dH
dt
u
T
y
Thus, selecting u = y, gives a closed-loop system where dH/dt 0. There-
fore, if H is the total energy, u drives the system to the minimum of its energy,
corresponding to a stable equilibrium.
In the case of the 3D rigid pendulum, the total energy
1
2
T
J mg
T
can be
selected as a storage function. Since it has a minimum at the hanging equilib-
rium (0,
h
), a control law based on angular velocity feedback is suggested.
Let : R
3
R
3
be a smooth function such that
1
|x|
2
x
T
(x)
2
|x|
2
, x R
3
(3.4)
where
2

1
> 0. Thus, we propose a class of controllers, referred to as
damping injection controllers, given by
u = () (3.5)
where satises (3.4). We next show that the above family of controllers,
which requires only angular velocity feedback, renders the hanging equilibrium
of a 3D asymmetric pendulum asymptotically stable, with a guaranteed domain
of attraction.
Lemma 3.1 Consider the fully actuated 3D asymmetric pendulum given by
(3.2). Let : R
3
R
3
be a smooth function satisfying (3.4) and choose a
controller in the form of (3.5). Then, the hanging equilibrium (0,
h
) is asymp-
totically stable.
Furthermore, for every (0, 2mgl), all solutions of the closed-loop system
given by (3.2) and (3.5), such that ((0), (0)) H, where
H =
_
(, ) R
3
S
2
:
1
2
T
J +
1
2
mgl
_

l
_
T
_

l
_
2mgl
_
satisfy lim
t
(t) = 0 and lim
t
(t) =
h
.
Proof. Consider the closed loop system given by (3.2) and (3.5). We propose
the following candidate Lyapunov function
V (, ) =
1
2
_
T
J +mgl
_

l
_
T
_

l
_
_
(3.6)
Note that the above Lyapunov function is positive denite on R
3
S
2
and
V (0,
h
) = 0. Furthermore, the derivative along a solution of the closed-loop
system given by (3.2) and (3.5) is
V (, ) =
T
(J ) +mg(l )
T

=
T
(J +mg ()) +mg(l )
T
( )
=
T
() +mg
T
( ) mg
T
( )
=
T
()
1
||
2
,
where the last inequality follows from (3.4). Thus, V (, ) is positive denite
and

V (, ) is negative semidenite on R
3
S
2
.
Next, consider the sub-level set given by H = (, ) R
3
S
2
: V (, )
2mgl . Note that the compact set H contains the hanging equilibrium
(0,
h
). Since,

V (, ) 0, all solutions such that ((0), (0)) H satisfy
((t), (t)) H for all t 0. Thus, H is an invariant set for solutions of (3.2)
and (3.5). Furthermore, from LaSalles invariant set theorem, we obtain that
solutions satisfying ((0), (0)) H converge to the largest invariant set in
39
(, ) H : = 0, that is = 0.
Therefore, either (t) /l =
h
or (t) /l =
i
as t . Since
(0,
i
) , H, it follows that (t)
h
as t . Thus, (0,
h
) is an asymptoti-
cally stable equilibrium of the closed-loop system given by (3.2) and (3.5), with
a domain of attraction H.
This result could be expected noticing that 2mgl is the total energy of the pen-
dulum at the inverted equilibrium. Thus, when ((0), (0)) is in H, injection
of dumping can only decelerate the rigid body, towards the other equilibrium,
represented by the hanging equilibrium. Dierent is the case when ((0), (0))
is such that the total energy is exactly 2mgl. In this case (t) can be shown
to approach
h
as t , only if the initial kinetic energy is not zero. The
nontrivial proof of this result is provided by the following Lemma.
Lemma 3.2 Consider the fully actuated 3D asymmetric pendulum given by
(3.2). Let : R
3
R
3
be a smooth function satisfying (3.4) and choose a
controller as given by (3.5). Then, all solutions of the closed-loop system given
by (3.2) and (3.5), such that (0) ,= 0 and ((0), (0)) /, where
/ =
_
(, ) R
3
S
2
:
1
2
T
J +
1
2
mgl
_

l
_
T
_

l
_
= 2mgl
_
satisfy lim
t
(t) = 0 and lim
t
(t) =
h
.
Proof. Consider the closed loop system given by (3.2) and (3.5), and the Lya-
punov function given in (3.6). As already shown,

V (, ) =
T
(). Thus,
V (, ) =
T
()
T

() =
T
_
() +
_
_
T
_
since is smooth. Furthermore, for all (, ) /, is bounded and
= J
1
(J +mg ())
is also bounded. Dene
N = sup
(,)A
__
_
_
_
_

T
_
() +
_
_
T
__
_
_
_
_
_
< .
Next, since ((0), (0)) /, V ((0), (0)) = 2mgl and |
V ((0), (0))| N.
Expanding V ((t), (t)) in a Taylor series, gives
V ((t), (t)) = 2mgl (0)
T
((0))t +R(t) 2mgl
1
|(0)|
2
t +N t
2
,
since the remainder necessarily satises |R(t)| Nt
2
.
Dene
= min
_
2mgl, 3
2
1
|(0)|
4
16N
_
.
It can be easily shown that for all t [T
1
, T
2
],
t
2
1
|(0)|
2
N
t +

N
0
where
T
1
=
1
|(0)|
2
2N
_
1
1
4 N
2
1
|(0)|
4
_
> 0
and T
2
T
1

1
(0)
2
2N
. Choose an (0, ). Then, for all t [T
1
, T
2
],
V ((t), (t)) 2mgl < 2mgl and hence, ((t), (t)) H, where H is
the invariant set given in Lemma 1. Thus, from Lemma 1, we obtain the result
that (t) 0 and (t)
h
, as t .
By the two Lemmas above, the hanging equilibrium is stable, with a guaranteed
domain of attraction. Indeed, when the initial conditions ((0), (0)) H, the
trajectories converge to the hanging equilibrium. Furthermore, when the initial
total energy is 2mgl and the initial kinetic energy is not zero, the trajectories
enter H in nite time. This result is obtained applying the simple damping
injection based control law (3.5). This result is summarized by next theorem.
Theorem 3.1 Consider the fully actuated 3D pendulum given by (3.2). Let
: R
3
R
3
be a smooth function satisfying (3.4) and choose controller as in
(3.5). Then, all solutions of the closed-loop system given by (3.2) and (3.5),
such that ((0), (0)) ^(0,
i
), where
^ =
_
(, ) R
3
S
2
:
1
2
T
J +
1
2
mgl
_

l
_
T
_

l
_
2mgl
_
(3.7)
satisfy lim
t
(t) = 0 and lim
t
(t) =
h
.
41
Proof. From Lemmas 3.1 and 3.2, we obtain the result that for every
(0, 2mgl) and ((0), (0)) H
/, where H and / are as dened in Lemmas

3.1 and 3.2, (t) 0 and (t)
h
as t . Since, ^ can be written as
^ =
_
(0,2mgl)
(H
_
/)
the result follows.
It is important to point out that ^(0,
i
) is an invariant set, but it is not
the maximal domain of attraction of the hanging equilibrium. Any solution of
(3.2) and (3.5), starting from an initial condition not in ^(0,
i
), that does
not pass through (0,
i
), must eventually enter ^(0,
i
). Hence, the maxi-
mal domain of attraction of the hanging equilibrium is a proper superset of the
invariant set ^(0,
i
).
In the subsequent theorem, we claim that for the control given in (3.5), the
equilibrium (0,
h
) of (3.2) and (3.5) is almost globally asymptotically stable,
and the domain of attraction of the hanging equilibrium is dense.
Theorem 3.2 Consider the fully actuated 3D asymmetric pendulum described
by (3.2). Let : R
3
R
3
be a smooth function satisfying (3.4) and
(0)
be positive denite and symmetric. Choose controller as given in (3.5). Let /
denote the stable manifold of the equilibrium (0,
i
). Then (R
3
S
2
)/is dense
in (R
3
S
2
), and all solutions of the closed-loop system given by (3.2) and (3.5),
such that ((0), (0)) (R
3
S
2
)/, satisfy lim
t
(t) = 0 and lim
t
(t) =
h
.
Proof. Consider the closed loop system given by (3.2) and (3.5), and the
Lyapunov function given in (3.6). Since S
2
is a compact set and the Lyapunov
function V (, ) is quadratic in , each sublevel set of V (, ) is a compact
set. Furthermore, since

V (, ) =
T
() 0, by LaSalles invariant set
theorem, all solutions converge to the largest invariant set in (, ) R
3
S
2
:
V (, ) = 0.
The largest such invariant set is given by (0,
h
)
(0,
i
). Hence, it is suf-
cient to show that the stable manifold of the inverted equilibrium (0,
i
), has
dimension less than the dimension of (R
3
S
2
), i.e. ve, since all other solutions
converge to the hanging equilibrium (0,
h
) by invariant set theorem [Kha01].
Using linearization, it can be shown that the equilibrium (0,
i
) of the closed
loop given by (3.2) and (3.5) is unstable and hyperbolic with nontrivial stable
and unstable manifolds. Hence, from theorem 3.2.1 in [GH83], it follows that
the dimension of the stable manifold of (0,
i
) is less than ve, so that the do-
main of attraction in theorem 3.1 is dense in (R
3
S
2
).
Comparing the conclusions in the two previous theorems we notice that, theo-
rem 3.1 provides an explicit description of a domain of attraction of the hanging
equilibrium; however that domain of attraction is not maximal. In contrast, the-
orem 3.2 shows that the maximal domain of attraction consists of all points in
(R
3
S
2
) that are not in the stable manifold / of the inverted equilibrium.
However, the stable manifold of the inverted equilibrium is not explicitly de-
scribed. We note that Lebesgue measure of / is zero.
3.3 Experiments on Stabilization of the Hanging Equi-
librium
In this section we present experimental results on the stabilization of the hanging
equilibrium of the asymmetric 3D rigid pendulum. They validate the controller
proposed in the previous section, showing that (3.5) asymptotically stabilizes the
hanging equilibrium with angular velocity feedback. The test-bed on which the
experiments are performed, is the Triaxial Attitude Control Testbed (TACT)
described above.
We are able to measure the angular velocity of the TACT thanks to three
orthogonal gyroscopes mounted on one of the two external plates, suitably cali-
brated. First, bias is eliminated keeping the TACT in a resting position. Then,
a rotation about each axis is performed, assuring output only from the corre-
sponding gyro, and no data from the others.
In order to measure the attitude of the TACT, a triaxial accelerometer is
mounted on one of the two external plates, aligned with the body-xed axes.
Before each test, the accelerometer triad is calibrated to read g on each indi-
43
vidual axis when aligned with the downward vertical. To do this, the TACT
is rotated about each axis independently, while taking continuous reading from
the accelerometer. The accelerometer axis is dened to be aligned with the
downward vertical when a maximum occurs in the data. Bias and scaling fac-
tors are then adjusted such that each accelerometer outputs g at the maximum
point.
Actually, the reduced attitude cannot be measured directly. Indeed, accelerom-
eter are not mounted at the center of gravity of the system. Thus, must be
estimated. We are able to estimate the reduced attitude using accelerometers
and gyroscopes together. First we read the accelerometer measure. It gives not
only the acceleration due to gravity, but also angular and centripetal accelera-
tion terms. Centripetal acceleration can be determined indirectly by using the
knowledge of the angular velocity. It correspond to ( r). The angular
acceleration term is ignored in the computation of the reduced attitude, but
preliminary calculations suggest that its contribution is quite small.
Hence, direction of gravity is calculated as
Gravity = Measure ( r)
representing the vector, since already expressed in body-xed coordinates.
Finally, inertia matrix J is identied as described in [CBA
+
06].
Results of the application of (3.5) to stabilize the hanging equilibrium, are
depicted in Figures 3.2 to 3.4. The controller used for this purpose is chosen
such that
() = P
where P = 10I
3
. This choice of equal gains in the three feedback chan-
nels does not represent a particularly eective choice in terms of closed-loop
performance. However, our objective in the experiment is to demonstrate
closed-loop stabilization; we do not seek to tune the gains to achieve closed-
loop performance. The desired reduced attitude vector is the hanging equi-
librium
0
= (0, 0, 1)
T
. The pendulum is released from the reduced attitude
(0) = (0.2936, 0.8579, 0.4217)
T
, with (0) = 0. These initial conditions are
chosen to excite all degrees of freedom of the system.
As seen from experimental closed loop responses shown in Figures 3.2 and 3.3,
0 5 10 15 20 25 30 35 40
5
0
5
x

(
D
e
g
/
s
e
c
)
Angular Velocities
0 5 10 15 20 25 30 35 40
5
0
5
y

(
D
e
g
/
s
e
c
)
0 5 10 15 20 25 30 35 40
50
0
50
100
z

(
D
e
g
/
s
e
c
)
Time [sec]
Figure 3.2. Experimental results for the evolution of the angular velocity of the 3D
pendulum in the body frame.
0 5 10 15 20 25 30 35 40
0.5
0
0.5
Reduced Attitude Vector in Body Frame
x
0 5 10 15 20 25 30 35 40
1
0.5
0
0.5
1
y
0 5 10 15 20 25 30 35 40
0
0.5
1
z
Time [sec]
Figure 3.3. Experimental results for the evolution of the components of the direction
of gravity in the body frame.
the angular velocity converges to zero and (t) converges to
0
= (0, 0, 1)
T
as
t . This is clear from Figure 3.4 which shows that the response of the
45
0 5 10 15 20 25 30 35 40
0
10
20
30
40
50
60
70
80
Time [sec]

[
D
e
g
]
Angle between (t) and
0
in Body Frame
Figure 3.4. Experimental result for the evolution of the angle between the reduced
attitude vector (t) and the desired reduced attitude vector
0
.
angle between (t) and
0
converges to zero as t . Notice that a residual
damped oscillation remains in the pitch axis angular velocity. It can be reduced
adjusting the feedback control gains. Notice also that measurements during the
initial transient period are especially noisy. This can be attributed to both gyros
noise and an attitude estimation algorithm that ignores angular acceleration,
which can add to noisy measurements during the initial transient period.
The fan thrusters mounted on the TACT provide approximately 5 Nm of con-
trol authority in each axis. Thrusters saturation can be an issue in these control
experiments and the feedback gains need to be tuned to minimize any nonlin-
earities arising from saturation. This is not critical in stabilizing the hanging
equilibrium of the 3D pendulum. Indeed, saturation might only produce a slower
convergence. On the other hand, it has to be taken into account while stabi-
lizing a generic attitude. Techniques to be used in the presence of saturation
constraints, for stabilizing spacecraft, will be presented in the second part of
this dissertation.
Finally, notice that in this experiment, initial conditions are inside the guaran-
teed domain of attraction ^. Several experiments have been performed on the
TACT showing that under (3.5), dynamics converge to the hanging equilibrium
also starting outside ^. They are presented in [SCB
+
05].
3.4 Conclusions
In this chapter we have introduced a passivity based control law for stabilizing
the hanging equilibrium of an asymmetric 3D rigid pendulum. The described
control technique is based on the observation that angular velocity feedback
renders the closed-loop input-output passive, with the total energy as storage
function. We have shown that this control law cannot be global, due to the
presence of the inverted equilibrium, but it is near-global. Finally, we have
presented results of the application of the technique to the TACT.
Even if the task of stabilizing the hanging equilibrium of the 3D rigid pendulum
can appear quite simple, since
h
is a minimum for the potential energy, this is
fundamental for introducing the control of a generic attitude described in the
following chapter.
47
Chapter
4
Attitude Control of an
Asymmetric 3D Rigid
Pendulum
In the previous chapter we have presented a technique for stabilizing the hanging
equilibrium of an asymmetric 3D pendulum, based on damping injection. The
proposed control law has been inspired by the fact that the hanging equilibrium
is a minimum of the potential energy of the system. In this chapter we gen-
eralize this concept, proposing a control law for an arbitrary reduced attitude
vector.
For controlling the reduced attitude of a 3D pendulum we mean pointing its
vector towards a pre-established direction. This is an usual task in satellite sys-
tems, corresponding to stabilize the orientation of spacecraft instrumentations
or antennas. Thus, this problem is strictly related to geo-stationary satellites
attitude control. Several local control laws for geo-stationary spacecraft, have
been proposed in the literature, based on Eulers angle error, quaternion error,
Euler axis of rotation and many other techniques
1
. In particular, here we focus
1
See [Kap76] and [Sid00] for a comprehensive overview.
our attention on the approach adopted in [Bur95], for a model described by
(3.2). We rst show that this control law does not guarantee convergence to
an arbitrary attitude. Therefore, we propose a class of controllers that suit-
ably modify the potential energy of the rigid body, giving a closed-loop system
whose potential has a minimum exactly at the desired attitude. This technique
is referred to as potential shaping [OSR98]. We show that it produces a nearly
global stabilization of any arbitrary attitude.
Finally, we apply the proposed technique to an asymmetric 3D rigid pendulum
and we depict simulation results, showing its eectiveness.
4.1 Local Asymptotic Stabilization of an Arbitrary Equi-
librium
As stated in the previous chapters, the mathematical model of a balanced 3D
rigid pendulum actually describes the motion of a geo-stationary spacecraft.
Indeed, the most common description of spacecraft evolution in composed by the
Poissons kinematics equation (1.6) and the dynamics equation (3.1), without
the gravity term mgR
T
e
3
. Moreover, the reduced kinematics equation (1.13)
is used as a simplied version of (1.6), since it describes the evolution of R
T
e
3
,
i.e the third row of the rotation matrix R.
Thus, controllers proposed in the literature of geo-stationary spacecraft, appear
suitable also for stabilizing the orientation of a generic unbalanced asymmetric
3D rigid pendulum. It can be made selecting a control input of the form
u = u mg R
T
e
3
(4.1)
viz. canceling the eects of gravity.
Besides the several control strategies described in [Sid00], here we consider a
simplied version of the technique presented in [Bur95]. In this reference, a
passivity based control input for complete attitude is designed, assuming no
availability of angular velocity measurements. Here, we remove this assumption
and we consider the reduced attitude kinematics model (1.13). It produces the
49
following control law for stabilizing an arbitrary orientation
0
u = P +k(
0
) (4.2)
with P a positive denite matrix and k a real positive constant.
Theorem 4.1 The closed-loop system (3.2) and (4.2) with u as given in (4.1),
possesses an asymptotically stable equilibrium in (, ) = (0,
0
) and an unstable
equilibrium in (, ) = (0,
0
).
Proof. Let
0
be an arbitrary reduced attitude. Note that the closed loop
system given by (3.2), (4.1) and (4.2) has two equilibria for (, ) = (0,
0
) and
(, ) = (0,
0
). Consider the following candidate Lyapunov function
V (, ) =
1
2
_
T
J +k(
0
)
T
(
0
)
(4.3)
The above Lyapunov function is positive denite on R
3
S
2
and V (0,
0
) = 0.
Furthermore, the derivative along a solution of the closed-loop system given by
(3.2), (4.1) and (4.2) is
V (, ) =
T
J +k(
0
)
T

=
T
(P +k(
0
)) +k(
0
)
T
( )
=
T
P +k
T
(
0
) k
T
0
( )
=
T
P 0
Thus, V (, ) is positive denite and

3
S
2
.
Hence, from LaSalles invariant set theorem, we obtain that (, ) converges to
the largest invariant set in R
3
S
2
such that = 0, i.e. (
0
) = 0. Moreover,
if ((0), (0)) is such that V ((0), (0)) < 2k, V ((0), (0)) is an invariant set
for the trajectories of the closed loop system, starting from ((0), (0)). Since
V (0,
0
) = 2k, it follows that (t)
0
as t . Thus, (0,
0
) is an
asymptotically stable equilibrium of the closed-loop system given by (3.2), (4.1)
and (4.2).
On the other hand, consider the following Lyapunov function
V (, ) =
1
2
_
T
J +k( +
0
)
T
( +
0
)
(4.4)
The above function is positive denite on R
3
S
2
and V (0,
0
) = 0. Fur-
thermore, it is straightforward to show that

V (, ) is not dened along the
trajectories of the closed-loop system given by (3.2), (4.1) and (4.2). Thus,
(0,
0
) is an unstable equilibrium of the closed-loop system given by (3.2),
(4.1) and (4.2).
Notice that the term P in (4.2) can be replaced by any function () as
dened by (3.5). Indeed, it plays the role of injecting damping into the system.
Hence, the above controller is strictly connected to the one presented in chapter
3, for stabilizing the hanging equilibrium
h
= /l. This can also be deduced
comparing equation (3.6) and equation (4.3). The fundamental dierence is that
an arbitrary
0
is not a minimum of the potential energy, unless the potential
in suitably modied. In some way, this is what equations (4.1) and (4.2) do.
They modify the system, for having a closed-loop whose potential energy has a
minimum at =
0
. This way to proceed, is referred to as potential shaping,
in the sense that the control law actually re-shape the potential energy of the
whole system. A comprehensive description of this technique, that is particu-
larly suitable in the presence of mechanical systems, can be found in [BLM00],
[BLM01] and related references.
theorem 4.1 states the capability of control law (4.1) and (4.2) of stabiliz-
ing a desired attitude
0
, when initial condition satises V ((0), (0)) < 2k,
with V (, ) dened by (4.3). Unfortunately, it does not guarantee that in
general, the evolution of the closed loop system converge to the desired atti-
tude. Indeed, initial condition cannot be assured to belong to the invariant set
V ((0), (0)) < 2k. Selecting a large k for enlarging the guaranteed domain
of attraction of (0,
0
) is not in accordance with constraints usually present
in practical problem. In fact, in real applications, thrusters cannot supply ar-
bitrary large torques. This is especially true in aerospace applications, where
minimizing consumption is also an issue.
When ((0), (0)) is not in V ((0), (0)) < 2k, an initial condition belonging
to the stable manifold of (0,
0
) cannot be avoided a priori. This situation is
depicted in Figure 4.1 where lines represent function V (, ) level sets.
Therefore, the technique proposed in [Bur95] and its simplied version described
51
G
-G
0
0
V( =2k w,G)
G
w
(w(0),G(0))
0
Figure 4.1. Evolution of the closed loop system (3.2), (4.1) and (4.2) towards the
unstable equilibrium.
above, represent a local attitude control law. In order to arbitrarily enlarge the
domain of attraction of the desired nal attitude, one should design a controller
that enforces the closed-loop trajectories to move away from (, ) = (0,
0
)
as soon as the closed-loop evolution approaches the unstable equilibrium. This
argument leads to the nearly global control law presented in the following sec-
tion.
4.2 Attitude Control of a Geo-Stationary Spacecraft
In the last section, we have presented a controller that asymptotically stabi-
lizes an arbitrary conguration of the asymmetric 3D rigid pendulum, with a
guaranteed domain of attraction. Since the desired orientation of the vector
is arbitrary, the controller requires the knowledge of in addition to , as it
could be expected. Hence, it extends (3.5), providing an angular velocity and
reduced attitude vector feedback based control.
Unfortunately, the proposed controller is local. This motivates us to study the
problem of designing an alternative, nearly globally stabilizing control law.
It has to be noted that we aim at obtaining a nearly global controller. It means
that we rule out the possibility of globally stabilizing an arbitrary attitude of the
3D asymmetric rigid body. The reason is that there exists a topological obstruc-
tion in designing a continuous time-invariant controller for global stabilization
of an arbitrary reduced attitude, as widely motivated in [BB00]. This reference
represents a basic result in the theory of control for mechanical systems with ro-
tational degrees of freedom. Indeed, in [BB00] it is shown that systems having a
compact conguration manifold cannot be globally asymptotically stabilized to
a rest conguration, using continuous state feedback. Since the two rotational
degrees of freedom manifold S
2
is compact, this result applies for the problem
at hand.
Thus, we propose a controller that almost globally asymptotically stabilizes the
reduced attitude vector to a desired value
0
. By almost global, we mean that
the domain of attraction of the equilibrium is dense in R
3
S
2
. In particular, we
propose a control law that stabilizes the desired attitude
0
from every initial
condition in R
3
S
2
, except one point.
Let : [0, 1) R be a C
1
monotonically increasing function such that (0) = 0
and (x) as x 1. Let : R
3
R
3
be a smooth function satisfying
(3.4). We propose a class of controllers given by
u =
_
_
1
4
(
T
0
1)
2
_
(
T
0
1)
0
+mg
_
() (4.5)
Theorem 4.2 Let
0
S
2
. Consider the fully actuated system given by (3.2).
Choose a controller as given in (4.5). Then (0,
0
) is an equilibrium of the closed
loop system (3.2) and (4.5) that is globally asymptotically stable with domain of
attraction R
3
(S
2
0
).
Proof. Consider the closed-loop system given by (3.2) and (4.5) and the candi-
date Lyapunov function
V (, ) =
1
2
T
J + 2
_
1
4
(
T
0
1)
2
_
(4.6)
Note that the Lyapunov function is positive denite on R
3
S
2
and V (0,
0
) = 0.
Furthermore, every sub-level set of the Lyapunov function in R
3
S
2
is compact,
and the closed-loop vector eld given by (3.2) and (4.5) has only one equilibrium
in each sub-level set, namely (0,
0
).
Computing the derivative of the Lyapunov function along the solution of (3.2)
53
and (4.5) yields
V (, ) =
T
J +
_
1
4
(
T
0
1)
2
_
(
T
0
1)
T
0

=
T
(J +mg () mg )
T
_
_
1
4
(
T
0
1)
2
_
(
T
0
1)(
0
)
_
+
_
1
4
(
T
0
1)
2
_
(
T
0
1)
T
0
( )
=
_
1
4
(
T
0
1)
2
_
(
T
(
0
)
+
T
0
( ))(
T
0
1)
T
()
=
T
()
1
||
2
Thus,

3
S
2
and hence, all solutions remain
in the compact sub-level set given by
^ = (, ) R
3
S
2
: V (, ) V ((0), (0))
Note that ^ R
3
(S
2
0
). Next, from LaSalles invariant set theorem,
for any arbitrary initial condition, the solutions converge to the largest invariant
set in (, ) ^ : = 0. Thus, from the second equation in (3.2), we obtain
that is constant and from the rst equation in (3.2), we obtain that either
T
0
= 1 or
0
= 0, or both. Therefore, either (t)
0
or (t)
0
as
t . However, (0,
0
) is not contained in ^, hence, (t)
0
as t .
It should be noted that the controller (4.5) is not dened if =
0
S
2
. For
all other initial conditions in R
3
(S
2
0
), the controller guarantees that
the solution of the closed-loop, given by (3.2) and (4.5), asymptotically con-
verges to (0,
0
). Indeed, thanks to its form, () pushes the pendulum away
from
0
as soon as evolution approaches this orientation. It guarantees that
closed-loop solution do not pass through the stable manifold of
0
.
It is insightful to view the class of controllers in (4.5) as providing a combination
of potential shaping, represented by the function () and the term containing
gravity, and damping injection, represented by the function (). Furthermore,
it may be noted that the argument of the potential function () is proportional
to the cosine of the angle between and
0
. The closed loop has the property
that if (0) = 0, then for all t 0, the angle between (t) and
0
is bounded
above by the angle between (0) and
0
.
Corollary 4.1 Consider the fully actuated system given by (3.2) with controller
as in (4.5). Furthermore, let (0) = 0 and (0) ,=
0
. Then, for all t 0,
(
0
, (t)) (
0
, (0))
Proof. Consider the candidate Lyapunov function (4.6), for the closed-loop
system. As already shown,

V (, ) =
T
(). Thus, since

V (, ) is negative
semidenite, V ((t), (t)) V ((0), (0)). Thus substituting (0) = 0 in (4.6),
we obtain the result that for all t 0,
1
2
(t)
T
J(t) + 2
_
1
4
(
T
0
(t) 1)
2
_
2
_
1
4
(
T
0
(0) 1)
2
_
Since, the kinetic energy term is strictly non-negative and () is a monotonic
function, we obtain
[
T
0
(t) 1[ [
T
0
(0) 1[, t 0
Thus,
T
0
(t)
T
0
(0), t 0,
which implies that
cos ((
0
, (t))) cos ((
0
, (0))) , t 0
Since (
0
, (t)) [0, ) and cos() is non-increasing in [0, ), the result follows.
The result stated by the previous corollary is prominent in aerospace control

theory. Indeed, it assures that controller (4.5) avoid the so called unwinding phe-
nomenon. This phenomenon, rstly exposed in [BB00], is produced by many
attitude control laws proposed in the literature. It consists of the case when a
control law produces an attitude response that starts close to the desired orien-
tation
0
, diverges from
0
and then converges once again to
0
.
Thus, in some sense, controller (4.5) is able to impress to the asymmetric 3D
pendulum, the correct direction of angular motion, that is the one reducing the
55
angle between (0) and
0
. Hence, controller (4.5) uses a suitable metric on
S
2
, to dene the distance between initial attitude and nal desired attitude.
Such a metric is (
T
0
(0) 1). On the contrary, corollary 4.1 does not hold true
for technique (4.1) and (4.2) illustrated in previous section. There, the term
describing distance between initial condition and nal desired conguration is
represented by (
0
(0)). It is worth pointing out that the problem of dening
a correct metric on SO(3), initially faced in [WD91] and [FF94], is currently a
subject of study.
Thus, the proposed control law (4.5) produces a closed-loop system converg-
ing to any arbitrary desired equilibrium
0
, from every initial condition on
R
3
(S
2
0
), i.e. a nearly global control law. Moreover, (4.5) provides a
control input that avoids unwinding. Hence, in some way, (4.5) minimizes the
path between ((0), (0)) and (0,
0
). This is one of the most important prob-
lems in spacecraft practical applications, where minimizing consumption while
controlling attitude is a fundamental task.
4.3 Simulation on Stabilization of an Arbitrary Equi-
librium
In the previous section, we introduced a class of controllers that guarantee near
global stabilization of the reduced system (3.2). In this section, we present
simulation results for a specic controller selected from the family of controllers
given by (4.5). We choose
(x) = k ln(1 x)
where k > 0, and
(x) = Px
where P is a positive denite matrix. The resulting control law (4.5) is given
by
u = P mg +k
(
T
0
1)
1
1
4
(
T
0
1)
2
(
0
) (4.7)
Consider model (3.2) where m = 140 kg, = (0, 0, 0.5)
T
mand J = diag(3, 40, 50)
kgm
2
. These parameters correspond to the ones of the TACT described above.
Let u be the controller (4.7) where P = 2.5I
3
and k = 5. The following gures
describe the evolution of the closed loop system (3.2) and (4.7) correspond-
ing to the inverted equilibrium
0
= (0, 0, 1)
T
. The initial conditions are
(0) = (1, 3, 1)
T
rad/s and (0) = (0.5, 0.7071, 0.5)
T
. Hence, control law (4.7)
is activated while the system is moving.
Simulation results in Figures 4.2 and 4.3 show that (t) 0, and (t)
0
as t . Indeed, expressing control input u as u = u
1
+ u
2
, the control
0 50 100 150 200 250 300
5
0
5
1

[
r
a
d
/
s
]
Angular Velocity
0 50 100 150 200 250 300
4
2
0
2
4
2

[
r
a
d
/
s
]
0 50 100 150 200 250 300
2
0
2
4
3

[
r
a
d
/
s
]
time [s]
Figure 4.2. Evolution of the angular velocity of the 3D pendulum in the body
frame.
component
u
1
=
_
_
1
4
(
T
0
1)
2
_
(
T
0
1)
0
+mg
_
shapes the 3D pendulum energy for generating a closed-loop system whose po-
tential energy has a minimum at the inverted equilibrium. In the meanwhile,
the control component
u
2
= ()
injects damping into the systems to reduce angular velocity. This is also clearly
seen from the plot of the angle between (t) and
0
in Figure 4.4. Moreover, the
vertical angular momentum approaches 0 and the total energy E 686.3 N
57
0 50 100 150 200 250 300
1
0.5
0
0.5
1
1
Components of in Body Frame
0 50 100 150 200 250 300
1
0.5
0
0.5
1
2
0 50 100 150 200 250 300
1
0.5
0
0.5
1
3
time [s]
Figure 4.3. Evolution of the components of the direction of gravity in the body
frame.
0 50 100 150 200 250 300
0
20
40
60
80
100
120
140
160
180
0
time [s]

[
d
e
g
]
Figure 4.4. Evolution of the angle between the reduced attitude vector (t) and
0
.
m, i.e. the value corresponding to the inverted equilibrium, as shown in Figure
4.5.
In order to illustrate the result described in Corollary 4.1, we repeat the sim-
ulation, activating (4.7) while the system is in a rest position, i.e. (0) = 0.
We select gains P = 2I
3
and k = 10. In this case evolution is similar to the
one depicted above. Thus, we only show the plot of the angle between (t) and
0
in Figure 4.7. As expected from Corollary 4.1, the angle between (t) and
0 50 100 150 200 250 300
600
400
200
0
200
400
600
800
Angular Momentum about Vertical Axis Total Energy
time [s]
h

E
Momentum
Total Energy
Figure 4.5. Vertical angular momentum (solid line) and total energy (dash line) of
the 3D pendulum.
0
remains bounded above by the angle between (0) and
0
which is 120 deg.
This is in contrast to Figure 4.4, where the excursion in angle exceeds 120 deg.
4.4 Conclusions
In this chapter we have introduced a control law for stabilizing an arbitrary
attitude
0
of an asymmetric 3D rigid pendulum. It is made by designing a
closed-loop system whose potential energy has a minimum exactly at the de-
sired attitude, i.e. shaping the potential. Meanwhile, damping is injected into
the system for dissipating kinetic energy.
The obtained control law distinguishes with respect to previous literature be-
cause it provides nearly global asymptotic stability. Indeed, evolution con-
59
1
0.5
0
0.5
1
1
0.5
0
0.5
1
1
0.5
0
0.5
1
y
3D Pendulum Motion: /|| vector in Inertial Frame
x
z
Figure 4.6. Motion of the vector between the pivot and the center of mass of the
3D pendulum in the inertial frame.
0 50 100 150 200 250 300
0
20
40
60
80
100
120
140
0
time [s]

[
D
e
g
]
0
for (0) = 0.
verges to the desired state (, ) = (0,
0
), from any initial condition except of
((0), (0)) = (0,
0
). On the contrary, many controllers proposed previously
are shown to be only local.
Besides the ability of nearly globally stabilize the desired attitude, the proposed
scheme, has been shown to provide input torques in the direction that reduces
the angle between (0) and
0
. Thus, it uses a suitable metric on S
2
. This
is prominent in GEO-stationary spacecraft attitude control problems, since it
avoids the unwinding phenomenon and then it reduces fuel consumption.
Finally, we have presented simulation results, that conrm the theory. This
suggests the application of a potential shaping and damping injection technique
also to the special cases of the 3D rigid pendulum, illustrated in Chapter 2. It
will be made in the following chapter.
61
Chapter
5
Attitude Control of a 3D Axially
Symmetric Rigid Pendulum
In Chapter 2 we have shown how, with simple assumptions, the 2D pendulum,
the 1D pendulum and the symmetric heavy top, can be seen as special cases of
the 3D rigid pendulum. In particular we have analyzed the axisymmetric case,
i.e. the case where two of the principal moments of inertia of the pendulum
are identical and the pivot is located on the axis of symmetry of the pendulum
model. This is an assumption usually adopted while facing attitude control
problems for underactuated geo-stationary spacecraft [DT00]. Indeed, when an
underactuated satellite is not symmetric with respect to the unactuated axis,
the whole system does not satisfy Broketts condition [BMS83] for continuous
feedback stabilization.
In this chapter we develop controllers for a 3D pendulum, under the above as-
sumption. Moreover, only two independent control moments are assumed to
act on the pendulum. In particular, we assume that there is no control moment
about the axis of symmetry of the pendulum.
We rst present a reduced model that forms the basis for controller design and
closed-loop analysis. Such a model follows directly from equation (1.12) and
equation (1.13) when inertia matrix has the form J = diag(J
t
, J
t
, J
a
). It is
based on a global representation of the reduced attitude. This is an important
feature of the development. Indeed, it avoids the use of Eulers angles and other
non-global attitude representations, leading to singularities, as seen in Chapter
2.
Then, we treat asymptotic stabilization of an equilibrium of the reduced equa-
tions of motion of the 3D pendulum. This corresponds to stabilization of a
relative equilibrium of the 3D pendulum. The relative equilibrium consists of a
specied constant value of the reduced attitude and an angular velocity that is
a pure spin about the pendulums axis of symmetry. It leads to the development
of controllers that asymptotically stabilize the hanging relative equilibrium, the
development of controllers that asymptotically stabilize the inverted relative
equilibrium, and for the special case that there is zero angular velocity about
the axis of symmetry of the pendulum, the development of controllers that as-
ymptotically stabilize an arbitrary reduced equilibrium.
We suggest a control technique based on potential shaping and damping injec-
tion as presented in Chapter 4. It is eective both for the case when the angular
velocity component about the axis of symmetry is nonzero and for the case when
it is zero. In the rst case, our control results can be compared with other results
in the literature on stabilization of heavy spinning top [SLRM92], [WTCB94],
[WCB95]. In the second case, our control results can be compared with other
results in the literature on stabilization of spherical pendulum [SEL99], [SEL04].
In all of these cases, our proposed stabilization results are new.
Finally, we depict simulations for all the illustrated cases. Simulation results
conrm that the proposed control technique provides near-global asymptotic
stabilization in a direct way using relatively simple nonlinear controllers. This
approach avoids the articial need to develop a swing-up controller, a locally
asymptotically stabilizing controller, and a strategy for switching between the
two [AF00], [SEL04].
63
5.1 Global Coordinates Model of the 3D Axially Sym-
metric Pendulum
In this section we introduce a reduced model for the controlled 3D axially sym-
metric pendulum. We also recall some important stability properties of the
uncontrolled 3D axially symmetric pendulum, introduced in Chapter 2.
Since the pendulum is assumed to be axially symmetric, there is no loss of
generality in assuming that the inertia matrix is J = diag(J
t
, J
t
, J
a
). As de-
scribed in Chapter 1, let = (0, 0, l)
T
denote again the vector from the pivot
to the center of mass of the pendulum, where l is a nonzero scalar. Moreover,
let = (
x
,
y
,
z
)
T
denote the angular velocity vector of the pendulum, and
= (
x
,
y
,
z
)
T
the reduced attitude vector of the pendulum, both expressed
in body-xed coordinates.
Taking into account the moment due to gravity and the control moments, Eulers
equations in scalar form for the rotational dynamics of the 3D axially symmetric
pendulum, are
J
t

x
= (J
t
J
a
)
z
y
mgl
y
+u
x
J
t

y
= (J
a
J
t
)
z
x
+mgl
x
+u
y
J
a

z
= 0
(5.1)
Here, u
x
and u
y
denote the control moments. Besides kinematics models illus-
trated in Chapter 2 in terms of Eulers angles, here we describe the kinematics
of the symmetric pendulum in terms of reduced attitude vector components.
Hence, we expand equation (1.13), obtaining the following three scalar dieren-
tial equations
x
=
y
y
=
x
z
+
z
z
=
x
x
(5.2)
Since || = 1, the motion of the 3D axially symmetric pendulum evolves on
R
3
S
2
.
It is clear that the above equations cannot be asymptotically stabilized in any
meaningful sense. Clearly, the third equation in (5.1) implies that the angular
velocity component about the axis of symmetry of the pendulum is
z
= c (5.3)
where c is a constant. Reduction
1
in this case is easily achieved by ignoring the
third equation in (5.1) and substituting equation (5.3) into (5.2). This leads to
the reduced dynamics equations
J
t

x
= c(J
t
J
a
)
y
mgl
y
+u
x
J
t

y
= c(J
a
J
t
)
x
+mgl
x
+u
y
(5.4)
and the reduced kinematics equations
x
= c
y
y
= c
x
+
z
z
=
x
x
(5.5)
Equations (5.4)(5.5) form a global-coordinates mathematical description of an
axisymmetric 3D pendulum, with two control torques. It can be shown that this
representation is equivalent to model (2.2). As a consequence, for any constant
c, the motion of the axially symmetric 3D pendulum can be viewed as evolving
on R
2
S
2
.
The uncontrolled equations (5.4)(5.5) have two dierent equilibrium solutions,
namely
x
=
y
= 0, =
h
= (0, 0, 1)
T
, and
x
=
y
= 0, =
i
=
(0, 0, 1)
T
. Note that an equilibrium of (5.4)(5.5) corresponds to a relative
equilibrium of equations (1.12) and (1.13). It consists of a pure spin of the 3D
pendulum about its axis of symmetry. Only if the constant c = 0, a relative
equilibrium solution of (5.4)(5.5) is an ordinary equilibrium solution of the 3D
pendulum.
Thus, in the general case of c , = 0, an axially symmetric 3D rigid pendulum,
without any control torque about its axis of symmetry, spins steadily. Con-
sequently, only stabilization of its sleeping motions, i.e. inverted and hanging
conguration, makes sense. Therefore, it is important to remind the result
below, following from Proposition 2.4 and from the analysis in [GPS80].
Proposition 5.1 Consider the 3D axially symmetric pendulum given by equa-
tions (5.4)(5.5). Then the hanging equilibrium is stable in the sense of Lya-
1
Consider again the reduction theory described in [MR99] and in [MRS00]
65
punov for all c R and the inverted equilibrium is stable in the sense of Lya-
punov if and only if J
2
a
c
2
4mglJ
t
.
This background provides motivation for study of controllers that asymptoti-
cally stabilize one of the natural equilibrium solutions, either the hanging equi-
librium or the inverted equilibrium, or asymptotically stabilize a new equilib-
rium induced by the control action.
5.2 Stabilization of the Hanging Equilibrium of the
Symmetric Spinning Top
In this section we assume that the constant angular velocity is c ,= 0. In this
case, the 3D axially symmetric pendulum described by equations (5.4)(5.5) is
eectively a heavy top, as described in Chapter 2. Hence, that terminology is
used in this section. We propose two classes of feedback controllers that asymp-
totically stabilize the hanging equilibrium of the reduced model described by
equations (5.4)(5.5). The rst result is based on feedback of the angular veloc-
ity of the top. The second result is based on feedback of both angular velocity
and the reduced attitude of the top. In each case, the domain of attraction of
the hanging equilibrium is shown to be nearly-global.
First, we consider controllers based on feedback of the angular velocity, of the
form
u
x
=
x
(
x
)
u
y
=
y
(
x
)
(5.6)
where
x
: R R and
y
: R R are smooth functions satisfying the following
sector inequalities with
2

1
> 0
1
[x[
2
x
x
(x), x
y
(x)
2
[x[
2
, x R (5.7)
and
x
(0),
y
(0) > 0.
Theorem 5.1 Consider the 3D axially symmetric pendulum given by the equa-
tions (5.4)(5.5). Let (
x
,
y
) be smooth functions satisfying (5.7) and choose
u
x
and u
y
as in (5.6). Then the hanging equilibrium of (5.4)(5.5) is asymptoti-
cally stable. Furthermore, for every (0, 2mgl), all solutions of the closed-loop
system given by (5.4)(5.5) and (5.6), such that (
x
(0),
y
(0), (0)) H, where
H =
_
(
x
,
y
, ) (R
2
S
2
) :
1
2
_
J
t
(
2
x
+
2
y
) +mgl|
h
|
2
_
2mgl
_
(5.8)
satisfy (
x
(t),
y
(t), (t)) H, t 0, and lim
t
x
(t) = 0, lim
t
y
(t) = 0 and
lim
t
(t) =
h
.
Proof. Consider the closed-loop system given by (5.4)(5.5) and (5.6). We
propose the following candidate Lyapunov function
V (
x
,
y
, ) =
1
2
_
J
t
(
2
x
+
2
y
) +mgl|
h
|
2
_
(5.9)
Note that the Lyapunov function is positive denite in R
2
S
2
and V (0, 0,
h
) =
0. Furthermore, the derivative

V along a solution of the closed-loop system is
V (
x
,
y
, ) = J
t
(
x

x
+
y

y
) +mgl(
h
)
T

= c(J
t
J
a
)
x
y
mgl
y
x
+c(J
a
J
t
)
x
y
+mgl
x
y
+u
x
x
+u
y
y
+mgl[c
x
y
c
x
y
+
y
x
+ (
z
1)(
x
x
)]
= mgl
y
x
+mgl
x
y
+u
x
x
+u
y
y
+mgl[
x
y
+
y
x
+
x
y
+
y
x
]
= u
x
x
+u
y
y
=
x
x
(
x
)
y
y
(
y
)
1
(
2
x
+
2
y
) 0
where the last inequality follows from (5.7). Thus, V is positive denite and

V
is negative semidenite on R
2
S
2
.
Next, consider the sub-level set given by
H =
_
(
x
,
y
, ) (R
2
S
2
) : V (
x
,
y
, ) 2mgl
_
Note that the compact set H contains the hanging equilibrium (0, 0,
h
). Since
V (
x
,
y
, ) 0 on H, all solutions such that (
x
(0),
y
(0), (0)) H satisfy
(
x
(t),
y
(t), (t)) H for all t 0. Thus, H is an invariant set for the solu-
tions of the closed-loop system.
Furthermore, from LaSalles invariant set theorem, we obtain that the solu-
tions satisfying (
x
(0),
y
(0), (0)) H converge to the largest invariant set
in (
x
,
y
, ) H : (
x
,
y
) = (0, 0). Next,
x

y
0 implies that
x
=
y
= 0 and

z
= 0, that is
z
= 1. Thus, as t , either =
h
or
=
i
. However, since (0, 0,
i
) , H it follows that =
h
as t .
67
Therefore,
h
is an asymptotically stable equilibrium of the closed-loop system
given by (5.4)(5.5) and (5.6), with a guaranteed domain of attraction H.
It is straightforward to notice that the result stated above holds true also for
the complete model given by (1.12)(1.13). Indeed, the candidate Lyapunov
function
V
1
(, ) =
1
2
[
T
J +mgl(
h
)
T
(
h
)] J
a
2
z
similar to (3.6), introduced in Chapter 3, is such that V
1
(, ) = V (
x
,
y
, )
(, ), where V (
x
,
y
, ) is given by (5.9). Hence, V
1
(, ) has a minimum
in the hanging equilibrium (, ) = ([0, 0, c]
T
,
h
) that is asymptotically stable
with guaranteed domain of attraction H.
We next strengthen the conclusion of above theorem, showing that the domain
of attraction of the hanging equilibrium is almost global.
Theorem 5.2 Consider the 3D axially symmetric pendulum given by the equa-
tions (5.4)(5.5). Let (
x
,
y
) be smooth functions satisfying (5.7) and choose
u
x
and u
y
as in (5.6). Then, all solutions of the closed-loop system given by
(5.4)(5.5) and (5.6), such that (
x
(0),
y
(0), (0)) R
3
(S
2
/) satisfy
lim
t
x
(t) = 0, lim
t
y
(t) = 0 and lim
t
(t) =
h
. Here /, a set of Lebesgue
measure zero, is the stable manifold of the closed loop inverted equilibrium.
Proof. Since, the proof is very similar to process leading to theorem 3.1, here
we present only an outline of the proof. Denote
^ =
_
(
x
,
y
, ) (R
2
S
2
) :
1
2
_
J
t
(
2
x
+
2
y
) +mgl|
h
|
2
_
2mgl
_
Then, it can be shown that all solutions of the closed-loop system (5.4)(5.5) and
(5.6), satisfying (
x
(0),
y
(0), (0)) ^(0, 0,
i
), enter the set H dened
in theorem 5.1 in nite time, for some > 0. Then, from theorem 5.1 and the
denition of ^, we note that for every (0, 2mgl) and (
x
(0),
y
(0), (0))
^(0, 0,
i
), (t) 0 and (t)
h
, as t . Next, since
^ =
_
(0,2mgl)
_
H
_
^
_
it follows that all solutions satisfying (
x
(0),
y
(0), (0)) ^(0, 0,
i
) con-
verge to the hanging equilibrium.
Furthermore, it can be shown that all solutions of the closed-loop (5.4)(5.5)
under (5.6), enter the set ^ in nite time. Hence, all solutions either converge
to the inverted equilibrium, or the hanging equilibrium. Thus, it is sucient
to show that the stable manifold of the inverted equilibrium (0, 0,
i
), has di-
mension less than the dimension of (R
2
S
2
), i.e. four, since all other solutions
converge to the hanging equilibrium.
Next, the equilibrium (0, 0,
i
) of the closed-loop system is unstable as shown
in Chapter 1. Using linearization it can be shown that its instability is hyper-
bolic with nontrivial stable and unstable manifolds. Thus, from theorem 3.2.1
in [GH83], it follows that the dimension of the stable manifold of (0,
i
) is less
than four, so that the domain of attraction in theorem 5.2 is (R
2
S
2
)/,
being / the stable manifold. Hence, the result follows.
Theorem 5.2 provides conditions under which the heavy spinning top is made
asymptotically stable by feedback of angular velocity. Any controller of the form
(5.6) can be viewed as providing damping injection. In theorem 5.1, the hang-
ing equilibrium has a domain of attraction that is easily computed. In theorem
5.2, the domain of attraction is almost global. We are not aware of any prior
literature on stabilizing the hanging equilibrium of an heavy top.
Next, we consider controllers based on feedback of the angular velocity and the
reduced attitude. These controllers provide more design exibility than the con-
trollers that depend on angular velocity only; hence they can provide improved
closed loop performance.
Theorem 5.3 Consider the 3D axially symmetric pendulum given by equations
(5.4)(5.5) with c ,= 0. Let : [0, 1) R be a C
1
monotonically increasing
function such that (0) = 0 and (x) as x 1. Furthermore, let (
x
,
y
)
be smooth functions satisfying inequality given by (5.7). Choose
u
x
=
x
+
x
() c(J
t
J
a
)
y
+J
t
(
z
1)(c
x
+
z
x
)
x
()
+
(
1
4
(
z
1)
2
) +mgl
y
u
y
=
y
+
y
() c(J
a
J
t
)
x
+J
t
(1
z
)(c
y
+
z
y
)
y
()
+
(
1
4
(
z
1)
2
) mgl
x
(5.10)
69
where = (
z
1)
y
and = (1
z
)
x
. Then, (
x
,
y
, ) = (0, 0,
h
) is an
equilibrium of the closed-loop given by (5.4)(5.5) and (5.10) that is asymptot-
ically stable with domain of attraction R
2
(S
2
i
).
Proof. Consider the system represented by (5.4)(5.5) and (5.10). We propose
the following candidate Lyapunov function along a solution of the closed-loop
V (
x
,
y
, ) =
J
t
2
[
x
x
()]
2
+
J
t
2
[
y
y
()]
2
+ 2
_
1
4
(
z
1)
2
_
2
S
2
with
V (0, 0,
h
) = 0.
Suppose that (
x
(0),
y
(0), (0)) ,= (0, 0,
i
). Computing the derivative of the
Lyapunov function along a solution of the closed-loop, we obtain
V [
x
x
()]
2
[
y
y
()]
2
_
1
4
(
z
1)
2
_
(
z
1)
2
(
2
x
+
2
y
) 0
Thus,

V is negative semidenite and hence, each solution remains in the com-
pact set / = (
x
,
y
, ) R
2
S
2
: V (
x
,
y
, ) V (
x
(0),
y
(0), (0)).
Since

V is negative semidenite and is monotonic with
(x) ,= 0 if x ,= 0, we
obtain that 0, 0,
x

x
(0) = 0 and
y

y
(0) = 0 as t . The
last two properties follow from (5.7) and the Sandwiching theorem for the limit
of a function. Furthermore, by LaSalles invariant set theorem, each solution
converges to the largest invariant set / (
x
,
y
, ) / :
x
=
y
= 0, =
0, = 0. Since any closed-loop solution of (5.4)(5.5) and (5.10) in /satises
x

y
0, we obtain that the solution also satises
z
= constant.
Next, = = 0 yields either
z
= 1, in which case =
h
, or it yields
x
= 0
and
y
= 0 and hence, =
h
or =
i
. However, since V (
x
(t),
y
(t), (t)) ,=
V (
x
(0),
y
(0), (0)), therefore (t) , =
i
for all t > 0. Thus
i
, /. Hence,
the only solution of the closed-loop system contained in the invariant set / is
x
=
y
= 0 and =
h
.
Result in theorem 5.3 follows from the observation that control model (5.4)
(5.5) is input-output passive if the couple (, ) is taken as the output. It
provides conditions under which the hanging equilibrium of the heavy top can
be made almost-globally stable by feedback of the angular velocity and feedback
of the reduced attitude. Any controller of the form (5.10) requires knowledge of
the axial and the transverse principal moments of inertia, mass, location of the
center of mass and spin rate of the rigid body. Notice that, in equation (5.10),
plays the role described in theorem 4.2, i.e. it shapes the potential energy of the
closed-loop system. Namely, it pushes the reduced attitude vector away from
the unstable equilibrium. In the case of the hanging equilibrium stabilization
described above, it is not necessary. Indeed, the same role could also be played
by gravity terms, as in theorem 5.2. Here, gravity is canceled by the control
law and the choice of u
x
and u
y
as in equation (5.10) is made so as to provide
a desired rate of convergence to the hanging equilibrium. Moreover, it is useful
for introducing the subsequent stabilization of the inverted equilibrium of the
heavy top, where canceling the gravity and shaping the potential is necessary.
5.3 Stabilization of the Inverted Equilibrium of the 3D
Axially Symmetric Pendulum
We consider the stabilization of the inverted equilibrium of the 3D axially sym-
metric rigid pendulum. First, as in previous section, we assume that the con-
stant c ,= 0, so that the 3D axially symmetric pendulum described by equations
(5.4)(5.5) is an heavy top. Then, we assume that c = 0. In this case, equations
(5.4)(5.5) describe eectively a spherical pendulum as illustrated in Chapter 2.
We propose feedback controllers that almost-globally asymptotically stabilize
the inverted equilibrium of the reduced model. The result is based on feedback
of both angular velocity and the reduced attitude of the top.
(5.4)(5.5) with c ,= 0. Let : [0, 1) R be a C
1
x
,
y
)
71
be smooth functions satisfying inequality given by (5.7). Choose
u
x
=
x
+
x
() c(J
t
J
a
)
y
+J
t
(
z
+ 1)(c
x
+
z
x
)
x
()
+
(
1
4
(
z
+ 1)
2
) +mgl
y
u
y
=
y
+
y
() c(J
a
J
t
)
x
J
t
(1 +
z
)(c
y
+
z
y
)
y
()
+
(
1
4
(
z
+ 1)
2
) mgl
x
(5.11)
where = (1 +
z
)
y
and = (1 +
z
)
x
. Then, (
x
,
y
, ) = (0, 0,
i
) is an
equilibrium of the closed-loop given by (5.4)(5.5) and (5.11) that is asymptot-
ically stable with domain of attraction R
2
(S
2
h
).
Proof. Consider the system represented by (5.4)(5.5) and (5.11) and consider
the following candidate Lyapunov function
V (
x
,
y
, ) =
J
t
2
[
x
x
()]
2
+
J
t
2
[
y
y
()]
2
+ 2
_
1
4
(
z
+ 1)
2
_
2
S
2
with
V (0, 0,
i
) = 0.
Suppose that (
x
(0),
y
(0), (0)) R
2
(S
2
h
). Computing the derivative
of the Lyapunov function along a solution of the closed-loop, we obtain
V [
x
x
()]
2
[
y
y
()]
2
_
1
4
(
z
+1)
2
_
(
z
+1)
2
(
2
x
+
2
y
) 0
Thus,

pact set / = (
x
,
y
, ) R
2
S
2
: V (
x
,
y
, ) V (
x
(0),
y
(0), (0)).
The remainder of the proof follows exactly the arguments used in theorem 5.3.
The only solution of the closed-loop system (5.4)(5.5) and (5.11) such that
(
z
+ 1)
y
0, (
z
+ 1)
x
0,
x

x
(0) = 0,
y

y
(0) = 0 as t ,
is the inverted equilibrium (
x
,
y
, ) = (0, 0,
i
).
Theorem 5.4 provides conditions under which the inverted equilibrium of the
heavy spinning top can be made asymptotically stable by feedback of the angular
velocity and feedback of the reduced attitude. The inverted equilibrium of the
top is guaranteed to have an almost-global domain of attraction . These results
can be compared with the extensive literature on stabilization of the upward
sleeping motion of the top. See, e.g. [SLRM92], [WTCB94], [WCB95] and
related references. Here, the main dierence with respect to previous works
is that the spinning top is retained in the more general framework of the 3D
pendulum. Hence, its motion can be described by global coordinates in the form
of (5.5). Consequently, the results in Theorem 5.4 are substantially dierent
from any of these cited results.
We consider now the case where the axially symmetric 3D pendulum does not
spin around the unactuated axis, i.e.
z
= c = 0. As reminded above, in this
case the 3D pendulum is eectively a spherical pendulum and model (5.4)(5.5)
reduces easily to the dynamics equation
J
t

x
= mgl
y
+u
x
J
t

y
= mgl
x
+u
y
(5.12)
and to the kinematics equation
x
=
z
y
=
z
z
=
x
x
(5.13)
It is easy to show that model (5.12)(5.13) is equivalent to representation (2.3)
introduced in Chapter 2. Indeed, it is sucient to express in terms of Eulers
angles. Moreover, (5.12)(5.13) represents a description of the 2D pendulum
motion, alternative to (2.5), that is the one commonly used in literature while
facing 2D pendulum control problems. It can be shown that the relation be-
tween the above model and (2.5) is nonlinear and non smooth. This result
could be expected, noticing that (2.3) and (2.5) have singularities in dierent
congurations. In particular, (2.5) is not dened in the inverted equilibrium, as
described in detail in Chapter 2. It leads to several control problems that here
are simply avoided using model (5.12)(5.13).
(5.4)(5.5) with c = 0. Let : [0, 1) R be a C
1
x
,
y
)
be smooth functions satisfying inequality given by (5.7). Assume (0) = c = 0,
and let
(1 +
z
)
x
(1 +
z
)
y
(5.14)
73
u
x
= mgl
y
+J
t
x
() (
x
x
()) +
_
1
4
(
z
+ 1)
2
_
u
y
= mgl
x
+J
t
y
()

(
y
y
()) +
_
1
4
(
z
+ 1)
2
_
(5.15)
where and

are obtained by dierentiating (5.14) and substituting from (5.5).
Then, (0, 0,
i
) is an equilibrium of the closed-loop given by (5.4)(5.5) and
(5.15) that is asymptotically stable with domain of attraction R
2
(S
2
h
).
Proof. Consider the system represented by (5.4)(5.5) and (5.15). Since
c = 0, (5.4)(5.5) reduces to (5.12)(5.13). We propose the following candidate
Lyapunov function
V (
x
,
y
, ) =
J
t
2
[
x
x
()]
2
+
J
t
2
[
y
y
()]
2
+ 2
_
1
4
(
z
+ 1)
2
_
2
S
2
with
V (0, 0,
i
) = 0. The derivative of the Lyapunov function along a solution of the
closed-loop, is
V [
x
x
()]
2
[
y
y
()]
2
_
1
4
(
z
+ 1)
2
_
(
2
+
2
) 0
Thus,

pact set / = (
x
,
y
, ) R
2
S
2
: V (
x
,
y
, ) V (
x
(0),
y
(0), (0)).
Next, since

V is semidenite and from properties of (), we obtain that 0,
0,
x

x
(0) = 0 and
y

y
(0) = 0 as t . The last two equalities
follow from the Sandwiching theorem for the limit of a function.
Furthermore, by LaSalles invariant set theorem, the solution converges to the
largest invariant set / (
x
,
y
, ) / :
x
=
y
= 0, = 0, = 0. Since
any closed-loop solution of (5.12)(5.13) and (5.15) in /satises
x

y
0,
we obtain that the solution also satises = constant.
Next, = = 0 yields either
z
i
, or it yields
x
= 0
and
y
= 0 and hence, =
i
or =
h
x
(t),
y
(t), (t))
V (
x
(0),
y
(0), (0)), therefore (t) ,=
h
for all t 0. Thus (0, 0,
h
) , /.
Hence, the only solution of the closed-loop system contained in the invariant set
/ is
x
=
y
= 0 and =
i
.
Previous theorem provides conditions under which the inverted equilibrium of
the spherical pendulum can be made asymptotically stable by feedback of the
angular velocity and feedback of the reduced attitude. It follows from the obser-
vation that (5.12)(5.13) is input-output passive if and are taken as outputs.
The proposed control law (5.15), injects damping and shapes the potential so
as to generate a closed-loop system whose potential energy has a minimum in
the inverted equilibrium. Notice that (5.15) is similar to (5.11) used in theorem
5.4 for stabilizing the upward sleeping motion of the heavy top. Indeed, both
problems arise from the more general context of the 3D pendulum. On the other
hand, they are substantially dierent from controllers for spherical pendulums
that have appeared in prior literature [SEL99], [SPLE00], [SEL04]. Indeed, our
results provide an almost-globally stabilizing controller that avoids the need to
construct a swing-up controller, a locally stabilizing controller, and a switching
strategy between the two. In this comparative sense, our results are direct and
simple.
5.4 Attitude Control of an Underactuated Geo-Stationary
Spacecraft
Treatment in the last section can be easily extended to the case where the
desired nal attitude of the 3D axisymmetric pendulum is not represented
just by the inverted equilibrium
i
= (0, 0, 1), but by any possible attitude
0
= (
x0
,
y0
,
z0
). Apart of the dierence between whole attitude control
and reduced attitude control, widely described in previous chapters, this prob-
lem consists eectively of the attitude control problem for underactuated geo-
stationary satellites [DT00]. Indeed, removing gravity terms from equations
(5.4)(5.5) gives the rotational model of a symmetric GEO spacecraft, without
actuation around its axis of symmetry.
Thus, we face the problem of controlling any arbitrary reduced attitude of an
axially symmetric 3D pendulum, with
z
= c = 0, viz. no spinning about the
axis of symmetry, and u
z
= 0 viz. no control action about the axis of symme-
try. Notice that, for our purposes, we assume c = 0. If
z
= c ,= 0, one should
rst apply to the system a detumbling control law. It consists of stabilizing the
angular velocity to zero before controlling the attitude. Detumbling controllers
75
for geo-stationary spacecraft have been proposed in prior literature. For the
problem at hand, one possibility is of breaking the symmetry of the system us-
ing proof mass as described in [CM02], then applying the control law described
in [Son00] and nally restoring the symmetry, again by proof mass movement.
We propose a class of controllers which almost globally asymptotically stabilize
any arbitrary orientation of the spherical pendulum on R
2
S
2
using the func-
tions (),
x
() and
y
() introduced above.
Theorem 5.6 Consider the 3D axially symmetric pendulum given by equa-
tions (5.12)(5.13). Let
0
= (
x0
,
y0
,
z0
)
T
be a desired orientation and
: [0, 1) R be a C
1
monotonically increasing function such that (0) = 0
and (x) as x 1. Furthermore, let (
x
,
y
) be smooth functions satis-
fying inequality given by (5.7). Assume (0) = c = 0, and let
(
T
0
1)(
z0
y0
z
)
(
T
0
1)(
x0
z0
x
)
(5.16)
u
x
= mgl
y
+J
t
x
() (
x
x
()) +
_
1
4
(
T
0
1)
2
_
u
y
= mgl
x
+J
t
y
()

(
y
y
()) +
_
1
4
(
T
0
1)
2
_
(5.17)
where and

are obtained by dierentiating (5.16) and substituting from
(5.13). Then, (0, 0,
0
) is an equilibrium of the closed-loop given by (5.12)
(5.13) and (5.17) that is asymptotically stable with domain of attraction R
2
(S
2
0
).
Proof. Consider the system represented by (5.12)(5.13) and the following can-
didate Lyapunov function
V (
x
,
y
, ) =
J
t
2
[
x
x
()]
2
+
J
t
2
[
y
y
()]
2
+ 2
_
1
4
(
T
0
1)
2
_
Note that the above Lyapunov function is positive denite and V (0, 0,
0
) = 0.
Furthermore, V (
x
,
y
, ) is radially unbounded on R
2
S
2
. Calculating the
derivative of the Lyapunov function along a solution of the closed-loop (5.12)
(5.13) and (5.17), we obtain
V = [
x
x
()]
2
[
y
y
()]
2
_
1
4
(
T
0
1)
2
_
(
x
() +
y
())
[
x
x
()]
2
[
y
y
()]
2
_
1
4
(
T
0
1)
2
_
(
2
+
2
) 0
Thus,

V is negative semidenite and hence, each solution remains in the compact
set / = (
x
,
y
, ) R
2
S
2
: V (
x
,
y
, ) V (
x
(0),
y
(0), (0)). Next,
since

V is semidenite and from properties of (), we obtain that 0, 0,
x

x
(0) = 0 and
y

y
(0) = 0 as t . The last two equalities follow
from (5.7) and from the Sandwiching theorem for the limit of a function.
Furthermore, by LaSalles invariant set theorem, the solution converges to the
largest invariant set / (
x
,
y
, ) / :
x
=
y
= 0, = 0, = 0. Since
any closed-loop solution of (5.12)(5.13) and (5.17) in /satises
x

y
0,
we obtain that the solution also satises = constant.
Next, 0 yields either
T
0
0
, or it yields
z0
y
=
y0
z0
x
=
x0
z
(5.18)
Since || = |
0
| = 1, taking the square of left hand side and right hand side
of (5.18), gives
2
z0
(1
2
z
) =
2
z
(1
2
z0
)
that means
2
z
=
2
z0
and hence,
z
=
z0
. Substituting
z
=
z0
in equa-
tion (5.18), we obtain that =
0
x
(t),
y
(t), (t))
V (
x
(0),
y
(0), (0)), therefore (t) ,=
0
for all t 0. Thus (0, 0,
0
) , /.
Hence, the only solution of the closed-loop system contained in the invariant set
/ is
x
=
y
= 0 and =
0
.
Theorem 5.6 shows that control law (5.17) stabilize an arbitrary reduced atti-
tude
0
of an underactuated axisymmetric spacecraft, whose attitude kinematics
is described by (5.13). Indeed, function in (5.17) shapes the closed-loop po-
tential energy so as to guarantee that the attitude cannot approach unstable
equilibrium
0
. On the other hand,
x
and
y
provide suitable damping in-
jection. As a consequence, lim
t
x
(t) = 0, lim
t
y
(t) = 0 and lim
t
(t) =
o
for any initial condition except of (
x
(0),
y
(0), (0)) = (0, 0,
0
). This re-
sult follows in a direct and easy way, retaining the problem in the framework
of controlling the axisymmetric 3D rigid pendulum. The above result can be
compared with other techniques presented in literature for controlling the atti-
tude of underactuated geo-stationary satellites. See, e.g. [GE95], [CLT00] and
77
[DT00].
It is worth pointing out that (5.17), which depends on two control inputs, can
be easily extended to the case where the 3D pendulum is fully actuated and
asymmetric, as discussed in Chapter 4. In this case J = (J
1
, J
2
, J
3
)
T
and the
system is described by model (3.2). Thus, dening
(
T
0
1)(
y0
x0
y
)
along with (5.16), i.e. (, , )
T
= (
T
0
1)(
0
), introducing another
smooth function
z
: R R, satisfying (5.7), and taking
u
x
= J
1
x
() (
x
x
()) +
_
1
4
(
T
0
1)
2
_
+mgl
y
u
y
= J
2
y
()

(
y
y
()) +
_
1
4
(
T
0
1)
2
_
mgl
x
u
z
= J
3
z
() (
z
z
()) +
_
1
4
(
T
0
1)
2
_
(5.19)
we obtain that (, ) = (0,
0
) is an almost-global equilibrium point of the
closed-loop system dened by (3.2) and (5.19), with domain of attraction R
3
(S
2
0
). The proof of this result is easily obtained proceeding as in theorem
5.6, with the following candidate Lyapunov function
V (, ) =
J
1
2
[
x
x
()]
2
+
J
2
2
[
y
y
()]
2
+
J
3
2
[
z
z
()]
2
+2
_
1
4
(
T
0
1)
2
_
The dierence between control (5.19) and control (4.5) lies in the rst two terms
of (5.19), that only mean a dierent damping injection with respect to (4.5).
Hence, properties of (4.5) are maintained. In particular, (5.19) utilizes the same
metric on S
2
for representing the distance between and
0
. Thus, unwinding
phenomenon is still avoided.
Therefore, the attitude of a geo-stationary spacecraft can be controlled using
(5.19) both in the fully and in the underactuated case. In the rst situation the
satellite can be asymmetric with general inertia matrix J. In presence of failures,
the satellite must be made symmetric with respect to the unactuated axis, e.g.
using proof mass movements. We are not aware of other control techniques
proposed in prior literature, able of controlling the attitude of a spacecraft both
in the absence and in the presence of actuator failures.
Simulation results on the application of the control techniques described above,
are presented in the following section.
5.5 Simulation results
In this section we present simulation results on the application of the control
techniques described in the previous sections. First, we stabilize the inverted
equilibrium of a heavy spinning top, using (5.11). Then, we control an arbitrary
attitude of a 2D pendulum, assuming
z
= 0 and applying control law (5.17).
We choose again
(x) = k ln(1 x)
where k > 0, and
i
(x) = p
i
x
where p
i
are positive constants.
Consider model (5.4)(5.5) where m = 140 kg, = (0, 0, 0.5)
T
m and J =
diag(40, 40, 50) kgm
2
. These parameters correspond to the ones of the TACT,
once it is made symmetric with respect to the vertical. It can be done applying
suitable weights to the external plates.
Let u
x
and u
y
be given by controller (5.11) where
x
() = 3 and
y
() = 3.
Moreover, we choose k = 5 and we assume
z
= c = 1. Hence, (5.4)
(5.5) represents eectively a heavy top, that spins steadily at velocity
z
= 1
around its axis of symmetry. The following gures show that in closed-loop
system (5.4)(5.5) and (5.11) the inverted equilibrium
i
= (0, 0, 1)
T
is as-
ymptotically stable. Initial conditions are (0) = (1, 3, 1)
T
rad/s and (0) =
(0.5627, 0.8501, 0.0033)
T
. Notice that control law (5.11) is activated while the
system is moving.
Simulation results in Figures 5.1 and 5.2 show that (t) (0, 0, 1)
T
, and
(t)
i
as t . Indeed, the two control components u
1
and u
2
in
(5.11), shape the energy of the symmetric top, so as to generate a closed-loop
system whose potential has a minimum in the inverted equilibrium. On the
other hand, the control law does not act on the third component of the angu-
lar velocity. This is evident from Figure 5.3, where angular momentum with
respect to the vertical converges to
T
J
i
= 50. The whole trajectory of the
closed-loop system, under controller (5.11), is depicted in Figure 5.4. There, the
motion is drowned from the inertial frame point of view. Remember that body
and inertial versors are related by B(t) = R(t)
T
I(t) and (t) = R(t)
T
e
3
.
79
0 20 40 60 80 100 120 140 160 180 200
4
2
0
2
1

[
r
a
d
/
s
]
Angular Velocity
0 20 40 60 80 100 120 140 160 180 200
2
0
2
4
2

[
r
a
d
/
s
]
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1
1.5
2
3

[
r
a
d
/
s
]
time [s]
Figure 5.1. Evolution of the angular velocity of the symmetric heavy top in the
body frame.
0 20 40 60 80 100 120 140 160 180 200
0
0.5
1
1
Components of in {B}
0 20 40 60 80 100 120 140 160 180 200
0.5
0
0.5
1
2
0 20 40 60 80 100 120 140 160 180 200
1.5
1
0.5
0
0.5
3
time [s]
frame.
Hence, in order to depict the motion in I, it is necessary to dene an initial
condition also for the rotation matrix R between I and B. In the present
0 20 40 60 80 100 120 140 160 180 200
200
100
0
100
200
300
400
500
600
700
800
Vertical Momentum Total Energy
time [s]
h

E
Momentum
Total Energy
Figure 5.3. Vertical angular momentum (solid line) and total energy (dash line) of
the symmetric top.
example we have selected the following initial rotation matrix between I and
B
R(0) =
_
_
_
_
_
0.8389 0.5191 0.1635
0.1373 0.0889 0.9865
0.5267 0.8501 0.0033
_
_
_
_
_
For the reason explained above, even if we control the third row of R, i.e. the
direction of gravity in B, Figure 5.4 illustrates its third column evolution, i.e.
evolution of the vector from the pivot to the center of mass of the 3D symmetric
pendulum. Notice that it converges to the upward vertical position, as expected.
Consider now the case where the axisymmetric 3D rigid pendulum does not
spin about its axis of symmetry. In this case, model (5.12)(5.13) describes
eectively a spherical pendulum and, removing gravity terms, an underactu-
ated geo-stationary spacecraft. Let u
x
and u
y
be given by controller (5.17)
where,
x
() = 3 and
y
() = 3. The following gures show the re-
sult of the simulation when initial condition is (0) = (1, 3, 0)
T
rad/s and
(0) = (0.5627, 0.8501, 0.0033)
T
and where parameters are the same of previ-
81
ous simulation:m = 140 kg, = (0, 0, 0.5)
T
m and J = diag(40, 40, 50) kg m
2
.
The nal desired attitude is
0
= (1/
5, 0, 2/
5). As expected, angular ve-

1
0.5
0
0.5
1
1
0.5
0
0.5
1
1
0.5
0
0.5
1
y
3D Pendulum Motion: /|| vector in {I}
x
z
Figure 5.4. Closed-loop trajectory of the symmetric top in the inertial frame.
locity 0 and reduced attitude
0
as t . This is illustrated in
Figure 5.5 and Figure 5.6 Moreover, the unwinding phenomenon is avoided
thanks to the term
_
1
4
(
T
0
1)
2
_
in (5.17). It guarantees that the controller
push the pendulum in the correct angular direction, in such a way that
(
0
, (t)) (
0
, (0))
It is clear observing Figure 5.7. The whole trajectory is depicted in Figure 5.8.
Again, in order to represent the vector between the pivot and the center of mass
of the pendulum, it is necessary to dene an initial rotation matrix R(0). We
select the same matrix of previous example.
5.6 Conclusions
In this chapter we have introduced control laws for an axisymmetric 3D rigid
pendulum, assuming the presence of only two control torques. This problem is
extremely important in aerospace literature, since the related model describes
0 50 100 150 200 250 300
0.5
0
0.5
1
1.5
1

[
r
a
d
/
s
]
Angular Velocity
0 50 100 150 200 250 300
0
1
2
3
2

[
r
a
d
/
s
]
0 50 100 150 200 250 300
1
0.5
0
0.5
1
3

[
r
a
d
/
s
]
time [s]
Figure 5.5. Evolution of the angular velocity of the 2D pendulum in the body
frame.
0 50 100 150 200 250 300
0.5
0
0.5
1
1
Components of in {B}
0 50 100 150 200 250 300
0
0.5
1
2
0 50 100 150 200 250 300
0.5
0
0.5
1
3
time [s]
frame.
an underatuated geo-stationary spacecraft, by simply removing gravity terms.
First, we have proposed a control law based on angular velocity feedback and we
83
0 50 100 150 200 250 300
0
20
40
60
80
100
120
Angle between Gravity and Desired Attitude
time [s]

[
d
e
g
]
0
for (0) = 0.
1
0.5
0
0.5
1
1
0.5
0
0.5
1
1
0.5
0
0.5
1
y
3D Pendulum Motion: /|| vector in {I}
x
z
Figure 5.8. Closed-loop trajectory of the 2D pendulum in the inertial frame.
have shown that it can easily stabilize the hanging equilibrium of the system,
both in the case where the pendulum does or does not spin steadily. In the
rst case the controller stabilizes the vertical conguration of a top under the
table, i.e. its downward conguration. Surprisingly, this problem has not been
studied in previous literature.
Then, we have presented a controller able to stabilize the sleeping motion of a
spinning top. The proposed control technique appears simple and direct, from
the 3D pendulum theory presented above. Finally, we have proposed a control
law that stabilizes an arbitrary attitude of the axisymmetric pendulum, when
z
= 0. It is based on damping injection and potential shaping, thanks to the
selection of suitable outputs that render closed-loop system, input-output pas-
sive. This controller has been generalized to the asymmetric case, producing a
control law useful both in the fully and in the underactuated case.
The possibility of designing simple continuous state-feedback controllers for sta-
bilizing any arbitrary attitude of 2D and 3D rigid bodies, appears a prominent
result in mechanics and aerospace literature. This is possible thanks to the idea
of viewing all the related models as special cases of the 3D pendulum. Its global
attitude representation, avoids the presence of singularities and the consequent
articial need to develop swing-up controllers, local controllers, and strategies
for switching between them.
85
Part II
MEO and LEO Spacecraft
Attitude Control
87
Chapter
6
Medium and Low Earth Orbit
Satellites
The 3D pendulum model presented in the rst part of this dissertation describes
the evolution of geo-stationary satellites. Since, geo-stationary spacecraft travel
on orbits, whose period can be made exactly equal to the time Earth takes
for rotating once about its axis, for an observer on Earth only attitude motion
makes sense. Thus, apart of gravity terms, controlling the attitude of a 3D
pendulum is equivalent to control the attitude of a geosynchronous satellite.
Beyond the huge literature on geo-stationary spacecraft [Hug86], [Kap76], [Sid00],
developed from the fties to present time, in the last few decades large eorts
have been made for studying the motion of satellites traveling on orbits at lower
distance from Earth. Indeed, progresses in telecommunications have suggested
the use of small space vehicles on Medium Earth orbits
1
(MEO) and Low Earth
orbits
2
(LEO). MEO and LEO satellites have produced a signicant reduction
of the costs related to launches, maneuvers and maintenance.
In this chapter we present a mathematical model for MEO and LEO spacecraft
dynamics. We rst propose a novel attitude model for satellites orbiting at ve-
1
Orbits between 3.000 and 30.000 km above the Earths surface
2
Orbits between 200 and 2.000 km above the Earths surface
89
locity
0
relative to Earth. It is designed with respect to an orbital reference
frame that spins at velocity
0
about its third axis. Small angle maneuvers are
taken into account. Then, we complete the mathematical model by presenting
the Hills model. Hills model describes the position of a satellite with respect
to a reference point on its orbit. It originates from the two body problem, well
known in literature [Sid00], [BMW71], [TIH02]. The need of a position model,
along with an attitude model, is due to the fact that in MEO and LEO orbits
the motion is not geosynchronous. Thus, describing the position is necessary
for modeling the whole system state.
Finally, we introduce the problem of controlling the attitude and the position
of a LEO spacecraft. Since there is no mathematical dierence between MEO
and LEO systems, apart of vehicles masses and orbiting velocities, we focus our
attention on LEO satellites. For these systems, involving small spacecraft, the
presence of limited thrusts cannot be left out of consideration.
6.1 LEO Spacecraft Attitude Model
Consider a satellite in Low Earth Orbit, traveling around Earth at a velocity
of
0
rad/s. It consists of a rigid body, with three rotational and three trans-
lational degrees of freedom, with respect to an Earth Centered Inertial (ECI)
reference frame. Following the same arguments of Chapter 1, let B be an
R(t)
{ECI}
0
R(t)
1
R(t)
0
=
Figure 6.1. Reference frame {B} motion with respect to an Earth centered reference
frame {ECI}
orthonormal reference frame aligned with the principal axes of the rigid body.
Hence, attitude can be represented describing the evolution of the rotation ma-
trix R(t) between B and ECI. However, it is immediate to notice that
this way to proceed, that in Chapter 1 leads to equations (1.2) and (1.6), is not
suitable in the present case. Indeed, as depicted in Figure 6.1, B rotates with
respect to ECI, even in the absence of external torques. On the other hand,
when 0, R(t) can be calculated at each time instant, given its initial value
R(t
0
). In fact, in this case R(t) depends only on
0
.
Thus, it is easier to analyze the rotations of B with respect to another ref-
erence frame O, that is not inertial and that orbits at velocity
0
in ECI.
This way to proceed is referred to as the leader following approach [Sid00],
[TIH02]. It means that the rotational and translational motion of the 3D rigid
body can be modeled with respect to a reference frame whose origin is xed at
the center of mass of a virtual satellite. The virtual satellite can be a moving
point on the 3D rigid bodys orbit, or possibly, another spacecraft. In the latter
case, the two satellites consist of a formation ying system with one leader and
one follower. Fleets of satellites ying in LEO orbits, have been collecting great
interest in the last few years [ENS
+
03]. It has been emphasized that they can
substitute large monolithic satellites, drastically reducing costs. Therefore, it is
expected that a great deal of future space missions will be based on formations
of satellites.
For our purposes, an orbiting reference frame O = X, Y, Z is dened. It
is selected with the rst axis in the opposite direction of Earth center. The
second axis is assumed aligned with the orbit and the third axis completes
an orthonormal right oriented versors set. Consequently, its center is assumed
moving in a circular orbit, i.e. O exhibits a constant rotational motion with
respect to the third axis of ECI. We dene the corresponding angular ve-
locity as
o
= (0, 0,
0
)
T
. Moreover, with the same notations of Chapter 1,
B = X
, Y
, Z
is introduced, as depicted in Figure 6.2.

The dynamics equations, concerning eects of forces on the motion of the satel-
lite, are again represented by equation (1.2). Thus,
J = J + (6.1)
91
{ECI}
w
0
w
0
Z
X
Y
Z
X
Y
{B}
{O}
Figure 6.2. Reference frame {B} with respect to an orbiting reference frame {O},
both moving in {ECI}
where J = diag(J
x
, J
y
, J
z
). We describe the kinematics of the rigid body using
Eulers angles , and , as introduced in Chapter 1. In particular here, they
are used to represent the orientation of O, with respect to B. Notice that,
for this reason, relations between O and B in present chapter are identical
to relations between frames I and B dened in Chapter 1. Indeed, B
does not rotate in O unless external torques are applied. However, we do not
use directly representation (1.8)(1.10), since we also have to consider relations
between O and ECI. Thus, we rewrite the absolute angular velocity of
B as the sum of the velocity of B with respect to O plus the velocity of
O with respect to ECI, all the variables expressed in B coordinates:
b
=
bo
+
o
(6.2)
Further,
bo
depends on the sequence of rotations that the orbit frame has to
perform in order to reach the body frame:
bo
=
bo
+
o
+
o
o
(6.3)
These rotations are related to the Eulers angles. In particular O
is the
reference frame obtained from O after a rst rotation of angle along the
rst axis and O
is the one obtained from O
after a second rotation of

angle . Hence,
bo
=
_
_
s
+cc

cs

+s

_
(6.4)
where

=
o
o
,

=
o
and

=
bo
. For brevity we have replaced cx =

cos(x) and sx = sin(x).
Furthermore, also
o
needs to be expressed in body coordinates:
o
= R
_
_
0
0
o
_
_
=
_
_
(ss csc)
o
(sc +css)
o
cc
o
_
_
where
R =
_
_
cc ssc +cs ss csc
cs cc sss sc +css
s sc cc
_
_
(6.5)
is the rotation matrix that leads O to coincide with B and
o
is the angular
velocity of O with respect to ECI, as dened above. Notice that (6.5) diers
from (1.7), since here we have selected a 1-2-3 sequence of rotations, while in
Chapter 1 the sequence is 3-2-1. Thus,
b
=
_
_
s
+cc

+ (ss csc)
o
c
cs

+ (sc +css)
o
+s

+cc
o
_
_
(6.6)
As it will be seen in the following chapter, for the problem at hand, it is con-
venient the use of controllers designed for linear systems. Hence, for the time
being, we assume small angular displacements between B and O. Therefore,
the previous expression can be linearized, considering cos() 1, sin() ,
and analogous for and . It results
+
o
+
o
_
_
(6.7)
93
Consequently, using (6.7) and its rst derivative into (6.1) and disregarding
nonlinear terms, the Eulers equations become
J
x

= (J
y
J
z
)
2
o
+ (J
x
+J
y
J
z
)
o

+
x
J
y
= (J
z
J
y
J
x
)
o

(J
z
J
x
)
2
o
+
y
J
z
=
z
(6.8)
Model (6.8) represents the linear attitude model for a spacecraft orbiting around
Earth at velocity
0
. It is equivalent at using equation (6.1) together with the
rst derivative of (6.7). This representation can be used for describing the
rotational motion of the orbiting rigid body as long as it spans small angles. In
this case any sequence in the rotations of angles , and describing O in
B gives the same result [GPS80].
6.2 LEO Spacecraft Position Model
Model (6.8) describes the rotational motion of a 3D rigid body traveling on
a Low Earth Orbit at velocity
0
. In order to complete the mathematical
description of this kind of systems, it is necessary to introduce a position model.
Indeed, in practical applications, a LEO satellite is allowed to move with respect
to its orbit by means of a suitable combination of control inputs. Therefore,
the coordinates of B origin, have to be described in the moving reference
frame O. The importance of a position model, besides attitude model (6.8),
is simply understood, thinking to the inuence of small perturbations on the
spacecraft, such as drug forces. They require some control action for moving
the rigid body back to its orbit. Furthermore, a position model has to be used
while facing practical problems involving LEO satellites, as the fundamental
rendezvous problem between two spacecraft. In practice, all the real missions in
LEO and MEO orbits, require the design of coordinated attitude and position
maneuvers. Earth observing systems and space interferometers [ENS
+
03] are
two examples.
Since the satellite orbits in close proximity, i.e. the distance between the LEO
orbit and the ECI reference frame is much greater than the one between
the satellite and the point selected as leader, a standard approach to model its
position is that of linearizing the dynamics of the spacecraft around the orbit. In
the case that the reference orbit is approximately circular, the related linearized
equations of motion, commonly referred as the Hills equations, are as follows:
m x = 3
2
0
mx + 2
0
m y +f
x
m y = 2
0
m x +f
y
m z =
2
0
mz +f
z
(6.9)
where m is the mass of the satellite. In (6.9) we have dened the three Cartesian
coordinates as x, y, z and the forces that act on the spacecraft as f
x
, f
y
, f
z
.
They can be obtained starting from the two body problem [BMW71]. Briey,
once represented the position of two close objects in neighboring near-circular
orbits, (6.9) is derived, linearizing the equations of motion under gravitational
attraction, around one of the two orbits, selected as leader. Indeed, the dis-
tance between the two orbits is negligible with respect to the distance between
the leader and Earth center. A comprehensive analytical description of this
approach can be found in [Sid00] and [TIH02].
Note that in the present description, model (6.8) and model (6.9) are decoupled.
However, components x ,= 0 would correspond to dierent angular velocities of
the satellite and the leader point, i.e.
0
would not be constant. Therefore,
the attitude of the vehicle would change with x. Anyway, in LEO systems, the
eects of relative position on the attitude are much lower than the ones due to
disturbances and they can be neglected.
Furthermore, note that usually, inputs are given in the body reference frame
B. For a correct use of (6.9), here they should be expressed in the orbit
frame O by means of an appropriate transformation. However, since we as-
sume small angle deviations between O and B there is no dierence in
representing these variables either in body coordinates or in orbit coordinates.
6.3 LEO Spacecraft Control Problem Formulation
In the following, we introduce the problem of controlling the attitude and the
position of a LEO spacecraft, both dened with respect to an orbiting point, or
95
to another satellite in a leader following framework.
Since equations (6.8) and (6.9) are linear, any control law designed for continu-
ous linear systems can be applied for stabilizing a desired state or pre-designed
path [Kha01]. However, in real missions, the simple stabilization problem is
not important, since it does not t with practical applications. Indeed, classical
attitude and position controllers may fail because of the presence of stringent
saturation constraints, always present with small space vehicles, such those in-
volved in the present kind of systems. Thus, we focus our attention on the
control problem in the presence of constraints. The same approach is adopted
in the most part of the literature on LEO and MEO space systems. See, for
instance, [VGRM01], [TIH02] and related references.
Consider again equations (6.8) and (6.9). For our purposes, they can be sum-
marized in the state-space time-continuous representation
s(t) = A
c
s(t) +B
c
(u(t) +(t)) (6.10)
where s is the whole state of the system, selected as
s = [ x, x, y, y, z, z,

, ,

, ,

, ]
T
R
12
As a consequence, u is composed by the six input components and f dened
in (6.8) and (6.9). Thus, u R
6
. Moreover, we have added input disturbances
forces and torques, always to be considered in practical LEO space systems.
Thus R
6
. Denition of matrices A
c
and B
c
in equation (6.10) is straight-
forward from (6.8) and (6.9).
Now, the general state-space model given in equation (6.10) can be discretized
using the well-known technique of approximately integrating the dynamics in
equations (6.8) and (6.9) over one sample period, T [FPW98]. Here, A
k
, B
k
correspond to A
k
= A(kT), B
k
= B(kT). In the case of system (6.10), A
k
= A
which is a constant and B
k
= B which is a constant. Therefore, the ZOH
sampled dynamical model of the satellite, takes the form
s(k + 1) = As(k) +B(u(k) +d(k)) (6.11)
where d(k) represents relative disturbance forces and torques accumulated dur-
ing a sampling period. We assume d(k) R
6
satisfying d(k) T, k Z
+
,
with T a specied convex and compact set such that 0
6
T.
We now formulate the problem of controlling a spacecraft in Low Earth Orbit.
Controlling a LEO spacecraft means driving the orbiting rigid body to a desired
state, i.e. solving a reconguration problem. Desired state is represented by a
reference position relative to the leader orbit and by a reference attitude relative
to the orbiting frame O. Moreover, the satellite can be required to cover a
pre-specied path, i.e. to solve a tracking problem. Finally, a LEO spacecraft
can be simply required to keep its initial state, in spite of disturbances acting on
it: keeping problem. These problems are especially important when the satellite
is a member of a formation ying system. In this case its maneuvers have to be
performed in coordination with all the other vehicles of the eet [ENS
+
03].
While facing a reconguration, a tracking or a keeping problem for a LEO satel-
lite, as dened above, it has to be taken into account that maneuvers are made
thanks to a combination of small jet actuators that are not able of supplying
arbitrary large forces and torques. Thus, input saturation constraints of the
form
[u
i
(k)[ u
i
max
, i = 1, , 6 (6.12)
are present. Furthermore, we want to handle state accuracy constraints, i.e.
state-related constraints
[y
i
(k) r
i
(k)[ < , i = 1, , 6 (6.13)
k Z
+
, with r(k) a reference signal and y(k) = Cs(k) some suitably selected
output.
Notice that the time-step T, is typically chosen to be small (i.e. on the order of
4 5 sec so that there are about 1.000 points per typical LEO orbit) to obtain
an accurate discretization. However, this results in a large number of inputs
and constraints for reconguration or keeping problems, which can take several
minutes to be solved. One method to reduce these computational times for
a real-time implementation is to only allow inputs every n-(integer) time-steps
and/or only enforcing the constraints every n-(integer) time-steps. For example,
making it every 10 steps, greatly reduces the size of the control problem, re-
sulting in solution times on the order of seconds. This decreased solution time
is obtained at the expense of system performance because the control inputs
are allowed less often (reduces eciency) and the constraints are only tested
97
at certain time-steps (violations are possible at intermediate time-steps). This
approach has been followed in [TIH02]. On the contrary, here we propose a
control technique able to guarantee low numerical burdens.
Model (6.11) under constraints (6.12) and (6.13) suggests the use of a model
based predictive control strategy [Mos95]. In particular, in the following chapter,
we propose a control strategy based on the Command Governor (CG) approach
[BCM97], [CMA00]. It consists in designing a primal control law that does not
take the constraints into account, and an external unit capable of taking care
of constraint fulllment by modifying, whenever necessary, the reference. Since
the primal controller is designed for the linear model (6.11), any simple con-
trol strategy can be selected. In this dissertation we solve the unconstrained
control problem using a simple Linear Quadratic (LQ) Regulator [Mos95]. The
LQ controller is used to stabilize the system in linear regimes, viz. when the
constraints are not active. To this end, by considering the following quadratic
cost function
J(u, s) =
+
j=0
(s
T
(j)Qs(j) +u
T
(j)Ru(j)) (6.14)
with the weights Q and R appropriately selected, e.g. for achieving, approxi-
mately, minimum energy control, the control law which minimizes (6.14) is given
by
u(k) = Fs(k) +Hr(k) (6.15)
where F = (R + B
T
P
B)
1
BP
A. Here r(k) is the reference, P
is the
solution of the Riccati Dierence Equation (RDE) and H is such that C(I
(ABF))
1
BH = I.
Hence, the evolution of the closed-loop system can be described by
s(k + 1) = s(k) +Gr(k) +G
d
d(k)
y(k) = H
y
s(k)
(6.16)
Clearly, the closed-loop system evolutions under LQ state feedback (6.16) not
necessarily satisfy the constraints (6.12) and (6.13) introduced above. Thus, we
need to equip the overall control strategy with a unit capable of taking care of
constraint fulllment.
6.4 Conclusions
In this chapter we have introduced the problem of controlling the attitude
of MEO and LEO spacecraft. We have dened a LEO satellite as a non-
geosynchronous 3D rigid body orbiting around Earth at a constant velocity,
depending on its orbits altitude. First, we have illustrated the mathematical
model of the system, focusing our attention on the main dierences with respect
to the geo-stationary case. In particular we have adopted the leader-following
approach, leading to a novel linear attitude model, besides the well known Hills
relative position model. Then, we have introduced input saturation and state
accuracy constraints, that have to be taken into account while facing MEO and
LEO spacecraft control problems. Finally, we have proposed the use of an MPC
control technique. It will be illustrated in more detail in the following chapter.
99
Chapter
7
Attitude Control of a Low
Earth Orbit Satellite
In Chapter 6 we have introduced the problem of reconguring and keeping the
state of a LEO spacecraft, as well as the problem of making it to track a pre-
specied trajectory. The mathematical model of LEO satellites is relatively
simple and it can be linearized, under common assumptions. Hence, it appears
that any control law designed for continuous linear systems, is suitable for this
problem. However, as explained in detail in Chapter 6, LEO satellites are
typically small and their thrusters are not able to provide large control eorts
[VGRM01], [TIH02]. Thus, in order to face practical applications, it is necessary
to design control strategies that take constraints into account. Therefore, for
the sake of simplicity and low numerical burden, we have proposed the use of
an MPC based technique: the Command Governor approach.
The main properties of such a control technique are illustrated in this chapter.
Then, a CG scheme is applied to a satellite orbiting at 600 m above Earths
surface and simulation results are depicted for illustrating the eectiveness of
the technique.
7.1 Command Governor Approach for LEO Spacecraft
Control
In Figure 7.1 a CG control scheme with plant, primal controller and Command
Governor unit [BCM97] is depicted. Equation (7.1) is a closed-loop state-space
[]
CG
r(k) g(k) u(k)
d(k)
y(k)
c(k)
Primal
Controller
Plant
c (k)
p
c (k)
c
c (k)
p
c (k)
c
[
s(k)=
]
r(k) ~
~
2D
2C
Figure 7.1. Command Governor scheme.
description of the plant regulated by the primal controller.
s(k + 1) = s(k) +Gg(k) +G
d
d(k)
y(k) = H
y
s(k)
c(k) = H
c
s(k) +Lg(k) +L
d
d(k)
(7.1)
In particular, s(k) R
n
is the state which includes plant and compensator states
(if any), g(k) R
m
, which would be typically g(k) = r(k) if no constraints were
present (no CG present), is the CG output, viz. a suitably modied version of
the reference signal r(k) R
m
. Note that r(k) can be any desired function,
even a constant, to be tracked by system output. Moreover, d(k) R
n
d
is an
exogenous disturbance satisfying d(k) T, k Z
+
, with T a specied convex
and compact set such that 0
n
d
T; y(k) R
m
is the output, viz. a performance
related signal which is required to track r(k) and c(k) R
n
c
is the vector to be
101
constrained. It is assumed that
1. (7.1) is asymptotically stable;
2. (7.1) is oset free, viz. H
y
(I )
1
G = I
m
(7.2)
The CG design problem consists of nding, at each time instant k, a command
g(k) as a function of the current state s(k) and reference r(k)
g(k) := g(s(k), r(k)) (7.3)
in such a way that g(k) is the best approximation of r(k) at time k, under
the constraint c(k) (, k, and all possible disturbance sequences d(k) T.
Moreover, it is required that:
1. g(k) r whenever r(k) r, with r the best feasible approximation of r;
2. the CG have a nite settling time, viz. g(k) = r for a possibly large but
nite k whenever the reference stays constant after a nite time.
By linearity, one is allowed to separate the eects of initial conditions and inputs
from those of disturbances, i.e. s(k) = s(k) + s(k), where s is the disturbance-
free component (depending only on initial state and input) and s depending
only on disturbances. Then, denote the disturbance-free steady-state solutions
of (7.1), for a constant command g(k) w, as follows
s
w
:= (I
n
)
1
Gw
y
w
:= H
y
(I
n
)
1
Gw
c
w
:= H
c
(I
n
)
1
Gw +Lw
(7.4)
Consider next the following set recursion
(
0
:= ( L
d
T
(
j
:= (
j1
H
c
j1
G
d
T
(
:=
j=0
(
j
(7.5)
where / c is dened as a / : a + e /, e c. It can be shown that
the sets (
j
are non-conservative restrictions of ( such that c(j) (
, j Z
+
,
implies c(j) (, j Z
+
. Thus, one can consider only disturbance-free
evolutions of the system and adopt a worst case approach. Next consider, for a
small enough > 0, the sets:
(
:= (
:= w R
m
: c
w
(
(7.6)
where B
is the ball of radius centered at the origin. In particular, J
, which
we assume non-empty, is the set of all commands whose corresponding steady-
state solution satises the constraints with margin .
The main idea is to choose at each time step a constant virtual command
v() w, with w J
, such that the corresponding virtual evolution fullls

the constraints over a semi-innite horizon and its distance from the constant
reference of value r(k) is minimal. Such a command is applied, a new state is
measured and the procedure is repeated. In this respect we dene the set 1(s)
as
1(s) = w J
: c(j, s, w) (
j
, j Z
+
(7.7)
where
c(j, s, w) := H
c
_
j
s(j) +
j1
i=0
ji1
Gw
_
+Lw (7.8)
has to be understood as the disturbance-free virtual evolution at time j of c
from the initial condition s at time j = 0 under the constant command v()
w. As a consequence, 1(s) J
. Moreover, if non-empty, it represents the

set of all constant virtual sequences in J
whose evolutions starting from s

satises the constraints also during transients. Thus, taking as a selection index
a quadratic cost, the CG output is chosen according to the solution of the
following constrained optimization problem
g(k) = arg min
wV(s(k))
|w r(k)|
2
(7.9)
where =
> 0
p
and |w|
2
:= s
s. It has been shown in [BCM97] and

[CMA00] that the CG technique described above, has the following properties.
Theorem 7.1 Let assumptions (7.2) be fullled. Consider system (7.1) along
with the CG selection rule (7.9), and let 1(s(0)) be non-empty. Then:
1. the minimizer in (7.9) uniquely exists at each k Z
+
and it can be ob-
tained by solving a convex constrained optimization problem, viz. 1(s(0))
103
non-empty implies 1(s(k)) non-empty along the trajectories generated by
the CG command (7.9);
2. the set 1(s), s R
n
, is nitely determined, viz. there exists an integer j
0
such that if c(j, s, w) (
j
, j 0, 1, . . . j
0
, then c(j, s, w) (
j
j Z
+
.
Such a constraint horizon j
0
can be determined o-line;
3. the constraints are fullled k Z
+
;
4. the overall system is asymptotically stable; in particular, whenever r(k)
r, g(k) converges in nite time either to r or to its best steady-state ad-
missible approximation,
g(k) r := arg min
wW
|w r|
2
(7.10)
Consequently, by the oset free condition (7.2.2), lim
k+
y(k) = r, where
y is the disturbance-free component of y.
[BCM97] also provides a comprehensive description on the algorithm for calcu-
lating the constraint horizon j
0
dened above. Other details on solvability and
computability of CG schemes, can be found in [GK92], [GKT95] and [GK99].
It is worth pointing out that, in order to accommodate coordination constraints
in the form of (6.13) one has to generalize the denition of c(k) in (7.1) as follows
c(k) = H
c
s(k) +L
g
g(k) +L
d
d(k) +L
r
r(k) (7.11)
Unfortunately, (7.11) diers from the third equation in (7.1) because of the
presence of L
r
r(k). Apart from cases where r(k) is constant, the condition
1(s(0)) non-empty, does not guarantee any longer that a solution exists and is
admissible at each future time instant as indicated in Theorem 7.1.
However, all the theorems stated properties can be restored acting as described
below. Consider a parametric description r((k)) of the reference trajectory,
with the introduction of a scalar parameter (k), where
(k) = (k 1) + (k) (k) [0, 1] (7.12)
r((k)) can be thought as any point R
m
between r(k 1) and r(k) along
the reference trajectory. In particular, one would be allowed to choose g(k) =
r((k)) corresponding to the nominal point r(k) in the reference trajectory,
taking (k) = 1, if no constraints were present. Therefore, we can substitute
(7.12) into both g(k) and r(k) in (7.11) obtaining
c(k) = H
c
s(k) +L
g
r((k)) +L
d
d(k) +L
r
r((k))
= H
c
s(k) +Lr((k)) +L
d
d(k)
that produces again an equation in the form of (7.1), with r((k)) in place of
g(k). It allows to successfully deal with the feasibility problems arising from
the presence of state accuracy constraints (6.13). Consequently, we can replace
(7.1) with
s(k + 1) = s(k) +Gr((k)) +G
d
d(k)
y(k) = H
y
s(k)
c(k) = H
c
s(k) +Lr((k)) +L
d
d(k)
(k) = (k 1) + (k)
(7.13)
where the parameter (k) represents the novelty of our approach with respect
to the classical CG strategy proposed in [BCM97].
The new CG problem is then
(k) := arg max
r((k1)+)V(s(k))
w(k) := r((k 1) + (k))

(7.14)
which can simply be solved via bisection. Thus, all properties pertaining to
the CG approach described above are restored and the state can be driven to a
desired nal value, along an a priori established trajectory, without constraints
violation. The same conclusion cannot be stated for other techniques proposed
in literature as in [TIH02] and [VGRM01].
The modied-CG scheme illustrated above, appears particularly suitable for
the problem at hand. Indeed, LEO spacecraft are often required to change
their orientation and position, with respect to their nominal orbit or with re-
spect to other satellites ying in formation. If they perform maneuvers moving
along pre-designed trajectories, constraints in the form of (6.13), actually en-
compass many other constraints. In particular they include collision-avoidance
constraints, that represent one of the main issues in formation ying control
[ENS
+
03], [SBW06].
105
Finally, notice that models (6.8) and (6.9) appear independent each other.
Therefore, it seems that input commands and f for attitude and position can
be provided separately. However, in LEO spacecraft, both translational and
rotational motion are obtained by suitable combinations of the same thrusts.
Moreover, attitude maneuvers and paths are usually dened in combination
with position maneuvers and paths. Thus, representing (6.8) and (6.9) as a
single model, as made above by (6.11), is necessary for the problem at hand.
Consequently, at any time instant k, the CG control law needs to select the most
restrictive (k) between the one resulting from the attitude maneuver and the
one resulting from the position maneuver.
7.2 Simulation of a LEO Spacecraft Reconguration
Maneuver
In the following, we present simulation results on the application of the modied-
CG technique described in previous section, to a LEO spacecraft. As illustrated
above, we assume small angle displacements between B and O, justifying
linear attitude description (6.8). Moreover, we assume small distances between
the orbiting 3D rigid body and its reference orbit. Hence, Hills model (6.9)
holds true.
From the arguments described in Chapter 6 and in previous section, we rewrite
(6.8)(6.9) in the more compact form (6.11) and we apply to it an LQ control
law, that does not take constraints into account. Hence, system object of study
becomes (6.16). Since we also want to handle saturation constraints in the form
of (6.12) and state accuracy constraints in the form of (6.13), the whole system
is nally described by (7.13). Thus, the control problem is eectively solved by
technique (7.14).
We refer to a LEO satellite that orbits around Earth at an altitude of about
600 Km, i.e. at a velocity
0
= 0.0011rad/sec. We assume, for the rigid body
a mass of 150 Kg and a diagonal inertia matrix J = diag(35, 16, 25) Kgm
2
.
These, are the typical dimensions of small satellites [ENS
+
03]. Further we sup-
pose the following saturation constraints: [f[
max
= 5 10
2
N and [[
max
=
10
3
Nm. The maximum amplitudes of disturbance forces are assumed to be
[d
i
[
max
= 10
3
N, i = 1, 2, 3. Moreover, the maximum amplitudes of distur-
bance torques are assumed to be [d
i
[
max
= 10
4
Nm i = 4, 5, 6. A value of
= 10
6
in equation (7.6) is selected.
We rst apply the technique described above to perform a reconguration ma-
neuver, i.e. to change the position and the attitude of the satellite. Notice that
this task can be interpreted as a tracking problem, where the reference signal to
be tracked is dened only by a nal state r and where constraints in the form
of (6.13) are absent.
A LEO spacecraft, leaving from position (x, y, z)
T
= (60, 5, 5)
T
m with re-
spect to the origin of O, selected as the leader orbiting point, has to be
driven to the nal desired position (x, y, z)
T
= (70, 0, 0)
T
m. In the meanwhile
it is asked to rotate, from its initial attitude corresponding to Eulers angles
(, , )
T
= (0.1, 0.1, 0.1)
T
rad in O, to a nal desired attitude correspond-
ing to Eulers angles (, , )
T
= (0.1, 0.1, 0)
T
rad. Position and attitude
evolutions under controller (7.14), are depicted in Figures 7.2 and 7.3. A con-
straint horizon of j
o
= 23 steps is determined by means of the appropriate
algorithm presented in [BCM97]. Moreover, Q and R in (6.14) are selected as
Q = I
12
and R = 50I
6
. Figures 7.4 and 7.5 show input forces and torques
related to the desired maneuver. We observe that there are instants in which
saturation constraints become active. Thanks to the action of the CG unit, such
constraints are never violated. If the CG unit is not used, the LQ control law
(6.14)-(6.15) is not able of driving the spacecraft towards the desired position
due to input saturation. It results from Figure 7.6, where components x and
y do not reach the desired values. The rationale is that input forces generated
by the LQ controller are no longer admissible. It can be seen from Figures
7.8 and 7.9 that show how forces and thrusts actually applied, are a saturated
version of those computed by the LQ control law. In this simulation, a dierent
behavior results for angles , and which reach the desired attitude in spite
of constraints violations during the transient. It is depicted in Figure 7.7.
107
0 1000 2000
60
62
64
66
68
70
72
Time [s]
x

[
m
]
0 1000 2000
2
1
0
1
2
3
4
5
Time [s]
y

[
m
]
0 1000 2000
5
4
3
2
1
0
1
Time [s]
z

[
m
]
Figure 7.2. Relative position components x(k), y(k), z(k) of the LEO satellite,
under the proposed LQ+CG control law (7.14).
0 1000 2000
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
Time [s]

[
r
a
d
]
0 1000 2000
0.1
0.05
0
0.05
0.1
0.15
Time [s]

[
r
a
d
]
0 1000 2000
0.02
0
0.02
0.04
0.06
0.08
0.1
Time [s]

[
r
a
d
]
Figure 7.3. Eulers angles (k), (k), (k) of the LEO satellite, under the proposed
LQ+CG control law (7.14).
7.3 Simulation of a LEO Spacecraft Tracking Maneu-
ver
Next, we present simulation results of the proposed control technique, when a
LEO satellite is required to reach a desired position and attitude, moving along
0 1000 2000
0.06
0.04
0.02
0
0.02
0.04
0.06
Time [s]
f
1

[
N
]
0 1000 2000
0.06
0.04
0.02
0
0.02
0.04
0.06
Time [s]
f
2

[
N
]
0 1000 2000
0.06
0.04
0.02
0
0.02
0.04
0.06
Time [s]
f
3

[
N
]
Figure 7.4. Input forces f
i
(t) generated by the proposed LQ+CG control law (7.14).
The dash lines represent the constraint boundaries.
0 1000 2000
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
x 10
3
Time [s]
1

[
N
]
0 1000 2000
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
x 10
3
Time [s]
2

[
N
]
0 1000 2000
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
x 10
3
Time [s]
3

[
N
]
Figure 7.5. Input torques
i
(7.14). The dash lines represent the constraint boundaries.
a pre-designed trajectory. In this case, constraints in the form of (6.13) are
present. Then, whole system is described by (7.13) and control action (7.14)
109
0 1000 2000
50
100
150
200
250
300
Time [s]
x

[
m
]
0 1000 2000
25
20
15
10
5
0
5
Time [s]
y

[
m
]
0 1000 2000
5
4
3
2
1
0
1
2
Time [s]
z

[
m
]
Figure 7.6. Relative positions x(k), y(k), z(k) of the LEO satellite under LQ
control law only.
0 1000 2000
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
Time [s]

[
r
a
d
]
0 1000 2000
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
Time [s]

[
r
a
d
]
0 1000 2000
0.04
0.02
0
0.02
0.04
0.06
0.08
0.1
Time [s]

[
r
a
d
]
Figure 7.7. Eulers angles (k), (k), (k) of the LEO satellite under LQ control
law only.
has to be applied in place of (7.9).
We suppose that the spacecraft, leaving from the state reached after previous
maneuver, is required to track a pre-specied circumference of radius 20 m in
0 1000 2000
30
25
20
15
10
5
0
5
Time [s]
f
1

[
N
]
0 1000 2000
1
0.5
0
0.5
1
1.5
2
Time [s]
f
2

[
N
]
0 1000 2000
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Time [s]
f
3

[
N
]
Figure 7.8. Input forces f
i
represent the constraint boundaries.
0 1000 2000
0.02
0.015
0.01
0.005
0
0.005
0.01
Time [s]
1

[
N
]
0 1000 2000
0.01
0.005
0
0.005
0.01
0.015
Time [s]
2

[
N
]
0 1000 2000
10
8
6
4
2
0
2
4
x 10
3
Time [s]
3

[
N
]
Figure 7.9. Input torques
i
represent the constraint boundaries.
the (x, y) orbit plan and centered in (x, y)
T
= (50, 0)
T
m, i.e. to track a variable
reference signal. Further, we require z, , and to maintain their initial
111
values, i.e. z = 0 m, and (, , )
T
= (0.1, 0.1, 0)
T
rad (keeping).
Here, besides the same saturation constraints of the rst example, we want to
consider also accuracy constraints [s
i
(k) r
i
((k))[ < 0.1 m for each of the
position components, i.e. for i = 1, 2, 3, and [s
i
(k) r
i
((k))[ < 0.01 rad for
each of the attitude components, i.e. for i = 4, 5, 6. This kind of accuracy is
required, as an example, in order to maintain the pointing of a geographic site,
despite of O motion around ECI.
Results of the application of technique (7.13)-(7.14) are illustrated below. Figure
7.10 shows evolution of position variables x, y and z from initial to nal state.
Note that the reference signal, depicted in dash lines, is tracked by x and y,
0 2 4 6
x 10
4
25
30
35
40
45
50
55
60
65
70
75
Time [s]
x

[
m
]
0 2 4 6
x 10
4
25
20
15
10
5
0
5
10
15
20
25
Time [s]
y

[
m
]
0 2 4 6
x 10
4
5
4
3
2
1
0
1
2
3
4
5
Time [s]
z

[
m
]
Figure 7.10. Evolution of the relative position variables x, y and z of the LEO
satellite with the application of LQ+CG (tracking problem) for a period of revolu-
tion.
while z(k) = z(0) is kept at any time instant k. Initial values of Eulers angles,
here not depicted, are kept as well. In the case of present simulation, the
active constraints are the coordination ones related to the position in the x
and y directions. This is evident from Figure 7.11, where e
1
and e
2
approach
constraints boundaries. Input variables, on the contrary, are well inside their
respective saturation boundaries. Notice that, because of the use of the CG
selection logic (7.14), the reference trajectory is tracked with variable speed,
0 2 4 6
x 10
4
0.1
0.05
0
0.05
0.1
0.15
e
1

[
m
]
Time [s]
0 2 4 6
x 10
4
0.1
0.05
0
0.05
0.1
0.15
e
2

[
m
]
Time [s]
0 2 4 6
x 10
4
0.1
0.05
0
0.05
0.1
0.15
e
3

[
m
]
Time [s]
Figure 7.11. Errors e
i
(t), i=1,2,3 in coordination accuracy for the position compo-
nents x(k), y(k) and z(k) related to the circumference tracking maneuver under
(7.14). The dash lines represent the constraint boundaries (|e
i
|
max
= 0.1).
lower than the planned velocity. This can be deduced from Figure 7.12 where
(k) is not constant and (k) is not a line. Thus, evolution in the (x, y) plane is
the one depicted on the left side of Figure 7.13, while reference evolution would
be the one represented on the right side, in dash line. It is evident that the
nominal trajectory r(k) is tracked slowly, with respect to planned velocity, but
desired path is covered, without accuracy constraints violation. As is previous
section, also in this case, if the CG unit is not used, constraints are not fullled.
In particular here, closed-loop system (6.16), without (7.14), does not satisfy
formation accuracy constraints related to the position components. On the other
hand, saturation constraints are always satised under this maneuver. Thus,
the trouble in this case is that coordination accuracy deteriorates up to a level
given in Figure 7.14. Eulers angles , and , input torques u
i
(i=4,5,6) and
attitude accuracy related to the present tracking maneuver, here not depicted,
stay inside their respective bounds.
113
0 2 4 6
x 10
4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time [s]
(
t
)
0 2 4 6
x 10
4
0
500
1000
1500
2000
2500
Time [s]
(
t
)
Figure 7.12. Values of (t) and (t) for the CG selection logic (7.14) corresponding
to the maneuver.
20 40 60 80
0
5
10
15
20
25
x

[m]
y

[
m
]
20 40 60 80
0
5
10
15
20
25
r
1
[m]
r
2

[
m
]
Figure 7.13. Trajectory in the (x, y) plane under (7.14) (left) in comparison with
reference signal (right).
7.4 Conclusions
In this chapter we have proposed a technique for controlling the attitude of
LEO satellites. The proposed control scheme is based on the MPC technique,
0 1 2
x 10
4
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
e
1

[
m
]
Time [s]
0 1 2
x 10
4
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
e
2

[
m
]
Time [s]
0 1 2
x 10
4
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
e
3

[
m
]
Time [s]
Figure 7.14. Errors e
i
(t) in coordination accuracy for the position components x,
y and z related to the circumference tracking maneuver under LQ only. The dash
lines represent the constraint boundaries (|e
i
|
max
= 0.1).
referred to as the Command Governor. It consists in designing a primal control
law and an external unit capable of taking care of constraint fulllment by mod-
ifying, whenever necessary, the reference signal. Besides classical CG scheme,
well known in literature and widely discussed in [BCM97], we have proposed
a parameterization of reference signal, in order to handle state accuracy con-
straints.
Since in LEO orbits, controlling attitude usually requires simultaneously con-
trolling relative position, we have applied the control scheme to a mathematical
description of the system, combining both position and attitude variables. Then,
we have shown simulation results of the application of the proposed technique,
both for reconguring the state of a LEO satellite and for making it to cover a
pre-designed trajectory.
The proposed LQ+CG algorithm needs the system to be linear. Hence, we
have assumed small angular displacements between initial body frame orienta-
tion ((0), (0), (0)) and nal desired orientation, leading to model (6.8). This
assumption will be removed in next chapter.
115
Chapter
8
Attitude Control of a LEO
Satellite for Large Angle
Maneuvers
In previous chapter, we have proposed a model predictive control technique,
namely a modied version of the Command Governor approach [BCM97], for
controlling the attitude of a LEO spacecraft. The presented scheme, has been
shown to be eective under the assumptions of small distances and small angu-
lar displacements between the body frame B and a leader orbiting reference
frame O. This allows at describing relative position by linear model (6.9) and
attitude by linear model (6.8).
However, the second assumption, requiring small angular displacements between
a LEO satellite body frame B and its leader reference frame O, might
not match with some demanding applications. Experimental results show that
model (6.8) has signicant discrepancy with respect to the real model for Eulers
angles deviations larger than 15 deg. Thus, in this chapter we remove the small
angles assumption, taking into account maneuvers requiring the space vehicle to
span large angles. For instance, this is an usual request, when a LEO satellite
has to observe a large area on the Earth surface, during an Observation Mission
[ENS
+
03].
On the other hand, the sake of simplicity and the need of low numerical burdens
still suggest the use of the CG techniques for linear systems, adopted in Chap-
ter 7. Thus, we propose a bank of controllers, each one designed with respect
to a pre-established orbital reference frame. A hybrid control scheme is then
applied. The bank of controllers is orchestrated by a supervisory logic which
at each time instant selects the most appropriate CG unit, on the basis of the
spacecraft state and the prescribed tracking paths.
8.1 LEO Spacecraft Nonlinear Attitude Model
Consider model (6.8)-(6.9) illustrated in Chapter 6, describing attitude and po-
sition variables dynamics of a LEO spacecraft, with respect to a leader reference
frame O. As described above, the assumption of close proximity between the
LEO satellite and its nominal orbit, under which model (6.9) can be used, is
eective for all the real missions, even when the space vehicle is a component
of a formation ying system. On the contrary, the hypothesis of small angles,
that leads to system (6.8), does not hold true in some practical applications.
On the other hand, the use of above model is convenient, since it guarantees the
possibility of using the CG-based control law (7.14), which provides simplicity
and law numerical burdens.
In order to maintain representation (6.8)-(6.9) in the general case, one possibil-
ity is that of parting any possible large angle attitude maneuver, in a sequence of
small angle maneuvers, dened with respect to a suitable set of reference frames
O
i
. In this way, any sub-maneuver describing B in O
i
, can satisfy the
small angular displacement assumption. Therefore, equation (6.8) can be used,
as long as Eulers angles , and , between O
i
and B, are small
1
. Then,
model (6.8) can be re-written with respect to another reference frame O
j
sat-
isfying the small angular displacement assumption. Finally, at any time instant
k, B attitude can be easily described in the leader frame, in the following
denoted as C, combining B orientation in O
i
(k) and O
i
(k) orientation
1
Less than 15 deg, from experimental results.
117
in C. Indeed, the constant attitude of any O
i
frame, can be expressed with
respect to C, by a constant rotation matrix.
Clearly, it is necessary that at any time instant, there exist at least one frame
O
i
, with small angular displacement with respect to B. An example of the
described approach is depicted in Figure 8.1.
w
0
Z
X
Y
Z
X
Y
{B}
{C}
{O }
1
{O }
2
{O }
3
{O }
4
{O }
5
{O }
6
...
Figure 8.1. Set of reference frames {O
i
} and leader orbiting reference frame {C}.
At any time instant, there exists at least one {O
i
} satisfying the small angular
displacement assumption with respect to {B}.
Thus, we consider again equation (6.4), expressing
bo
in body coordinates. As
introduced above, we denote the leader reference frame as C and we dene a
set of reference frames O
i
, (i = 1, ,
i), orbiting in ECI along with C,

and covering the space SO(3). It means that we take a number

i of reference
frames O
i
, in such a way that
SO(3)
_

i
_
i=1
_
R
o
i
c
R
o
i
c
O
i
_
_
(8.1)
where R
o
i
c
is the constant rotation matrix that changes C in O
i
and
R
o
i
c
= R
o
i
c
(, , ) spans angles up to 15 deg in , and directions.
Moreover, any O
i
is motionless in C. Hence,
o
i
=
c
i = 1, ,
i
Next, considering the rotation matrix R
bc
between the body frame B and the
leader frame C, we have
o
i
= R
bc
_
_
0
0
0
_
_
= RR
o
i
c
_
_
0
0
0
_
_
where R is the rotation matrix between O
i
and B, dened by (6.5), and
R
o
i
c
is the constant rotation matrix that changes C in O
i
.
Now, since both the velocity
0
of C in ECI and the attitude of any O
i
in C are constant and known, we can dene the angular velocity of C in

O
i
coordinates as
c
= R
o
i
c
_
_
0
0
0
_
_
=
_
c1
c2
c3
_
_
(8.2)
It follows that
o
i
corresponds to
o
i
=
_
_
cc
c1
+ (cs +ssc)
c2
+ (ss csc)
c3
cs
c1
+ (cc sss)
c1
+ (sc +css)
c3
s
c1
sc
c2
+cc
c3
_
_
(8.3)
Finally, adding (6.4) and (8.3), taking the rst derivative of
b
in (6.2), and
substituting into (6.1), yields to a nonlinear attitude model. However, we as-
sume small angular deviations between B and the selected O
i
(k), around
which B orientation is expressed at time k. Thus, cos() 1, sin() 0 and
analogous for and . Therefore, Eulers equations (6.1) become
= (
2
c3
2
c2
)

J
1
+
c3
(1 +

J
1
)
+
c1
c2

J
1
+
c2
(

J
1
1)

c1
c3

J
1
+
1
=
c3
(

J
2
1)

c1
c2

J
2
+ (
2
c1
2
c3
)

J
2
+
c1
(1 +

J
2
)

+
c2
c3

J
2
+
2
=
c2
(1 +

J
3
)

+
c1
c3

J
3
+
c1
(

J
3
1)
c2
c3

J
3
+ (
2
c2
2
c1
)

J
3
+
3
(8.4)
where

J
i
=
J
j
J
k
J
i
(i, j, k = 1, 2, 3, i ,= j ,= k).
System (8.4) can be placed in (6.10) in place of (6.8). Then, all the properties
119
of the scheme presented in Chapter 7 for controlling B orientation in O,
are restored. Clearly, at time k they hold true in a suitable O
i
, satisfying
the small angular displacement assumption. As soon as , or become large,
a technique for switching to another CG controller, has to be adopted. This
new controller needs to be dened in a suitable O
j
, j (1, ,
i), by the
corresponding (8.4)-(6.9) model.
A method for switching between the CG controller designed in O
i
, to the one
designed in O
j
is proposed in the next section. It addresses present control
problem in a hybrid framework.
8.2 Hybrid Command Governor for LEO spacecraft
In this section we propose a control scheme still based on the CG technique
(7.14) for linear systems, even when the LEO satellite has to perform large
angle maneuvers. As described above, this can be made introducing a bank of
linearized models in the form of (8.4), each one representing the attitude relative
to a specic reference frame O
i
. The position components are not aected
by this problem and just one single model (6.9) can be used along with each
member of the given set of attitude models.
Consequently, referring to the orbiting rigid body, a single CG unit can be
designed for each linearized model and a suitably designed supervisory unit
can take care of orchestrating the switching among the CG candidates during
the on-line operations. We refer to the overall technique as hybrid CG control
scheme (HCG). A similar approach has been previously used in [ABC
+
03].
Consider the following set of reference set-points r which are desired to be
tracked without oset
r R
m
Assume, for the problem at hand, that , J
, with J
dened in (7.6). Thus,

the requirement that all the set-points in will be tracked without error cannot
be satised. A way to overcome this limitation is that of covering the set with
a collection of J
i
(i = 1, ,
i), with overlapping interior corresponding to

i
dierent CGs such that

i
_
i=1
J
i
(8.5)
and InteriorJ
i
J
j
,= 0, for at least a pair (i, j) 1, ,
i. Clearly, CG
i
operates properly when initial and nal set-points belong to J
i
. If the nal
set-point belongs to a dierent set J
j
, a procedure for switching between CG
i
and CG
j
has to be dened. To this end, let us consider the output admissible
set Z
i
R
m
R
n
for CG
i
. It consists of the set of all pairs [r, s]
T
whose
evolutions satisfy the constraints for all t Z
+
.
Hence, we can dene the set of all states which can be steered to feasible equi-
librium points without constraints violation
A
j
:= x R
n
: [w, s]
T
Z
i
for at least one w R
m
Now, if (i, j) is such that InteriorJ
i
J
j
,= 0 then also InteriorA
i

A
j
,= 0. Thus, one can a-priori dene a convenient transition reference r
ij

InteriorJ
i
J
j
such that
s
ij
InteriorA
i
A
where s
ij
is the equilibrium disturbance-free steady-state corresponding to r
ij
(using the worst case approach stated in Chapter 7).
Finally, [r
ij
, s
ij
]
T
Z
i
Z
j
and the transfer strategy is simply dened.
Assume to be at instant k, be using CG
i
and let r(k) J
i
, r(k +1) J
j
with
J
i
J
j
,= 0. Hence, a possible switching logic is as follows:
1. Solve and apply
g(k +h) = arg min
wV
i
(s(k+h))
|w r(k)|
2
, h = 1, . . . , h
2. At k = k +h, as soon as
s(k) InteriorA
i
A
j
(8.6)
switch to CG
j
and solve
g(k) = arg min
wV
j
(s(k))
|w r(k + 1)|
2
, k k +h + 1
121
The illustrated scheme is motivated by the fact that for any s R
n
the state
evolution will enter in InteriorA
i
A
j
within a nite number of time in-
stants. An upper bound to this integer can be computed o-line with respect
to all s R
n
in a way similar to that used to determine the constraint horizon
j
0
dened in Chapter 7.
Then, instead of checking the set-membership condition in step 1), one could
determine such an upper bound h o-line and exploit it during the on-line op-
erations by waiting for exactly h steps before switching to step 2) above.
Of course, other possibilities for the switching logic exist, which could be more
eective for some applications. In all cases, the above guidelines, inspired by
[GK99], allows one to retain in a hybrid general context the same stability and
feasibility properties pertaining to the basic CG approach (7.14).
In conclusion, a hybrid CG scheme can be adopted dening a correct decom-
position of r into a well suited sequence of transition reference signals r
ij
and
a criterion for switching between the corresponding controllers. For planned
missions, the transition reference sequence can be dened a priori. Further, a
graph of connections and a look-up table can replace the switching criterion. In
the general case, however, a switching logic has to be employed. We propose
a criterion to characterize InteriorA
i
A
j
in (8.6), based on the Euclidean
norm between the state and the linearization point. The supervisor can main-
tain CG
i
as long as the distance between O
i
and present attitude, assumed
to be available, is minimal. On the contrary it can switch to the j th model,
centered in O
j
where CG
j
corresponds to
j = arg min
j(1, ,
i)
_
_
(
j
,
j
,
j
)
T
_
_
(8.7)
The criterion is depicted is Figure 8.2. Simulations results show that this switch-
ing criterion is eective in performing large angle maneuvers.
Finally notice that the above control scheme easily encompasses the control tech-
nique described in Chapter 7. Indeed, when J
, a single model linearized

around (, , )
T
= (0, 0, 0)
T
rad, covers the entire attitude reconguration
maneuver and no switching is required.
Figure 8.2. Switching logic for attitude reconguration.
8.3 Simulations of a LEO spacecraft performing Large
Angle Maneuvers
We refer again to a LEO satellite orbiting around Earth at a velocity
0
=
0.0011 rad/s. It could represent, for instance, a slave satellite in an Earth Ob-
serving Formation Flying System. For the space vehicle, we assume the same
mass and inertia matrix of previous examples, and the following saturation
constraints: [f[
max
= 5 10
2
N, [[
max
= 2 10
3
Nm. The maximum ampli-
tudes of disturbance forces are assumed to be [d
i
[
max
= 6 10
3
N, i = 1, 2, 3.
Moreover, the maximum amplitudes of disturbance torques are assumed to be
[d
i
[
max
= 2 10
4
Nm, i = 4, 5, 6. A value of = 10
4
in equation (7.6) is
selected.
The LEO spacecraft, leaving from positions (x, y, z)
T
= (0, 10, 0)
T
m in C, is
required to move away from the formation center, along the second axis direc-
tion, in order to reach positions (x, y, z)
T
= (0, 200, 0)
T
m. Further, it has to
maintain its pointing towards position (x, y, z)
T
= (100, 0, 0)
T
m. This could
be the position of another satellite, with which the spacecraft has to exchange in-
formation. It means that the satellite is required to follow a prescribed attitude
trajectory in accordance with the maneuver, consisting in a rotation around its
own third axis. Thus, only the angle has to be modied, whereas angles
and have to be maintained constantly at zero. Besides saturation constraints
123
w
0
w
0
Z
X
Y
{-100,0,0}
{0,10,0}
{0,200,0}
Figure 8.3. Reconguration for a LEO spacecraft, requiring a large angle maneuver.
we want to consider accuracy constraints [s
i
(k) r
i
((k))[ < 0.5 m for each of
the position components, i.e. for i = 1, 2, 3, and [s
i
(k) r
i
((k))[ < 0.1 rad
( 0.0017 rad) for each of the attitude components, i.e. for i = 4, 5, 6.
Results of the application of the hybrid CG controller (7.14) and (8.7) to the
satellite are illustrated in Figures 8.4 and 8.5. Here, the constraint horizon
0 0.5 1 1.5 2 2.5
x 10
4
50
100
150
y

[
m
]
0 0.5 1 1.5 2 2.5
x 10
4
0.05
0
0.05
f
2

[
N
]
0 0.5 1 1.5 2 2.5
x 10
4
0.5
0
0.5
Time [s]
e
2

[
m
]
Figure 8.4. Relative position y(t) (desired reference in dash lines), control force
f
2
(t) and error e
2
(t) in coordination accuracy for y(t) (constraint boundaries in
dash lines) under HCG. The vertical dash lines indicate instants of switching.
introduced in theorem 7.1, results j
0
= 51.
Values of (t) and (t) relative to equation (7.12), corresponding to the maneu-
0 0.5 1 1.5 2 2.5
x 10
4
60
40
20
0

[
d
e
g
]
0 0.5 1 1.5 2 2.5
x 10
4
1
0.5
0
0.5
1
x 10
3
3

[
N
m
]
0 0.5 1 1.5 2 2.5
x 10
4
0.1
0.05
0
0.05
0.1
Time [s]
e
3

[
d
e
g
]
Figure 8.5. Eulers angle (t) (desired reference in dash lines), control torque
3
(t)
and error e
3
(t) in coordination accuracy for (t) (constraint boundaries in dash
lines) under HCG. The vertical dash lines indicate instants of switching.
ver, are depicted in Figure 8.6. The active constraints are the coordination ones
0 0.5 1 1.5 2 2.5
x 10
4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time [s]
(
t
)
0 0.5 1 1.5 2 2.5
x 10
4
0
200
400
600
800
1000
1200
1400
1600
Time [s]
(
t
)
Figure 8.6. Values of (t) and (t) for the HCG selection logic (7.14), (8.7).
related to the position y(t). This is evident in Figure 8.4. Thus, in the present
125
example, the constraints that inuence the values of (t) in (7.14) are the ones
related to the position accuracy. Input forces and torques, on the contrary, are
well inside their respective saturation boundaries. Moreover, since the position
constrains are very stringent, input forces and torques take values close to zero
during the entire reconguration. The dash vertical lines depicted in Figures 8.4
and 8.5, indicate the time instants when the supervisor switches from the acting
CG to another one, selected according to rule (8.7). In the present simulation,
a dierent linear model has been associated to the system, for Eulers angles of
(0, 0, 0)
T
rad, (0, 0, 20)
T
rad, (0, 0, 40)
T
rad and so on.
Again, notice that, because of the use of the HCG, the reference trajectory is
tracked with a speed typically lower than the planned velocity. This is evident
in Figure 8.6 where (t) is less than one ((t) = 1 would correspond to nominal
tracking) and in Figure 8.4.
If the hybrid CG scheme is not used, the formation accuracy constraints related
to the position are no longer fullled as it results from Figure 8.7.
0 5000 10000 15000
50
100
150
y

[
m
]
0 5000 10000 15000
0.05
0
0.05
f
2

[
N
]
0 5000 10000 15000
1
0.5
0
0.5
1
Time [s]

e
2

[
m
]
Figure 8.7. Relative position y(t), control force f
2
(t) and error e
2
(t) in coordination
accuracy for y(t) under LQ (constraint boundaries in dash lines).
8.4 Conclusions
In this chapter we have proposed a technique for controlling the attitude of LEO
satellites, when they are required to span large angles. The presented scheme is
a natural extension of the technique illustrated in previous chapter. It consists
of a bank of Command Governors controllers, each one designed with respect
to a suitable attitude model. At any time instant, a supervisor selects the most
appropriate CG controller from the bank, on the basis of current state and pre-
scribed constraints. The proposed switching logic acts in such a way that the
supervisor always selects the i th CG, designed around the reference frame
O
i
, that is the closest to B in the bank.
The whole technique, addressing the attitude control problem for LEO space-
craft in a hybrid framework, allows at restoring the properties of the classical
CG control scheme, stated in theorem 7.1. Thus, simplicity and reduced numer-
ical burdens, proper of CG, are restored, with the only additional cost of solving
(8.7) at any time instant. Eectiveness of the technique has been illustrated by
simulation results.
127
Conclusion
Spacecraft motion has been subject of study for many years. Prior to the de-
velopment of space vehicles, scientist analyzed motion of heavy orbiting bodies,
giving rise to the discipline called celestial mechanics. Nowadays, two general
areas of study on the subject of spacecraft motion can be identied. The rst
one, namely astrodynamics, considers spacecraft motion in gravity elds. The
second one is attitude control. It is concerned with motion about the vehicles
center of mass. In this dissertation we have presented theory lying in this last
eld of research. In particular, we have proposed attitude mathematical models
and control techniques for satellites, traveling around Earth on geo-synchronous
and non geo-synchronous orbits.
Behavior of satellites on geo-stationary orbits is quite dierent with respect the
one pertaining to satellites moving on Medium and Low Earth orbits. Thus,
this dissertation has been broadly divided into two parts. The rst part has
dealt with the problem of modeling and controlling the attitude of spacecraft
on GEO orbits. Since in this case the satellite period can be made exactly equal
to the time it takes the Earth for rotating once about its axis, no translational
motion has been taken into account. As a consequence, we have addressed the
GEO spacecraft control problem in the more general framework of controlling a
rigid body, constrained to rotate about a pivot. Therefore, we have introduced
the new concept of 3D pendulum.
The second part of this dissertation, has dealt with the problem of modeling
and controlling the attitude of spacecraft on MEO and LEO orbits. In this case,
beyond rotational motion, a rigid body translational motion is observed from
Earth. Thus, we have introduced an orbital model, describing satellite rotations
and translations, with respect to a non-inertial reference frame.
Geo-stationary Spacecraft and the 3D Pendulum
A geo-stationary spacecraft is an orbiting rigid body, whose center of mass is
motionless with respect to Earth surface. Thus, it possesses only three rota-
tional degrees of freedom. They correspond to rotations generated by suitable
combinations of thrust actions. As a consequence, its dynamics can be repro-
duced on Earth, pivoting a 3D rigid body on its center of mass. This argument
has suggested the introduction of a new system, that we have dened in Chapter
1 as the 3D pendulum. Introduction of this new concept, has also been suggested
by a laboratory process, available at the Attitude Dynamics and Control Lab-
oratory, Department of Aerospace Engineering, University of Michigan. In this
dissertation we have referred to this system as Triaxial Attitude Control Testbed.
It actually consists of a 3D pendulum whose pivot can be located on the center
of mass or on another arbitrary point. This second case, i.e. the unbalanced
one, generalizes the geo-stationary spacecraft model, since it also considers the
presence of gravity.
In the rst part of this thesis we have analyzed in detail, mathematical descrip-
tions and control techniques for unbalanced 3D rigid pendulum. First of all,
in Chapter 1 we have analyzed several possible attitude models based both on
local and global coordinates. We have used Eulers angles, for showing that the
3D pendulum model embodies many other classical mechanical systems. In-
deed, in Chapter 2, we have proven that the spherical 2D pendulum model, the
planar 1D pendulum model and the heavy top model, are special cases of the
3D pendulum model. Moreover, we have shown that a symmetric 3D pendulum
can easily describe the motion of a spinning top under the table. This system
has a great theoretical value. Surprisingly, it has not been studied in previous
literature.
On the other hand, we have shown that global coordinates are more suitable for
describing the 3D pendulum motion and for relating it to attitude models for
CONCLUSION 129
geo-stationary satellites, proposed in prior literature. Indeed, they avoid singu-
larities, that are one of the main issues in controlling mechanical systems. Thus,
we have provided an attitude description based on Poissons variables. More-
over, noticing that the momentum about the vertical is a prime integral of the
system, we have performed a Routh reduction. We have obtained an attitude
model lying on S
2
R
3
. It describes the evolution of the gravity direction in
body-xed coordinates. Expressing the involved quantities in body-coordinates
is particularly convenient, because in space systems, torques are provided by
actuators mounted on the body.
Further, in Chapter 3, we have proposed passivity based control techniques for
stabilizing the natural equilibria of the system, i.e. the hanging equilibrium and
the inverted equilibrium. We have faced stabilization of arbitrary attitudes, in
Chapter 4. Selecting passivity-based control has been natural. Indeed, angular
velocity feedback renders the closed-loop, input-output passive, with the total
energy as storage function. Thus, suitably shaping the potential energy and
injecting damping, stabilizes any arbitrary orientation. Contrary to previous
literature, the control law that we have proposed in Chapter 4, provides nearly
global asymptotic stability. Moreover, the control scheme, has been shown
to generate input torques in the direction that avoids the unwinding phenom-
enon. Thus, it is particularly suitable in spacecraft control, since it reduces
fuel consumption. Finally, in Chapter 5, we have proven that a similar contin-
uous state feedback control law, again based on potential shaping and damping
injection, can be used for re-orienting a symmetric rigid body with only two
control torques. This is the classical problem of controlling underactuated GEO
satellites.
The possibility of designing simple continuous state-feedback controllers for sta-
bilizing any arbitrary attitude of 2D and 3D rigid bodies, appears a prominent
result in mechanics and aerospace literature. In this dissertation it has been
possible thanks to the idea of viewing all the related models as special cases of
the 3D pendulum. Its global attitude representation, has avoided the presence of
singularities and the consequent articial need to develop swing-up controllers,
local controllers, and strategies for switching between them.
Constrained MEO and LEO Spacecraft
We have faced the problem of controlling the attitude of MEO and LEO space-
craft, in the second part of this dissertation. Since they represent non-geo-syn-
chronous 3D rigid bodies, we have proposed a mathematical model describing
both attitude and position, with respect to an orbital reference frame. This
way to proceed is known as the leader following approach. MEO and LEO
satellites are usually small, if compared to GEO satellites. Indeed, they have
been designed for performing tasks, previously demanded to large space vehicles,
reducing costs for launch and maneuvers and increasing reliability. As a conse-
quence, they cannot be equipped with powerful thrusters. Thus, in Chapter 6,
we have introduced input saturation constraints. Moreover, we have described
the meaning and the role of state accuracy constraints, usually to be included
in keeping, reconguration and tracking maneuvers.
Assuming small angles, the orbital model describing LEO attitude dynamics, is
linear. Therefore, any control law designed for linear systems would be appro-
priate for stabilizing any desired attitude in the absence of constraints. Hence,
the presence of constraints is actually the main issue in LEO and MEO space
systems. As a consequence, in Chapter 7 we have proposed the use of an MPC
technique, referred to as command governor. It consists in designing a primal
control law and an external unit capable of taking care of constraint fulllment
by modifying, whenever necessary, the reference signal. It guarantees low nu-
merical burdens, that are fundamental in real time missions. Besides classical
CG scheme, we have proposed a parameterization of reference signal, in order
to handle state accuracy constraints.
Finally, we have extended the promising obtained results, to the case when MEO
and LEO satellites are required to span large angles. In this case, their attitude
orbital model is not linear. In order to restore results of Chapter 7, in Chapter
8 we have proposed a hybrid control technique. It consists of a bank of CG
controllers, each one designed with respect to a suitable attitude conguration.
At any time instant, a supervisor selects the most appropriate CG controller
from the bank, on the basis of current state and prescribed constraints. We
have shown that this control law allows at retaining all the results obtained in
the presence of small angles, with the only additional price of evaluating the
CONCLUSION 131
minimum of a quadratic function, at any time instant.
Research Contribution and Future Developments
This dissertation has presented several results on controlling space vehicles.
Both geosynchronous and non-geosynchronous satellites have been taken into
account, dividing the thesis into two parts. The main contribution of the rst
part, is the idea of describing GEO satellites as a special case of a 3D pendu-
lum. Although many works in literature use an attitude model similar to the
one proposed in Chapter 1, the idea is new. As a consequence, it produces new
mechanical systems, as the heavy top rotating below the pivot. Our control
technique for attitude stabilization, based on the idea of pushing the rigid body
away from unstable equilibria, by shaping the potential, is also new. It guaran-
tees the presence of a continuous state-feedback, almost globally stabilizing any
desired attitude. In the special case of the 2D pendulum, it solves the swing-
up problem, using one single control law. On the contrary, previous literature
proposes controllers that drive the pendulum in proximity of the upward cong-
uration, local controllers around the upward conguration, and techniques for
switching between them. Furthermore, the proposed control technique, avoids
the unwinding phenomenon. Thus, it represents an optimal strategy, in the
sense that it always pushes the rigid body in the right direction.
The TACT, available at the Attitude Dynamics and Control Laboratory, De-
partment of Aerospace Engineering, University of Michigan, is a natural lab-
oratory process for testing all the control laws proposed in the rst part of
this dissertation. Many experiments have already been performed in the un-
balanced case, conrming theoretical results. On the other hand, work still has
to be done, for making experiments in the balanced case. The TACT has to
be set-up for making gravity ineective. This represents a crucial step in the
present eld of research, before applying the proposed controllers to real space
systems. Studying strategies for selecting optimal and functions, making
controllers proposed in Chapter 4 and in Chapter 5 minimizing some suitable
cost function, is also an interesting subject of future research. Moreover, eorts
can be directed to the modication of present results to dierent actuation tech-
nologies, as rotation wheels or proof masses.
As widely discussed above, the second part of this dissertation treats aspects of
spacecraft motion, more technological than the ones analyzed in the rst part.
Beyond the attitude orbital model presented in Chapter 6, that appears new,
the attention has been focused on constraints, that are naturally present in all
the real missions. The main contribution of part II, lies on the hybrid command
governor (HCG) approach illustrated in Chapter 8. Although it is based on
an MPC strategy, as other control techniques proposed in previous literature,
it represents a novelty in non-geostationary space systems. Indeed, it allows
at facing large angle maneuvers using a combination of simple linear attitude
models. Moreover, it can be directly extended to formation ying systems, just
enclosing attitude and position of all the members of the formation, in the sys-
tem state s(k).
On the other hand, some aspects need future development. For instance, ef-
fectiveness of the switching technique, proposed in Chapter 8, has been proven
only by simulation results. Proof of the properties of such a switching logic rep-
resents a rst step of future research. Moreover, the study of dierent switching
criterions, possibly associated with common Lyapunov functions, appears an
interesting direction of development.
The research on attitude spacecraft control is a fervent eld of research. Many
aspects, both theoretical and technological are under development. It is testied
by the several recent works referred by this dissertation. In this thesis we have
faced some of these aspects. From a theoretical point of view, we have proposed
a new concept related to mechanical systems, dened 3D pendulum. From a
technological point of view, we have proposed a hybrid control strategy for low
altitude spacecraft. Along with the development of nano and pico-satellite, it
represents a starting point for designing future formation ying systems, both
orbiting around Earth and traveling in deep space.
CONCLUSION 133
Bibliography
[ABC
+
03] A. Albertoni, A. Ballucchi, A. Casavola, C. Gambelli, E. Mosca, and
A.L. Sangiovanni Vincentelli. Hybrid command governor approach
for idle speed control in gasoline direct injection engines. In Proc.
of American Control Conference, pages 773778, 2003.
[AF00] K.J. Astrom and K. Furuta. Swinging up a pendulum by energy
control. Automatica, 36(2):287295, 2000.
[Ang01] D. Angeli. Almost global stabilization of the inverted pendulum via
continuous state feedback. Automatica, 37(7):11031108, 2001.
[BB00] S.P. Bhat and D.S. Bernstein. A topological obstruction to con-
tinuous global stabilization of rotational motion and the unwinding
phenomenon. Systems & Control Letters, 39:6370, 2000.
[BCM97] A. Bemporad, A. Casavola, and E. Mosca. Nonlinear control of con-
strained linear systems via predictive reference menagement. IEEE
Transactions on Automatic Control, 42:340349, 1997.
[BLM00] A.M. Bloch, N.E. Leonard, and J.E. Marsden. Controlled la-
grangians and the stabilization of mechanical systems i: The rst
matching theorem. IEEE Transactions on Automatic Control,
45(12):148, 2000.
[BLM01] A.M. Bloch, N.E. Leonard, and J.E. Marsden. Controlled la-
grangians and the stabilization of mechanical systems ii: Potential
shaping. IEEE Transactions on Automatic Control, 46(10):1556
1571, 2001.
[BMB01] D.S. Bernstein, N.H. McClamroch, and A. Bloch. Development of
air spindle and triaxial air bearing testbeds for spacecraft dynamics
and control experiments. In Proc. of American Control Conference,
pages 39673972, 2001.
[BMS83] R.W. Brockett, R.S. Millman, and H.J. Sussmann. Asymptotic sta-
bility and feedback stabilization. In Dierential Geometric Control
Theory, pages 181191, 1983.
[BMW71] R.R. Bate, D.D. Mueller, and J.E. White. Fundamentals of Astro-
dynamics. Dover Publications, New York, USA, 1971.
[BS96] A. Bradshaw and J. Shao. Swing-up control of inverted pendulum
systems. Robotica, 14:397405, 1996.
[Bur95] I.V. Burkov. Asymptotic stabilization of the position of a rigid
body with xed point without velocity measurements. System and
Control Letters, 25:205209, 1995.
[CBA
+
06] N.A. Chaturvedi, D.S. Bernstein, J. Ahmed, F. Bacconi, and N.H.
McClamroch. Globally convergent adaptive tracking of angular ve-
locity and inertia identication for a 3-dof rigid body. IEEE Trans-
actions on Control Systems Technology, 14(5):841853, 2006.
[CLT00] M. Corless, J.M. Longuski, and P. Tsiotras. A novel approach
to the attitude control of axi-symmetric spacecraft. Automatica,
31(8):10991112, 2000.
[CM02] S. Cho and N.H. McClamroch. Feedback control of triaxial attitude
control testbed actuated by two proof mass devices. In Proceedings
of 41st IEEE Conference on Decision and Control, pages 498503,
2002.
[CMA00] A. Casavola, E. Mosca, and D. Angeli. Robust command governors
for constrained linear systems. IEEE Transactions on Automatic
Control, 45:20712077, 2000.
BIBLIOGRAPHY 135
[CSM03] S. Cho, J. Shen, and N.H. McClamroch. Mathematical models for
the triaxial attitude control testbed. Mathematical and Computer
Modeling of Dynamical Systems, 9(2):165192, 2003.
[CSMB01] S. Cho, J. Shen, N.H. McClamroch, and D.S. Bernstein. Equations
of motion of the triaxial attitude control testbed. In Proc. of of 40th
IEEE Conference on Decision and Control, pages 34293434, 2001.
[DT00] V. Doumtchenko and P. Tsiotras. Control of spacecraft subject to
actuator failures: State-of-the-art and open problems. Journal of
the Astronautical Sciences, 48(2,3):337358, 2000.
[ENS
+
03] J. Esper, S. Neeck, J.A. Slavin, J. Leitner, W. Wiskombe, and F.H.
Bauer. Nano/micro satellite constellations for earth and space sci-
ence. Acta Astronautica, 52:785791, 2003.
[FF94] O.E. Fiellstad and T.I. Fossen. Comments on the attitude control
problem. IEEE Transactions on Automatic Control, 39(3):699
700, 1994.
[FPW98] G. Franklin, G. Powell, and M. Workman. Digital Control of Dy-
namic Systems (3rd edn). Addison-Wesley, Reading, MA, 1998.
[Fur03] K. Furuta. Control of pendulum: From super mechano-system to
human adaptive mechatronics. In Proc. of 42
nd
IEEE Conference
on Decision and Control, pages 14981507, 2003.
[GD99] F. Ghorbel and J.B. Dabney. Spherical pendulum system to teach
key concepts in kinematics, dynamics, control, and simulation.
IEEE Transactions on Education, 42(4):359, 1999.
[GE95] J.M Godhavn and O. Egeland. Attitude control of underactuated
satellites. In Proc. of 34
nd
IEEE Conference on Decision and Con-
trol, 1995.
[GH83] J. Guckenheimer and P. Holmes. Nonlinear Oscillations, Dynamical
Systems, and Bifurcations of Vector Fields. Springer-Verlag, Berlin,
1983.
[GK92] T.J. Graettinger and B.H. Krogh. On the computation of reference
signal constraints for guaranteed tracking performance. Automatica,
28:11251141, 1992.
[GK99] E.G. Gilbert and I.V. Kolmanovsky. Set-point control of nonlinear
systems with state and control constraints: A lyapunov-function,
reference governor approach. In Proceedings of the 37th IEEE Con-
ference on Decision and Control, pages 25072512, 1999.
[GKT95] E.G. Gilbert, I. Kolmanovsky, and K. Tin Tan. Discrete-time refer-
ence governors and the nonlinear control of systems with state and
control constraints. International Journal of Robust and Nonlinear
Control, 1995.
[GPS80] H. Goldstein, C. Poole, and J. Safko. Classical Mechanics. Addison
Wesley, 1980.
[Hal02] C. Hall. Spacecraft Attitude Dynamics and Control. Lecture Notes,
Chapter 3, 2002.
[Hat01] A. Hatcher. Algebraic Topology. Cambridge University Press, Cam-
bridge, UK, 2001.
[HRW04] D. Halliday, R. Resnick, and J. Walker. Foundametals of Physics.
John Wiley & Sons, New York, USA, 2004.
[Hug86] P.C. Huges. Spacecraft Attitude Dynamics. John Wiley & Sons,
New York, USA, 1986.
[Kap76] M.H. Kaplan. Modern Spacecraft Dynamics and Control. John
Wiley & Sons, New York, USA, 1976.
[Kha01] H.K. Khalil. Nonlinear Systems. Prantice Hall, Third Edition, New
Jersey, USA, 2001.
[KK73] D. Kleppner and R. Kolenkov. An Introduction to Mechanics.
McGraw-Hill Science, 1973.
[MNF76] S. Mori, H. Nisihara, and K. Furuta. Control of unstable mechanical
systems: Control of pendulum. International Journal of Control,
23:673692, 1976.
BIBLIOGRAPHY 137
[Mos95] E. Mosca. Optimal, Predictive, and Adaptive Control. Prentice Hall,
Englewood Clis, NJ, 1995.
[MR99] J.E. Marsden and T.S. Ratiu. Introduction to Mechanics and
Symmetry: A Basic Exposition of Classical Mechanical Systems.
Springer Verlag, California, USA, 1999.
[MRS00] J.E. Marsden, T.S. Ratiu, and J. Scheurle. Reduction theory and
the lagrange-routh equations. Journal of Mathematical Physics,
41(6):33793429, 2000.
[OSR98] R. Ortega and H. Sira-Ramirez. Passivity-Based Control of Euler-
Lagrange Systems: Mechanical, Electrical and Electromechanical
Applications. Springer Verlag, California, USA, 1998.
[SBW06] G.L. Slater, S.M. Byram, and T.W. Williams. Collision avoidance
for satellites in formation ight. Journal of Guidance, Control, and
Dynamics, 29(5):11401146, 2006.
[SCB
+
05] M.A. Santillo, N.A. Chaturvedi, F. Bacconi, N.H. McClamroch, and
D.S. Bernstein. Experiments on stabilization of the hanging equi-
librium of a 3d asymmetric rigid pendulum. In Proceedings of IEEE
Conference on Control Applications, 2005.
[SEL99] A.S. Shiriaev, O. Egeland, and H. Ludvigsen. Global stabilization
of unstable equilibrium point of pendulum. In Proc. of American
Control Conference, pages 40344038, 1999.
[SEL04] A.S. Shiriaev, O. Egeland, and H. Ludvigsen. Swinging up the
spherical pendulum via stabilization of its rst integrals. Automat-
ica, 40(1):7385, 2004.
[Sid00] M.J. Sidi. Spacecraft Dynamics and Control: a Practical Engineer-
ing Approach. Cambridge University Press, Cambridge, UK, 2000.
[SLRM92] J.C. Simo, D. Lewis, T. Ratiu, and J.E. Marsden. The heavy top:
a geometric treatment. Nonlinearity, 5:22532270, 1992.
[Son00] E.D. Sontag. The iss philosophy as a unifying framework for
stability-like behavior. Nonlinear Control in the Year 2000, 2:443
468, 2000.
[SPLE00] A.S. Shiriaev, A. Progomski, H. Ludvigsen, and O. Egeland. On
global properties of passivity-based control of an inverted pendulum.
International Journal of Robust and Nonlinear Control, 10(4):283
300, 2000.
[TIH02] M.T. Tillerson, G. Inalhan, and J.P. How. Co-ordination and con-
trol of distributed spacecraft systems using convex optimization
techniques. International Journal of Robust and Nonlinear Con-
trol, 12:207242, 2002.
[VGRM01] S.M. Veres, S.B. Gabriel, E. Rogers, and D.Q. Mayne. Analysis of
formation ying control of a pair of nano-satellites. In Proceedings
of the 40th IEEE Conference on Decision and Control, pages 1095
1100, 2001.
[WCB95] C.J. Wan, V.T. Coppola, and D.S. Bernstein. Global asymptotic
stabilization of the spinning top. Optimal Control Applications &
Methods, 16:189215, 1995.
[WD91] J.T.Y. Wen and K.K. Delgado. The attitude control problem. IEEE
Transactions on Automatic Control, 36(10):11481162, 1991.
[WTCB94] C.J. Wan, P. Tsiotras, V.T. Coppola, and D.S. Bernstein. Global
asymptotic stabilization of the spinning top with torque actuators
using stereographic projection. In Proc. of American Control Con-
ference, pages 536540, 1994.
BIBLIOGRAPHY 139

Spacecraft Attitude Dynamics Bacconi

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Spacecraft Attitude Dynamics Bacconi

Загружено:

Авторское право:

Доступные форматы

UNIVERSIT

. It is useful to control the attitude of satellites in

and then a third rotation of roll angle

I. Thus, R in (1.3) corresponds to

. However, this is not a limitation in

V is not denite. Hence, V is a continuously dierentiable function such that

is the constant generalized momentum corresponding to :

/, where H and / are as dened in Lemmas

The result stated by the previous corollary is prominent in aerospace control

5). As expected, angular ve-

is introduced, as depicted in Figure 6.2.

is the one obtained from O

after a second rotation of

. For brevity we have replaced cx =

A. Here r(k) is the reference, P

is the ball of radius centered at the origin. In particular, J

, such that the corresponding virtual evolution fullls

. Moreover, if non-empty, it represents the

whose evolutions starting from s

s. It has been shown in [BCM97] and

w(k) := r((k 1) + (k))

i), orbiting in ECI along with C,

in C are constant and known, we can dene the angular velocity of C in

dened in (7.6). Thus,

i), with overlapping interior corresponding to

Now, if (i, j) is such that InteriorJ

, a single model linearized

Вам также может понравиться