-
Communication Loss Management and Analysis for
MultipleSpacecraft Formation Flying Missions
by
Mohamed Elnabelsya
A thesis submitted in conformity with the requirementsfor the
degree of Master of Applied Science
Graduate Department of Aerospace Science and
EngineeringUniversity of Toronto
Copyright c© 2010 by Mohamed Elnabelsya
-
Abstract
Communication Loss Management and Analysis for Multiple
Spacecraft Formation
Flying Missions
Mohamed Elnabelsya
Master of Applied Science
Graduate Department of Aerospace Science and Engineering
University of Toronto
2010
This thesis presents a method for managing periods of
communication loss between
multiple spacecraft in formation flying (MSFF), and analyzes the
effects of this method
on the stability of the formation keeping control algorithm. The
controller of interest in
this work in an adaptive nonlinear controller, where
synchronization is also incorporated
to force the position tracking errors to converge to zero at the
same rate. The com-
munication loss compensation technique proposed in this thesis
is to use the previously
communicated data in lieu of the lost data, which is an
effective and computationally-
efficient technique that is advantageous for small satellites.
The performance parameter
of interest in this research is the maximum rate of
communication loss that an MSFF sys-
tem can withstand before going unstable, and this is analyzed
theoretically and through
simulations. Finally, experiments involving multiple robots in
formation with communi-
cation loss are conducted, and the results are presented.
ii
-
Dedication
I dedicate this thesis to my Family - my mother Ihsan, my father
Gamal, and my two
little sisters Reem and Nada.
iii
-
Acknowledgements
This work would not have been possible without the support of
many in my life. I
would like to thank God for giving me the strength to complete
yet another endeavour
in my life, and to thank my family for their continuous love and
support.
I would like to thank Professor Hugh H. T. Liu for his
tremendous support and for
believing in me throughout my research period. His continuous
guidance was an integral
part of this work’s completion, and his high expectations and
inspiring supervisory style
have made this an exceptional learning experience. I always
sensed how much he cares
for his students and how his main goals are to help them become
great researches and
contribute strongly to their fields of studies, and for that I
will be forever grateful.
I would like to thank Professor Chris Damaren for creating two
excellent courses -
Spacecraft Dynamics and Control I and II - and for his great
teaching style and careful
selection of the material to cover. Those were two of the most
educational courses I have
ever taken, and I know I will continuously refer back to this
material long after I leave
UTIAS.
I would like to thank Keith Leung for guiding me throughout the
experimental phase
of my research. Without his patience and continuous assistance,
the experiments would
have taken a much longer time to get completed. I would like to
also thank Valentin
Peretroukhin for assisting me with some of the experimental
work.
Finally, I would like to thank all my friends at UTIAS for
making my stay a truly
wonderful and unforgettable experience.
iv
-
Contents
1 Introduction 1
1.1 Literature Review . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 2
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 4
1.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 5
2 System Modelling 6
2.1 MSFF Position Dynamics Model . . . . . . . . . . . . . . . .
. . . . . . 6
2.2 Adaptive Control Formulation . . . . . . . . . . . . . . . .
. . . . . . . . 8
2.3 Integrating Synchronization Control . . . . . . . . . . . .
. . . . . . . . . 9
2.4 Simulation Results . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 11
2.4.1 Adaptive Control Simulation . . . . . . . . . . . . . . .
. . . . . . 11
2.4.2 Adaptive Synchronization Control Simulation . . . . . . .
. . . . 14
3 Communication Loss Stability Analysis 18
3.1 Asynchronous Dynamical Systems . . . . . . . . . . . . . . .
. . . . . . . 19
3.1.1 Stability Criteria . . . . . . . . . . . . . . . . . . . .
. . . . . . . 20
3.1.2 Controller Estimate . . . . . . . . . . . . . . . . . . .
. . . . . . . 24
3.1.3 Discretization . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 24
3.2 Numerical Example . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 26
3.2.1 Theoretical Analysis . . . . . . . . . . . . . . . . . . .
. . . . . . 27
3.2.2 Linear System Simulation . . . . . . . . . . . . . . . . .
. . . . . 27
v
-
3.2.3 Nonlinear System Simulation . . . . . . . . . . . . . . .
. . . . . 29
4 ADS Stability Analysis with Synchronization Control 31
4.1 ADS Model with Synchronization . . . . . . . . . . . . . . .
. . . . . . . 31
4.2 Numerical Example . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 34
4.2.1 Theoretical Analysis . . . . . . . . . . . . . . . . . . .
. . . . . . 34
4.2.2 Linear System Simulation . . . . . . . . . . . . . . . . .
. . . . . 35
4.2.3 Nonlinear System Simulation . . . . . . . . . . . . . . .
. . . . . 38
5 Robots in Formation Analysis and Experiments 42
5.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 42
5.1.1 UTIAS Resources . . . . . . . . . . . . . . . . . . . . .
. . . . . . 42
5.1.2 Test Scenario . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 45
5.2 System Model . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 45
5.3 Controller Formulation . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 46
5.3.1 Control Law . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 46
5.3.2 Smooth Differentiation . . . . . . . . . . . . . . . . . .
. . . . . . 48
5.4 Theoretical Analysis . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 50
5.5 Code Implementation . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 51
5.5.1 Formation Code . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 51
5.5.2 Communication Loss Code . . . . . . . . . . . . . . . . .
. . . . . 56
5.6 Experimental Work . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 57
6 Conclusion and Future Work 63
6.1 Thesis Motivation . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 63
6.2 Thesis Content . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 63
6.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 65
A Formation Code for Follower 1 66
vi
-
List of Tables
5.1 A050166 Formatted as a Square Array . . . . . . . . . . . .
. . . . . . . 49
5.2 Differentiation Filter Example . . . . . . . . . . . . . . .
. . . . . . . . . 50
vii
-
List of Figures
2.1 Schematic representation of an MSFF system . . . . . . . . .
. . . . . . 6
2.2 Relative Trajectory of the MSFF System with No
Synchronization . . . . 12
2.3 Tracking Error of Follower Spacecraft 1 with No
Synchronization . . . . . 12
2.4 Internal Sync Error of Follower Spacecraft 1 with No
Synchronization . . 12
2.5 Tracking Error of Follower Spacecraft 2 with No
Synchronization . . . . . 12
2.6 Internal Sync Error of Follower Spacecraft 2 with No
Synchronization . . 12
2.7 Tracking Error of Follower Spacecraft 3 with No
Synchronization . . . . . 13
2.8 Internal Sync Error of Follower Spacecraft 3 with No
Synchronization . . 13
2.9 Tracking Error of Follower Spacecraft 4 with No
Synchronization . . . 13
2.10 Internal Sync Error of Follower Spacecraft 4 with No
Synchronization . . 13
2.11 External Sync Error of the x Axis with No Synchronization .
. . . . . . . 13
2.12 External Sync Error of the y Axis with No Synchronization .
. . . . . . . 13
2.13 External Sync Error of the z Axis with No Synchronization .
. . . . . . . 14
2.14 Relative Trajectory of the MSFF System with Synchronization
. . . . . . 14
2.15 Tracking Error of Follower Spacecraft 1 with
Synchronization . . . . . . . 15
2.16 Internal Sync Error of Follower Spacecraft 1 with
Synchronization . . . . 15
2.17 Tracking Error of Follower Spacecraft 2 with
Synchronization . . . . . . . 15
2.18 Internal Sync Error of Follower Spacecraft 2 with
Synchronization . . . . 15
2.19 Tracking Error of Follower Spacecraft 3 with
Synchronization . . . . . . . 15
2.20 Internal Sync Error of Follower Spacecraft 3 with
Synchronization . . . . 15
viii
-
2.21 Tracking Error of Follower Spacecraft 4 with
Synchronization . . . . . . . 16
2.22 Internal Sync Error of Follower Spacecraft 4 with
Synchronization . . . . 16
2.23 External Sync Error of the x Axis with Synchronization . .
. . . . . . . . 16
2.24 External Sync Error of the y Axis with Synchronization . .
. . . . . . . . 16
2.25 External Sync Error of the z Axis with Synchronization . .
. . . . . . . . 16
3.1 ADS Communication Loss Model . . . . . . . . . . . . . . . .
. . . . . . 19
3.2 Relative Trajectory of the Linearized System with 98%
Communication Loss 28
3.3 Tracking Error of the Linearized System with 98%
Communication Loss . 28
3.4 Relative Trajectory of the Linearized System with 99%
Communication Loss 28
3.5 Tracking Error of the Linearized System with 99%
Communication Loss . 28
3.6 Relative Trajectory of the Nonlinear System with 98%
Communication Loss 29
3.7 Tracking Error of the Nonlinear System with 98%
Communication Loss 29
3.8 Relative Trajectory of the Nonlinear System with 99%
Communication Loss 29
3.9 Tracking Error of the Nonlinear System with 99%
Communication Loss 29
4.1 Relative Trajectory of the Linearized System with 89%
Communication Loss 36
4.2 Tracking Error of Follower Spacecraft 1 with 89%
Communication Loss . 36
4.3 Tracking Error of Follower Spacecraft 2 with 89%
Communication Loss . 36
4.4 Tracking Error of Follower Spacecraft 3 with 89%
Communication Loss . 36
4.5 Tracking Error of Follower Spacecraft 4 with 89%
Communication Loss . 36
4.6 Relative Trajectory of the Linearized System with 90%
Communication Loss 37
4.7 Tracking Error of Follower Spacecraft 1 with 90%
Communication Loss . 37
4.8 Tracking Error of Follower Spacecraft 2 with 90%
Communication Loss . 37
4.9 Tracking Error of Follower Spacecraft 3 with 90%
Communication Loss . 37
4.10 Tracking Error of Follower Spacecraft 4 with 90%
Communication Loss . 37
4.11 Relative Trajectory of the Linearized System with 94%
Communication Loss 38
4.12 Tracking Error of Follower Spacecraft 1 with 94%
Communication Loss . 38
ix
-
4.13 Tracking Error of Follower Spacecraft 2 with 94%
Communication Loss . 38
4.14 Tracking Error of Follower Spacecraft 3 with 94%
Communication Loss . 38
4.15 Tracking Error of Follower Spacecraft 4 with 94%
Communication Loss . 39
4.16 Relative Trajectory of the Linearized System with 95%
Communication Loss 39
4.17 Tracking Error of Follower Spacecraft 1 with 95%
Communication Loss . 39
4.18 Tracking Error of Follower Spacecraft 2 with 95%
Communication Loss . 39
4.19 Tracking Error of Follower Spacecraft 3 with 95%
Communication Loss . 39
4.20 Tracking Error of Follower Spacecraft 4 with 95%
Communication Loss . 40
5.1 A Vicon Camera . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . 43
5.2 Lab Area . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 44
5.3 An iRobot Create Roomba with Markers and a Laptop . . . . .
. . . . . 44
5.4 Simplified Robot Model . . . . . . . . . . . . . . . . . . .
. . . . . . . . 46
5.5 Schematic representation of the MRF Configuration . . . . .
. . . . . . . 58
5.6 No Communication Loss, Robots’ Trajectories . . . . . . . .
. . . . . . . 58
5.7 No Communication Loss, Follower 1 Tracking Error . . . . . .
. . . . . . 58
5.8 No Communication Loss, Follower 2 Tracking Error . . . . . .
. . . . . . 59
5.9 No Communication Loss, Follower 3 Tracking Error . . . . . .
. . . . . . 59
5.10 No Communication Loss, Follower 4 Tracking Error . . . . .
. . . . . . . 59
5.11 50% Communication Loss, Robots’ Trajectories . . . . . . .
. . . . . . . 59
5.12 50% Communication Loss, Follower 1 Tracking Error . . . . .
. . . . . . 59
5.13 50% Communication Loss, Follower 2 Tracking Error . . . . .
. . . . . . 59
5.14 50% Communication Loss, Follower 3 Tracking Error . . . . .
. . . . . . 60
5.15 50% Communication Loss, Follower 4 Tracking Error . . . . .
. . . . . . 60
5.16 95% Communication Loss, Robots’ Trajectories . . . . . . .
. . . . . . . 60
5.17 95% Communication Loss, Follower 1 Tracking Error . . . . .
. . . . . . 60
5.18 95% Communication Loss, Follower 2 Tracking Error . . . . .
. . . . . . 60
5.19 95% Communication Loss, Follower 3 Tracking Error . . . . .
. . . . . . 60
x
-
5.20 95% Communication Loss, Follower 4 Tracking Error . . . . .
. . . . . . 61
5.21 99% Communication Loss, Robots’ Trajectories . . . . . . .
. . . . . . . 61
5.22 99% Communication Loss, Follower 1 Tracking Error . . . . .
. . . . . . 61
5.23 99% Communication Loss, Follower 2 Tracking Error . . . . .
. . . . . . 61
5.24 99% Communication Loss, Follower 3 Tracking Error . . . . .
. . . . . . 61
5.25 99% Communication Loss, Follower 4 Tracking Error . . . . .
. . . . . . 61
xi
-
List of Acronyms
MSFF: Multiple spacecraft in formation flying
LGQ: Linear quadratic Gaussian
MIMO: Multiple-input multiple-output
ADS: Asynchronous dynamical systems
NCS: Networked control systems
BMI: Bilinear matrix inequality
LQR: Linear quadratic regulator
LQG: Linear quadratic Gaussian
UTIAS: University of Toronto institute for aerospace studies
MRF: Multiple robots in formation
xii
-
Chapter 1
Introduction
Multiple spacecraft in formation flying (MSFF) is defined as a
set of multiple spacecraft
where their states are coupled through a common control law
[44]. MSFF have been a
topic of major interest since the 1960s when some projects
required multiple spacecraft
to perform rendezvous manoeuvres [48]. Advancement in this field
did not progress in
a significant manner since then until interest was renewed in
the mid 1990s with the
launch of Earth Orbiter-1, the first autonomous formation flying
mission. Ever since,
major strides and development have occurred and as research
continues, the technology
would reach the level where MSFF can be integrated into many
proposed space projects
[52].
MSFF has many advantages, paramount amongst them is the ability
to perform
special missions which would not be possible using a single
large spacecraft. Formation
flying can enhance a space mission’s performance by distributing
mission tasks over many
small spacecraft, and if accurate MSFF is achieved, multiple
satellites can be arranged as
to create a single large entity; this would result in space
instruments with unprecedented
abilities, such as higher resolution imagery and interferometry.
MSFF could also be
utilized to optimize planetary surveillance, exploration, and
observation missions, and
because they have the ability to achieve a longer mission life,
MSFF could generate
greater scientific return [48] [52] [7] [57].
Having multiple spacecraft perform a mission increases the
redundancy and reliability
of the system if something goes wrong, and a defected component
would mean sending
a replacement satellite, rather than redesigning and building
the whole system. It also
allows the whole system to be reconfigured by simply sending the
appropriate commands,
and combined with the structural flexibility of an MSFF system,
this means the same
1
-
Chapter 1. Introduction 2
group of spacecraft could be used for various different missions
interchangeably. Sensing
capabilities of the system also expand with MSFF, and
maintenance becomes more man-
ageable and cost efficient. Finally, autonomous formation
reduces ground maintenance
requirements, lowering the operational costs [48] [28] [57]
[59].
1.1 Literature Review
For autonomous MSFF to be realistic and successful in space
missions, communication
between the spacecraft also needs to be autonomous. A
ground-based command and
control system for relative positioning of the spacecraft will
be very complex and may
not be able to provide sufficiently rapid corrective control
commands for manoeuvres,
formation reconfiguration, or collision avoidance [24]. Of the
ongoing research currently
taking place for autonomous MSFF, the main areas being looked at
are formation sensing
(where technology is used to develop more accurate sensors) and
formation control (where
control algorithms are derived for the spacecraft to maintain
the desired formation) [22].
Research on formation sensing attempts to improve the autonomy
of spacecraft naviga-
tion through the use of wide area differential and differential
carrier phase GPS (global
positioning system.) On the formation control front, work is
being conducted on the use
of absolute vs. relative control architectures, including the
communication requirements
and formation trajectories generated by these architectures.
This thesis will focus on
MSFF control.
Five main architectures for controlling MSFF have been studied
and examined in
the literature: multiple-input multiple-output (MIMO), virtual
structure, behavioural,
cyclic, and leader-follower [44]. MIMO control techniques
require the MSFF system to
be modeled as a single plant with multiple inputs and multiple
outputs, and some of
the control techniques used with MIMO architectures include
linear quadratic Gaussian
(LQG) and directed graphs [18] [47] [51]. Virtual structure
architecture assumes the
MSFF system is a rigid body, where the overall motion of the
system include rigid body
motions with expansions and contractions; Controlling the
virtual structure of an MSFF
mission requires fitting a formation template to the position of
the spacecraft within
the formation at every time step [29] [30]. Behavioural
architecture combines different
controllers, each with different objectives, to achieve an
overall state for the MSFF system
[4] [5]. In the cyclic architecture, the controller is formed by
connecting controllers of
the spacecraft in the formation non-hierarchically, and some of
the controllers derived for
-
Chapter 1. Introduction 3
this architecture include potential field and rule-based
controllers [34] [40] [50]. Finally,
the leader-follower approach is similar to the cyclic approach
in that the controller is
formed by connecting controllers of the spacecraft within the
formation, however this
is done hierarchically. Various control laws have been studied
for the leader-follower
architecture including sliding mode control, model predictive
control, adaptive control,
intelligent control, synchronization control and H∞ control [41]
[33] [9] [53] [20] [56][58]. The leader-follower architecture is
the most developed and examined of the five
architectures mentioned [44].
The dynamics and control of a single spacecraft is well
established and understood,
mainly due to the rigid body assumption which denotes that no
relative motion exists
between 2 points on the body. In MSFF, the ultimate goal is for
the spacecraft to act like
a single rigid body in their movement and manoeuvres, however in
reality a formation
of spacecraft behaves like a deformable body due to relative
position tracking errors and
synchronization errors, and so control forces are required to
restore the spacecraft to
their desired formation [35]. An adaptive nonlinear controller
has been proposed that
guarantees tracking errors of the relative positions converge to
zero, which adopts the
leader-follower approach, and simulation results have proven the
effectiveness of this
technique [9]. This controller however does not force follower
satellites to track their
trajectories at the same rate, thus during manoeuvres the
geometric configuration gets
distorted. A synchronization technique and algorithm can be
integrated into the adaptive
nonlinear control strategy which guarantees that the
synchronization error of the relative
positions also converge to zero [20]. This means that the
formation configuration can be
kept during manoeuvres, which is highly desirable in MSFF, and
again simulation results
have proven the effectiveness of this technique.
Precise MSFF requires the follower spacecraft to sense changes
in the states of the
leader spacecraft, and so inter-satellite communication becomes
integral to the success of
autonomous MSFF missions. This means that the response of the
system when commu-
nication loss occurs needs to be analyzed and understood.
Temporarily and periodic loss
of communication can occur at any time due to many factors,
including solar winds and
component failures [22]. Communication loss analysis has been
studied for some dynamic
systems, but very little work has been conducted on studying the
effects of communi-
cation loss on MSFF. The majority of work on communication loss
analysis have been
conducted on linear time-invariant systems, and most of the
published work on communi-
cation loss has dealt with multi-agent systems [38] [36] [17]
[37], asynchronous dynamical
-
Chapter 1. Introduction 4
systems (ADS) [1] [2] [39], Markov jump linear systems [10] [15]
[6], and networked con-
trol systems (NCS) [21] [26] [3] [55]. In [39], stability
criteria are established for nonlinear
systems, however applying these criteria to non-trivial dynamic
systems is challenging.
Studying the different communication loss analysis methods
proposed in those papers
can lead to techniques for examining the effects of
communication loss in MSFF.
1.2 Objectives
Various controllers have been proposed for MSFF, including
adaptive nonlinear control
[9]. The advantage of using an adaptive controller is that the
relative positions of the
spacecraft are guaranteed to converge for a suitably smooth
trajectory even in the pres-
ence of constant, slow varying, or unknown parameters (i.e. the
masses of the spacecraft).
This controller is developed utilizing a leader-follower
architecture, where the equations
of dynamics describe the states of the follower spacecraft
relative to the leader spacecraft.
Since the control law depends on the relative states of the
follower spacecraft with respect
to the leader, it is important to understand how the system
behaves during periods of
communication loss, and how robust the controller is to periodic
packet loss between the
spacecraft. In [20], a synchronization algorithm is integrated
into the nonlinear adaptive
controller to force the position tracking errors in the MSFF
system to converge to zero
at the same rate, thus maintaining the geometric configuration
of the spacecraft when
executing manoeuvres. This synchronization algorithm will
require multiple spacecraft
within the formation to communicate their states to each other,
and so communication
loss will have an even stronger effect on the system when
synchronization is incorporated
into the controller.
The objective of this paper is to propose a technique to
managing MSFF systems
during periods of communication loss, and to investigate the
effects of this technique on
the stability of the system. The communication loss compensation
technique proposed in
this thesis is to use the previously communicated data in place
of the lost data, and since
this technique is computationally efficient, it becomes
advantageous for small satellites
not equipped with high on-board processing budgets. The
stability of MSFF systems
denotes the ability of the follower spacecraft to converge to
their desired trajectories, thus
maintaining the desired configuration of the formation. Another
objective is to devise
an experimental plan for ground robots in formation, and to
analyze the behaviour of
the system under communication loss both experimentally and
theoretically using the
-
Chapter 1. Introduction 5
proposed communication loss management technique. The reason the
experiments will
be conducted using ground robots is because it is not feasible
to perform experimental
testing on real satellites (due to time constraints, high risk
levels, and high costs) and the
university of Toronto institute for aerospace studies (UTIAS) is
not yet equipped with
hardware suitable for satellite formation testing.
1.3 Overview
The rest of this thesis is presented in 4 more chapters. Chapter
2 presents the system
model and the formulation for the adaptive nonlinear controller
with synchronization,
and simulation results showing the effectiveness of the
controller. Chapter 3 introduces
the proposed communication loss management technique, analyzes
the stability of MSFF
system with an approximation of the adaptive controller, then
contrasts the theoretical
analysis with simulation results. In Chapter 4, MSFF systems
with synchronization
are again analyzed using the proposed communication loss
management technique, and
the theoretical analysis is contrasted with simulation results.
Chapter 5 introduces the
experimental setup for ground robots in formation by presenting
the system model, and
analyzing the system under communication loss to compare the
experimental outcome
with the theoretical results. Finally, concluding remarks and
future work are discussed
in Chapter 6.
-
Chapter 2
System Modelling
2.1 MSFF Position Dynamics Model
In this section, the system model for MSFF will be derived. A
drawing depicting a leader-
follower system is shown in Fig. 2.1 on the following page. The
inertial coordinate system
(X, Y, Z) is attached to the center of the Earth, R(t) is the
position vector from the origin
of the inertial coordinate system to the leader spacecraft, the
relative coordinate system
(xl, yl, zl) is attached to the leader spacecraft such that xl
is in the opposite direction
of the velocity vector, yl is along the direction of R(t), and
zl completes the right hand
coordinate frame, and ρ is the position vector from the origin
of the relative coordinate
system to the follower spacecraft [9].
The nonlinear position dynamics of the leader and follower
spacecraft with respect to
Figure 2.1: Schematic representation of an MSFF system
6
-
Chapter 2. System Modelling 7
the inertial coordinate system (X, Y, Z) are:
ml··R+ml (M +ml)G
(R/ ‖R‖3
)+ Fdl = ul (2.1)
mf
(··R+
··ρ)
+mf (M +mf )GR + ρ
‖R + ρ‖3+ Fdf = uf (2.2)
where ml,mf are the masses of the leader and the follower, Fdl,
Fdf are disturbance force
vectors, ul(t), uf (t) are the control input vectors of the
leader and follower spacecraft, M
= 5.97 × 1024kg is the Earth’s mass, and G = 6.672 ×
10−11m3/kg.s2 is the universalgravity constant. Because M >>
ml,mf , equations (2.1) and (2.2) can be simplified as:
ml··R+mlMG
(R/ ‖R‖3
)+ Fdl = ul (2.3)
mf
(··R+
··ρ)
+mfMGR + ρ
‖R + ρ‖3+ Fdf = uf (2.4)
The dynamic equations describing the position of the follower
spacecraft relative to the
leader spacecraft can then be written as:
mf··ρ+mfMG
(R + ρ
‖R + ρ‖3− R‖R‖3
)+mfml
ul + Fdf −mfml
Fdl = uf (2.5)
Equation (2.5) can be written in terms of the relative
coordinate system (xl, yl, zl) as:
mf··q+C (ω)
·q+N (q, ω,R, ul) + Fd = uf (2.6)
where q(t) = [x, y, z]T is the relative position vector, and
C(ω) is the Coriolis-like matrix
given by:
C (ω)∆= 2mfω
0 −1 01 0 0
0 0 0
(2.7)N(.) is a nonlinear term defined as:
N(q, ω,
·ω,R, ul
)∆=
mfMG
x‖R+q‖3 −mfω
2x+mf·ω y +
mfmlulx
mfMG(y+‖R‖‖R+q‖3 −
1‖R‖2
)−mfω2y −mf
·ω x+
mfmluly
mfMGz
‖R+q‖3 +mfmlulz
(2.8)
where Fd is the total, constant disturbance force defined
as:
Fd∆=Fdf − (mf/ml)Fdl (2.9)
-
Chapter 2. System Modelling 8
The left hand side of equation (2.6) can be linearly
parameterized as:
mf··q+C (ω)
·q+N
(q, ω,
·ω,R, ul
)+ Fd
= W(ξ,·q, q, ω,
·ω,R, ul
)θ
(2.10)
where W(.) is the regression matrix composed of known functions,
ξ is a dummy variable,
and θ is the system’s constant parameter vector. W(.) and θ are
defined as:
W(ξ,·q, q, ω,
·ω,R, ul
)∆=
ξx − 2ω
.y−ω2x− .ω y + x‖R+q‖3 ulx 1 0 0
ξy − 2ω.x−ω2y + .ω x+
(y+‖R‖‖R+q‖3 −
1‖R‖2
)uly 0 1 0
ξz +z
‖R+q‖3 ulz 0 0 1
(2.11)
and
θ∆=[mf mf/ml Fdx Fdy Fdz
]T(2.12)
2.2 Adaptive Control Formulation
The main objective now is to design a control input uf so that q
(t)→ qd (t) as t→∞ .The performance of the control can be evaluated
using the tracking error e(t) defined as:
e (t)∆= qd (t)− q (t) (2.13)
To account for the system parametric uncertainty due to
instrumentation error, an adap-
tation law for on-line estimation of the unknown parameters is
introduced and defined
as:≈θ (t)
∆= θ −
∧θ (t) (2.14)
where≈θ denotes the parameter estimation error vector and
∧θ is the dynamic estimate
of θ . The filtered tracking error r(t) is defined as:
r (t)∆=·e (t) + Λe (t) (2.15)
where Λ is a constant, diagonal, positive-definite, control gain
matrix. Differentiating
equation (2.15) and multiplying it by mf gives:
mf·r = mf
( ··qd +Λ
·e)−mf
··q (2.16)
Equation (2.16) can be substituted into equation (2.6) to
get:
mf·r = mf
( ··qd +Λ
·e)+C (ω)
·q+N (q, ω,R, ul)+Fd−uf = W
(··qd +Λ
·e,·q, q, ω,R, ul
)θ−uf
(2.17)
-
Chapter 2. System Modelling 9
The control input uf (t) is then designed as:
uf = W (.)∧θ+Mr (2.18)
where M is a constant, diagonal, positive definite matrix.∧θ (t)
is updated using the
algorithm:·∧θ = ΓW
T (.) r (2.19)
where Γ is a constant, diagonal, positive-definite, adaptation
gain matrix. Equations
(2.15), (2.18) and (2.19) constitute the proposed adaptive
nonlinear controller. Finally,
by differentiating equation (2.14) with respect to time, the
parameter
·≈θ can be described
as:·≈θ = −ΓW T (.) r (2.20)
By defining the Lyapunov function:
V (r,≈θ)
∆=
1
2rTmfr +
1
2
≈θT Γ−1
≈θ (2.21)
it is shown in [9] that global asymptotic convergence of the
position and velocity tracking
errors can be accomplished by the adaptive controller. In
addition, global asymptotic
convergence is also proved for an adaptive controller through
the passivity theorem [8]
[27].
2.3 Integrating Synchronization Control
As mentioned in Chapter 1, for the formation to keep its
configuration during a manoeu-
vre, synchronization is needed between the spacecraft in an MSFF
system. To integrate
the synchronization technique into the proposed MSFF adaptive
controller, first the syn-
chronization error, which identifies the performance of the
synchronization controller, is
defined as [20]:
Ξ (t) = Te (t) (2.22)
where T is a generalized synchronization transformation matrix.
Ξ (t) represents how
one trajectory converges with respect to another. It is now
important to introduce the
concepts of internal and external synchronization. If for a
specific follower spacecraft e(t)
in Eq. 2.22 is defined as:
-
Chapter 2. System Modelling 10
e(t) =
ex
ey
ez
then the synchronization error is titled internal
synchronization, since the synchronization
occurs between the axes of the same follower spacecraft. If
however for a system with n
follower spacecraft e(t) is defined as:
e(t) =
e1x
e2x...
enx
then the synchronization is between the same axis of different
follower spacecraft, and
this is termed external synchronization (in the example above,
the synchronization is
for the x-axis, and can similarly be written for the y and z
axes). The objective then
becomes to achieve Ξ (t)→ 0 as t→∞ . The following
synchronization transformationmatrix is chosen:
T =
2 −1 −1−1 2 −1
. . . . . . . . .
−1 2 −1−1 −1 2
(2.23)
This synchronization transformation denotes that each follower
spacecraft will be syn-
chronized with the follower spacecraft before it and the
spacecraft after it. The coupled
position error e* which contains the tracking error e(t) and the
synchronization error
Ξ (t) is defined as:
e∗ (t) = e (t) + BTTt∫
0
Ξdτ (2.24)
where B is a positive coupling gain matrix. The coupled filtered
tracking error r∗(t) is
defined as:
r∗ (t) = ė∗ (t) + Λe∗ (2.25)
where Λ is a constant, diagonal positive-definite, control gain
matrix. The control input
uf is then defined as:
uf (t) = W (.)∧θ (t) + Mr (t) + KsT
TΞ (t) (2.26)
-
Chapter 2. System Modelling 11
where Γ is a constant, diagonal, positive-definite, adaptation
gain matrix. Equations
(3.24, 3.25, 3.26) constitute the adaptive synchronization
controller, and by defining the
following Lyapunov function:
V (r,≈θ,Ξ)
∆=
1
2rTmfr +
1
2
≈θT Γ−1
≈θ +
1
2ΞTKsΞ +
1
2
t∫0
TTΞdτ
T BΛKs t∫
0
TTΞdτ
(2.27)
global asymptotic convergences to zero of e(t) and Ξ (t) are
shown in [20].
2.4 Simulation Results
In this section, the effectiveness of the adaptive nonlinear
synchronization controller is
demonstrated through Matlab simulations of an MSFF system with
one leader and four
follower spacecraft. The goal of the controller is to drive the
follower spacecraft from
their initial position to a desired final positions relative to
the leader spacecraft, and to
maintain the desired final position. Having a follower
spacecraft maintain a constant
position relative to the leader spacecraft is termed an
along-track trajectory, and in the
simulations presented here all four follower spacecraft have an
along-track trajectory.
The parameters of the MSFF system, adopted from [20] are as
follows:
mf1 = mf2 = mf3 = mf4 = 410kg M = diag[0.13, 0.12, 0.09]
µ = 9.86× 1014m3/s2 Λ = diag[0.04, 0.04, 0.04]R = [0, 4.224×
107, 0]Tm B = diag[0.0008, 0.0008, 0.0008]
ml = 1550kg Fd = [−1.025, 6.248,−2.415]T × 10−5Nqd1 = [100, 0,
0]
Tm q1(0) = [150, 10, 20]Tm
qd2 = [0,−100, 0]Tm q2(0) = [−10,−130,−20]Tmqd3 = [−100, 0, 0]Tm
q3(0) = [−140, 10,−20]Tmqd4 = [0, 100, 0]
Tm q4(0) = [30, 160, 20]Tm
q̇d1 = q̇d2 = q̇d3 = q̇d4 = [0, 0, 0]Tm/s q̇1(0) = q̇2(0) =
q̇3(0) = q̇4(0) = [0, 0, 0]
Tm/s
ω = 7.272× 10−5rad/s
2.4.1 Adaptive Control Simulation
The first simulation is run for the MSFF described above using
only an adaptive con-
troller, without synchronization. The following figures outline
the results of the simula-
tions:
-
Chapter 2. System Modelling 12
Figure 2.2: Relative Trajectory of the MSFF System with No
Synchronization
Figure 2.3: Tracking Error of Follower
Spacecraft 1 with No Synchronization
Figure 2.4: Internal Sync Error of Follower
Spacecraft 1 with No Synchronization
Figure 2.5: Tracking Error of Follower
Spacecraft 2 with No Synchronization
Figure 2.6: Internal Sync Error of Follower
Spacecraft 2 with No Synchronization
-
Chapter 2. System Modelling 13
Figure 2.7: Tracking Error of Follower
Spacecraft 3 with No Synchronization
Figure 2.8: Internal Sync Error of Follower
Spacecraft 3 with No Synchronization
Figure 2.9: Tracking Error of Follower
Spacecraft 4 with No Synchronization
Figure 2.10: Internal Sync Error of Fol-
lower Spacecraft 4 with No Synchroniza-
tion
Figure 2.11: External Sync Error of the x
Axis with No Synchronization
Figure 2.12: External Sync Error of the y
Axis with No Synchronization
-
Chapter 2. System Modelling 14
Figure 2.13: External Sync Error of the z Axis with No
Synchronization
As can be observed from the simulation results, the adaptive
nonlinear controller
succeeds in forcing the tracking errors of the follower
spacecraft to converge to zero. The
internal synchronization for the follower spacecraft and
external synchronization of the
three axes are also presented to verify the effectiveness of the
synchronization algorithm
after it gets incorporated into the simulation.
2.4.2 Adaptive Synchronization Control Simulation
The same MSFF system is simulated once again using the full
adaptive nonlinear syn-
chronization controller, and the results are as follows:
Figure 2.14: Relative Trajectory of the MSFF System with
Synchronization
-
Chapter 2. System Modelling 15
Figure 2.15: Tracking Error of Follower
Spacecraft 1 with Synchronization
Figure 2.16: Internal Sync Error of Fol-
lower Spacecraft 1 with Synchronization
Figure 2.17: Tracking Error of Follower
Spacecraft 2 with Synchronization
Figure 2.18: Internal Sync Error of Fol-
lower Spacecraft 2 with Synchronization
Figure 2.19: Tracking Error of Follower
Spacecraft 3 with Synchronization
Figure 2.20: Internal Sync Error of Fol-
lower Spacecraft 3 with Synchronization
-
Chapter 2. System Modelling 16
Figure 2.21: Tracking Error of Follower
Spacecraft 4 with Synchronization
Figure 2.22: Internal Sync Error of Fol-
lower Spacecraft 4 with Synchronization
Figure 2.23: External Sync Error of the x
Axis with Synchronization
Figure 2.24: External Sync Error of the y
Axis with Synchronization
Figure 2.25: External Sync Error of the z Axis with
Synchronization
-
Chapter 2. System Modelling 17
As can be observed from the simulations results, the adaptive
synchronization control
law is able to force the internal and external synchronization
errors to converge to zero
faster than the adaptive controller alone, and the tracking
errors of the four spacecraft
still converge to zero. Since the external synchronization error
converge to zero at a faster
rate, this means the MSFF geometric configuration is maintained
for a longer time, a
feature desirable in MSFF systems.
This chapter defines an MSFF system model, and presents the
formulation for an
adaptive nonlinear controller and an adaptive nonlinear
synchronization control which
guarantees convergence in the presence of constant, slow
varying, or unknown parameters.
Simulation results show the adaptive controller’s ability to
force the tracking errors of
the follower spacecraft to converge to zero, and the adaptive
synchronization controller’s
ability to decrease the synchronization errors and force them to
converge faster than the
adaptive control alone. Chapter 3 will introduce ADS and show
how they can be used
to model communication loss within a networked system.
-
Chapter 3
Communication Loss Stability
Analysis
In this chapter, the foundation for the theoretical analysis of
communication loss between
spacecraft will be presented. In particular, the behaviour and
stability of a follower space-
craft will be analyzed when communication is lost between it and
the leader spacecraft.
It is assumed that the leader spacecraft communicates its states
(position and velocity
data) to the follower spacecraft within the MSFF system, and the
follower spacecraft
then calculate their relative states and determine the
appropriate control forces; thus,
when communication is lost between a follower and the leader,
the follower will not be
aware of its relative states and will not be able to calculate
the required control forces. It
is also assumed that the communication loss is stochastic in
nature (since external events
such as solar winds and flares could occur at any time), and
thus the amount of time
where communication is lost within the mission life is presented
as a probability value.
To manage periods of communication loss, it is proposed that the
follower spacecraft
use the last communicated data in lieu of the lost data to
calculate the required con-
trol forces. The main advantage of this technique is that it is
computationally simple,
thus becoming a very attractive method for small spacecraft with
limited computational
resources. In the literature studied, authors have used various
methods for modelling
communication loss within a system, including Bernoulli
processes [45], Markov chain
models [25] [43], and asynchronous dynamical systems (ADS) [1]
[2]. In the following
section, an MSFF system with communication loss will be modelled
as an ADS, and the
stability of the system when the proposed communication loss
management technique is
employed will be examined and analyzed.
18
-
Chapter 3. Communication Loss Stability Analysis 19
3.1 Asynchronous Dynamical Systems
ADS are systems that incorporate discrete and continuous
dynamics. The discrete dy-
namics are driven by external events with fixed rates, and the
continuous dynamics are
governed by differential equations (or difference equations for
the discretized form) [1]
[2]. ADS with rate constraints on events can be used to model
communication loss in a
networked control system, and stability criteria of such systems
have been studied and
presented in literature [1] [2] [60]. The following figure,
which has been studied in [60],
illustrates a networked control system with data packet
loss:
Figure 3.1: ADS Communication Loss Model
This example shall be used to model an MSFF system with
communication loss. As
can be seen in Fig. 3.1, communication loss is modeled as a
switch that closes randomly
with a certain probability, represented by the constant term as
rate r; it is important
to clarify that the rate r does not indicate the frequency at
which the switch closes,
but rather the probability that at any given time, the switch
will be closed. The closed
position of the switch (position S1) represents the dynamics of
the system when the
communication exists, thus the states of the leader spacecraft
are transmitted. The
open position of the switch (position S2) represents the
dynamics of the system when
transmission is lost between the spacecraft. Since the
compensation technique being
proposed during periods of communication loss is to use the last
communicated data,
then the dynamics of the switch can be modeled as follows:
S1 → x(kh) = x(kh)S2 → x(kh) = x((k − 1)h)
(3.1)
-
Chapter 3. Communication Loss Stability Analysis 20
3.1.1 Stability Criteria
The main attraction of using ADS to model communication loss is
that the stability
criteria for ADS have been studied and are available in the
literature. These criteria
and any associated theorems can therefore be used to analyze and
study the stability of
MSFF systems with communication loss to understand the behaviour
of those systems
when data transfer between the spacecraft is not always
available. This subsection will
highlight these main criteria, which are fully presented in [1]
[2] [60].
Given the ADS in Fig. 3.1 where the equations of dynamics for
the continuous states
are linear and time invariant, if there exists a Lyapunov
function V(x) and scalars a1, a2
such that:
ar1a1−r2 > a > 1 (3.2)
where r is the switching rate as mentioned previously and
V (x(k + 1))− V (x(k)) ≤ (a−2s − 1)V (x(k)), s = 1, 2 (3.3)
then the ADS is exponentially stable, with decay rate greater
than a. This criteria
guarantees the ADS to be stable on the whole. If for s the
discrete state dynamics can
be represented in the following form:
x((k + 1)h) = Φsx(kh) (3.4)
then the scalars a1, a2 and a Lyapunov function in the following
form:
V (x(kh)) = xT (kh)Px(kh) (3.5)
where P is a positive-definite symmetric matrix can be
formulated as a bilinear matrix
inequality (BMI) problem by rewriting Eq. 3.2 and Eq. 3.3 as
r log a1 + (1− r) log a2 > 0 (3.6)
and
ΦTs PΦs ≤ a−2s P (3.7)
where a BMI is defined as a function of the form
F (x, y) = F0 +m∑i=1
xiFi+n∑j=1
yjGj+m∑i=1
n∑j=1
xiyjHij > 0
-
Chapter 3. Communication Loss Stability Analysis 21
where Gj and Hij are symmetric matrices with the same dimensions
as Fi, x ∈
-
Chapter 3. Communication Loss Stability Analysis 22
Theorem: If the closed-loop system (Φ − ΓK) of the system in
Fig. 3.1 with nocommunication loss is stable, then:
- If the open-loop system(Φ) is marginally stable, then the
system is exponentially
stable for all 0 < r ≤ 1- If the open-loop system is
unstable, then the system is exponentially stable for all
1
1− γ1/γ2< r ≤ 1 (3.13)
where
γ1 = log [λ2max(Φ− ΓK)]
γ2 = log [λ2max(Φ)]
(3.14)
As already stated, the conditions of the theorem are necessary
for the stability of an
ADS, and for all r values that don’t fall within the specified
range, the system is unstable.
Using the communication loss management technique proposed, this
theorem allows for
calculating the maximum rate of communication loss that the
system can tolerate. If
the expected rate of communication loss for a particular MSFF
system is less than the
rate determined by the theorem, then the system is guaranteed to
maintain its desired
formation, otherwise the system will become unstable and the
follower spacecraft would
not converge to their desired trajectories. The setback of using
this theorem is that
it is developed for linear discrete systems, and assumes a
simple gain feedback control
law. Since this study is meant as a preliminary analysis of MSFF
stability under the
effects of communication loss, the equations of dynamics for
MSFF systems presented
in Chapter 2 shall be linearized and discretized, and the
adaptive control law shall be
approximated as a constant gain control law in order to use the
ADS analysis theorem.
Future studies shall relax these constraints to fully analyze
the nonlinear system with
the adaptive control law.
Linearization can be performed on the nonlinear dynamics of MSFF
systems using
the first-order terms of their Taylor series expansion. The
following represents the general
equation for linearizing a multivariable function:
f(X) ≈ f(X0) +∇f |p · (X −X0) (3.15)
where X is the states vector and X0 is the linearization point
of interest. For the MSFF
nonlinear equations presented in Chapter 2, assuming ul = 0, the
linearized equations of
-
Chapter 3. Communication Loss Stability Analysis 23
dynamics for MSFF systems around the point (0,0,0) are as
follows:
mf q̈ + Cl(.)q̇ +Nl(.)q + Fd = uf (3.16)
where
Nl(.) =
mf
(µR3− ω2
)mf ω̇ 0
−mf ω̇ −mf(
2µR3
+ ω2)
0
0 0mfµ
R3
(3.17)
Cl(.) =
0 2mfω 0
−2mfω 0 00 0 0
(3.18)The augmented states vector can be represented as:
ql =
qq̇
(3.19)then:
q̇l =
[03×3] [I3×3][−Nl(.)/mf] [−Cl(.)/mf] ql +
[03×3][I3×3/mf
]u−
Fdx
Fdy
Fdz
(3.20)
The adaptive control law from Chapter 2 is defined as:
uf = Wl(.)θ̂ + Mr (3.21)
where the regressor matrix for the linearized equations is:
W (.) =
−Λẋ+ 2ωẏ +
(µR3− ω2
)x+ ω̇y 1 0 0
−Λẏ − 2ωẋ−(
2µR3− ω2
)y − ω̇x 0 1 0
−Λż +(µR3
)z 0 0 1
(3.22)
and∧θ is the estimate for the linear system’s constant parameter
vector θ defined as:
θ∆=[mf Fdx Fdy Fdz
]T(3.23)
-
Chapter 3. Communication Loss Stability Analysis 24
3.1.2 Controller Estimate
Since the stability criteria and the theorem stated above are
based upon the assumption
that the control law for the plant is in the form −K −x(kh), the
objective is now tosimplify the adaptive control law as a constant
feedback control law to properly apply
the theorem and analyze the effects of communication loss on an
MSFF system. To do
so, the MSFF system is first assumed to be in a circular orbit (
therefore·ω = 0), and
then the parameter variable∧θ is assumed to be a constant (the
validity of this assumption
stems from the definition of the parameter variable, which is
required to be a constant
or slow-varying). To properly select the values for the
slow-varying parameters, possible
values for those parameters would be tested to determine the
value that results in the
largest r when the theorem in section 3.1.1 is used; this
constitutes the worst-case value
of that parameter, and for other values the system would still
be stable. For instance,
when the mass of the spacecraft is tested, it is determined that
the worst-case value for
that parameter is its minimum value, or the value of the
spacecraft with no fuel, and so
once the stability criteria are derived for that worst-case
value, the system will still be
stable for other values of the parameter. While this gives the
system robust attributes,
it is important to point out that by keeping the∧θ value a
constant, the controller ceases
to be adaptive. The control law can now be approximated as:
uf = Wl(.)θ + Mr = −Klql (3.24)
where Kl is a constant matrix for the linearized system. The
control law is now a
constant feedback law, where the gain K is an approximation to
the system’s behaviour
under adaptive control law, and the theorem can be used.
3.1.3 Discretization
The final step that needs to be taken before the theorem could
be applied is to discretize
the MSFF system. While discretizing the state space model is a
straight forward task, the
discrete equivalent of the K matrix is also required before the
theorem could be applied;
this is because the continuous equivalent of the K matrix in a
feedback control law will
not necessarily be the same in the discrete form, as is observed
in the difference between
the continuous-time and discrete-time Riccati equations used to
calculate the feedback
control law for linear-quadratic-regulator (LQR) and
linear-quadratic-Gaussian (LQG)
-
Chapter 3. Communication Loss Stability Analysis 25
controllers [49]. In this subsection, the bilinear equivalence
of the linearized MSFF state
space model will be derived, along with the discrete form of the
constant Kl matrix.
The full derivation of the bilinear equivalence of a continuous
system is available in
[16] and will be highlighted here. Starting from the continuous
state space equation in
the following form:
ẋ = Ax+Be
u = Cx+De(3.25)
the Laplace transform is taken:
sX = AX +BE
U = CX +DE
and s can be substituted for by the z-transform equivalent of
the trapezoid/bilinear rule:
2(z − 1)h(z + 1)
X = AX +BE (3.26)
U = CX +DE (3.27)
The equations are then transformed back to the time domain:
x(k + 1)− x(k) = Ah2
[x(k + 1) + x(k)] +Bh
2[e(k + 1) + e(k)] (3.28)
The k + 1 terms are now grouped together and defined as w(k +
1):
x(k+ 1)− Ah2x(k+ 1)− Bh
2e(k+ 1) = x(k) +
Ah
2x(k) +
Bh
2e(k)
∆=√hw(k+ 1) (3.29)
where the√h is a factor introduced to balance the gain of the
discrete equivalence
between the input and output [16]. Writing w at time k and
solving for x yields:
x(k) =
(I − Ah
2
)−1√hw(k) +
(I − Ah
2
)−1Bh
2e(k) (3.30)
Eq. (3.30) is now substituted into Eq. (3.29):
w(k + 1) =
(I +
Ah
2
)(I − Ah
2
)−1w(k) +
(I − Ah
2
)−1B√he(k) (3.31)
Finally, the output equation for the bilinear equivalence can be
obtained by substituting
Eq. (3.30) into Eq. (3.27):
u(k) =√hC
(I − Ah
2
)−1w(k) +
D + C(I − Ah
2
)−1Bh
2
e(k) (3.32)
-
Chapter 3. Communication Loss Stability Analysis 26
Since the state space representation of the linearized MSFF
system is in the following
form:
ẋ = Ax+Bu (3.33)
then the bilinear equivalence is in the following form:
w(k + 1) = Φw(k) + Γu(k) (3.34)
whereΦ = (I + Ah
2)(I − Ah
2)−1
Γ = (I − Ah2
)−1B√h
(3.35)
Since the output control law is of the form
u = −Kx (3.36)
then to get the discrete equivalence of the K matrix, x(k) is
substituted with the w(k)
equivalence:
u(k) = −Kx(k) = −K(I + AT2
)−1[w(k)− BT2u(k)] (3.37)
u(k) = −[I −K(I + AT2
)−1BT
2]−1K(I +
AT
2)−1w(k) (3.38)
Therefore the discrete form of the constant K matrix, or Kd is
in the following form:
[I −K(I + AT2
)−1BT
2]−1K(I +
AT
2)−1 (3.39)
Now the MSFF system can be fully discretized, and the theorem
for determining the
maximum rate of communication loss a system can tolerate can be
applied.
3.2 Numerical Example
A one-leader-one-follower example using one of the 4 follower
spacecraft used in the
simulations from Chapter 2 is analyzed in this section. This
example will be simplified and
discretized using the techniques discussed in this chapter, and
the theorem for analyzing
the stability of the system in the presence of communication
loss will be applied. In
addition, a simulation of the linearized system and the
nonlinear system in the presence
of communication loss will be shown to verify the results
obtained from the theoretical
analysis.
-
Chapter 3. Communication Loss Stability Analysis 27
3.2.1 Theoretical Analysis
The parameters of the system that will be analyzed are as
follows:
mf = 410kg q(0) = [150, 10, 20]Tm
µ = 9.86× 1014m3/s2 q̇d = [0, 0, 0]Tm/sR = [0, 4.224× 107, 0]Tm
q̇(0) = [0, 0, 0]Tm/sω = 7.272× 10−5rad/s M = diag[0.13, 0.12,
0.09]qd = [100, 0, 0]
Tm Λ = diag[0.04, 0.04, 0.04]
Fd = [−1.025, 6.248,−2.415]T × 10−5N
Using a 1 second sampling period, this system is discretized as
outlined in subsection
3.1.3, and when the stability criteria theorem is applied to
this example, r is calculated to
be 0.01; since r represents the rate at which the switch in Fig.
3.1 closes, this means that
the MSFF system will remain stable using the proposed
communication loss management
technique for communication loss rates that are less than 99%.
The effective sampling
period for this example is then calculated to be just under 100
seconds, or in other words
if the sampling period is chosen to be for example 99 seconds,
and the system experiences
no communication loss, it will remain stable and the follower
spacecraft will converge to
its desired states. This suggests that another use for the
theorem is to determine the
appropriate sampling period for the system (i.e. how often the
leader spacecraft needs to
transmit its states to the follower spacecraft) especially if
the expected communication
loss rate is known ahead of time.
3.2.2 Linear System Simulation
As a validation to the results produced by the theorem, the
linearized system with the
constant feedback control law was simulated using the proposed
communication loss
management technique, and the figures below show the simulation
results for 98% and
99% communication loss rates:
As can be observed from Fig. 3.2-3.5, the linearized MSFF system
maintains stability
when the communication loss rate is 98% and the follower
spacecraft is able to track its
desired trajectory, but the system is unstable for the case when
the rate is 99%. This
complies with the results obtained from the theoretical analysis
of the system.
-
Chapter 3. Communication Loss Stability Analysis 28
Figure 3.2: Relative Trajectory of the Lin-
earized System with 98% Communication
Loss
Figure 3.3: Tracking Error of the Lin-
earized System with 98% Communication
Loss
Figure 3.4: Relative Trajectory of the Lin-
earized System with 99% Communication
Loss
Figure 3.5: Tracking Error of the Lin-
earized System with 99% Communication
Loss
-
Chapter 3. Communication Loss Stability Analysis 29
3.2.3 Nonlinear System Simulation
The theoretical analysis in this section is conducted for the
linearized MSFF system, and
uses an approximation of the adaptive controller. It remains to
be seen how close the
analysis on the linearized system matches the behaviour of the
nonlinear system. To do
so, the full nonlinear system with adaptive control will be
simulated as in Chapter 2,
with communication loss introduced to the system, and using the
proposed management
technique. The theoretical analysis of the linearized system
indicate that the system
will remain stable if the communication loss rate is less than
99%, and below are the
simulation results for the MSFF system under 98% and 99%
communication loss rates:
Figure 3.6: Relative Trajectory of the Non-
linear System with 98% Communication
Loss
Figure 3.7: Tracking Error of the Nonlinear
System with 98% Communication Loss
Figure 3.8: Relative Trajectory of the Non-
linear System with 99% Communication
Loss
Figure 3.9: Tracking Error of the Nonlinear
System with 99% Communication Loss
As can be observed from the plots in Fig. 3.6-3.9, the nonlinear
MSFF system main-
-
Chapter 3. Communication Loss Stability Analysis 30
tains stability when the communication loss rate is 98% and the
follower is again able to
track its desired trajectory, but the system is unstable for the
case when the rate is 99%.
This complies with the theoretical analysis conductd on the
linearized system, and sug-
gests that the analysis of the simplified system is a good
approximation of the behaviour
of the nonlinear system with the adaptive control law in the
presence of communication
loss.
This chapter outlines the communication loss management
technique proposed by this
thesis, which is to use the latest communicated data in lieu of
the lost data. The chapter
also introduces ADS, and demonstrates how the can be used to
model communication
loss within MSFF systems. In order to use the ADS stability
criteria to analyze the
effect of communication loss on MSFF systems, the system model
is linearized and the
controller is simplified as a constant feedback control law.
Simulation results of the
linearized and nonlinear systems are also presented and
contrasted with the theoretical
analysis. Chapter 4 will perform similar theoretical analysis on
MSFF systems with
synchronization control between the follower spacecraft.
-
Chapter 4
ADS Stability Analysis with
Synchronization Control
In Chapter 3, a theoretical approach to analyzing the behaviour
of an MSFF system
when communication is lost between the leader and a follower is
presented. However, to
incorporate synchronization control into MSFF systems, follower
spacecraft are required
to communication with other follower spacecraft in addition to
the leader. This chapter
will extend the theoretical analysis from the previous chapter
to account for the case when
synchronization is integrated into the controller. Using the
linearization and controller
simplification techniques from Chapter 3, the system model for
an MSFF system with
synchronization control will be derived, and a numerical example
for the theoretical
analysis will be provided. The theoretical analysis of the
numerical example will then be
contrasted with simulation results of the linearized and the
nonlinear MSFF system as
was done in the previous chapter.
4.1 ADS Model with Synchronization
Similar to the approach taken in Chapter 3, the main goal again
becomes to represent the
MSFF system with synchronization control as a linear time
invariant system of equations
in the following form:
ẋ(t) = Ax(t) +Bu(t) (4.1)
where the control force u can be approximated as a constant gain
feedback law in the
31
-
Chapter 4. ADS Stability Analysis with Synchronization Control
32
following form:
u(t) = −Kx(t) (4.2)
The linearized system model for a 1-leader 1-follower system
shown in Eq. 3.20 can
be extended for n-follower spacecraft systems as shown
below:
q̇l =
[03×3] [I3×3][03×(6×(n−1))
][−Nl(.)/mf] [−Cl(.)/mf] [03×(6×(n−1))]
[03×3][03×(6×1)
][I3×3]
[03×(6×(n−2))
][03×(6×1)
] [−Nl(.)/mf] [−Cl(.)/mf] [03×(6×(n−2))]. . . . . . . . . . .
.
[03×3][03×(6×(n−1))
][I3×3][
03×(6×(n−1))] [−Nl(.)/mf] [−Cl(.)/mf]
ql+
[03×(3×n)
][I3×3/mf
] [03×(3×(n−1))
][03×(3×n)
][03×3]
[I3×3/mf
] [03×(3×(n−2))
]. . . . . . . . .[
03×(3×n)]
[03×(3×(n−1))
] [I3×3/mf
]
u− (4.3)
[ [Fdx1 Fdy1 Fdz1
] [Fdx2 Fdy2 Fdz2
] . . . [ Fdxn Fdyn Fdzn ]]T
where
q =[x1 y1 z1 ẋ1 ẏ1 ż1 x2 y2 z2 ẋ2 ẏ2 ż2 · · · xn yn zn ẋn
ẏn żn
]T(4.4)
and
u =[ufx1 ufy1 ufz1 ufx2 ufy2 ufz2 · · · ufxn ufyn ufzn
]T(4.5)
The control input (Eq. 2.26) can be simplified in the same
manner highlighted in the
previous chapter to be approximated as a feedback control law in
the form of Eq. 4.2,
for which K will be defined as:
-
Chapter 4. ADS Stability Analysis with Synchronization Control
33
K =
V1 0 0 V2 V3 0 V4 0 0 V5 0 0 V0 V6 0 0 V7 0 0
0 V8 0 V9 V10 0 0 V11 0 0 V12 0 V0 0 V13 0 0 V14 0
0 0 V15 0 0 V16 0 0 V17 0 0 V18 V0 0 0 V19 0 0 V20
V6 0 0 V7 0 0 V1 0 0 V2 V3 0 V4 0 0 V5 0 0 V0
0 V13 0 0 V14 0 0 V8 0 V9 V10 0 0 V11 0 0 V12 0 V0
0 0 V19 0 0 V20 0 0 V15 0 0 V16 0 0 V17 0 0 V18 V0... ... ...
... ... ... ... ... ... ... ... ... ... ... ... ... ...
V4 0 0 V5 0 0 V0 V6 0 0 V7 0 0 V1 0 0 V2 V3 0
0 V11 0 0 V12 0 V0 0 V13 0 0 V14 0 0 V8 0 V9 V10 0
0 0 V17 0 0 V18 V0 0 0 V19 0 0 V20 0 0 V15 0 0 V16
(4.6)
for the synchronization transformation matrix
T =
2 −1 0 0 −1−1 2 −1 0 0
. . . . . . . . .
0 0 −1 2 −1−1 0 0 −1 2
(4.7)
where:
V0 = 01×(6×(n−3))
V1 =(µR3− ω2
)mf +M1(2B − 1)Λ
V2 = M1(2B − 1)− ΛmfV3 = 2ωmf
V4 = V6 = −BΛM1V5 = V7 = −BM1
-
Chapter 4. ADS Stability Analysis with Synchronization Control
34
V8 = M2(2B − 1)Λ−(
2µR3
+ ω2)mf
V9 = −2ωmfV10 = M2(2B − 1)− ΛmfV11 = V13 = −BΛM2V12 = V14 =
−BM2V15 = M3(2B − 1)Λ + µR3mfV16 = M3(2B − 1)− ΛmfV17 = V19 =
−BΛM3V18 = V20 = −BM3
The MSFF system with synchronization incorporated into the
controller is now in
the form of Eq. 4.1 and 4.2, and the stability criteria for ADS
can be applied; more
specifically, the theorem for determining that maximum rate of
communication loss the
system can withstand can now be applied to analyze the behaviour
of the MSFF system
when communication is lost between the leader and follower
spacecraft, as well as between
follower spacecraft and other follower spacecraft.
4.2 Numerical Example
The five-satellite system from Chapter 2 will be used to conduct
a stability analysis in
the presence of communication loss. First, the theoretical
analysis will be conducted
using the ADS stability criteria for the linearized model with
the simplified controller
that includes synchronization between the follower spacecraft,
and then simulation re-
sults for the linearized system will be contrasted with the
theoretical analysis. Finally,
the nonlinear system with the adaptive synchronization control
from Chapter 2 will be
simulated again, this time with communication loss, to determine
how the behaviour of
the system compares with the theoretical analysis conducted on
the linearized system.
4.2.1 Theoretical Analysis
The parameters of the 1-leader four-follower system are shown
below:
-
Chapter 4. ADS Stability Analysis with Synchronization Control
35
mf1 = mf2 = mf3 = mf4 = 410kg M = diag[0.13, 0.12, 0.09]
µ = 9.86× 1014m3/s2 Λ = diag[0.04, 0.04, 0.04]R = [0, 4.224×
107, 0]Tm B = diag[0.0008, 0.0008, 0.0008]ω = 7.272× 10−5rad/s Fd =
[−1.025, 6.248,−2.415]T × 10−5Nqd1 = [100, 0, 0]
Tm q1(0) = [150, 10, 20]Tm
qd2 = [0,−100, 0]Tm q2(0) = [−10,−130,−20]Tmqd3 = [−100, 0, 0]Tm
q3(0) = [−140, 10,−20]Tmqd4 = [0, 100, 0]
Tm q4(0) = [30, 160, 20]Tm
q̇d1 = q̇d2 = q̇d3 = q̇d4 = [0, 0, 0]Tm/s q̇1(0) = q̇2(0) =
q̇3(0) = q̇4(0) = [0, 0, 0]
Tm/s
After discretizing Eq. 4.3 and Eq. 4.6 using the bilinear
equivalence with a 1 second
sampling period and applying the ADS stability theorem, r is
calculated to be 0.1, which
means that the system shall remain stable for communication loss
rates less than 90%.
Since all the follower spacecraft in this example are similar to
the one studied in section 3.2
(with the exception of the initial position and the final
desired positions of the follower
spacecraft) it can be observed that addting synchronization
attributes to the MSFF
system increased its sensitivity to communication loss. With the
system in which each
follower is only required to communicate with the leader, the
theoretical analysis shows
the system to be capable of maintaining stability for
communication loss rates lower
than 99%, but with the system where the follower spacecraft also
communicate with
other follower spacecraft, the MSFF system is now only capable
of maintaining stability
for communication loss rates lower than 90%. This means that
missions which would
potentially use synchronization techniques need to assess the
probability of packet loss
occuring between the spacecraft and determine whether the system
would remain stable
under the worst-case conditions before integrating
synchronization into the controller.
4.2.2 Linear System Simulation
To validate the theoretical results, the linearized system was
simulated using the pro-
posed communication loss management technique. The figures below
show the simulation
results for 89% and 90% communication loss rates:
As can be observed, the simulation results match the results
obtained from the theo-
retical analysis. The MSFF system maintains stability using the
proposed management
technique when the communication loss rate is 89%, but becomes
unstable when the
-
Chapter 4. ADS Stability Analysis with Synchronization Control
36
Figure 4.1: Relative Trajectory of the Lin-
earized System with 89% Communication
Loss
Figure 4.2: Tracking Error of Follower
Spacecraft 1 with 89% Communication
Loss
Figure 4.3: Tracking Error of Follower
Spacecraft 2 with 89% Communication
Loss
Figure 4.4: Tracking Error of Follower
Spacecraft 3 with 89% Communication
Loss
Figure 4.5: Tracking Error of Follower Spacecraft 4 with 89%
Communication Loss
-
Chapter 4. ADS Stability Analysis with Synchronization Control
37
Figure 4.6: Relative Trajectory of the Lin-
earized System with 90% Communication
Loss
Figure 4.7: Tracking Error of Follower
Spacecraft 1 with 90% Communication
Loss
Figure 4.8: Tracking Error of Follower
Spacecraft 2 with 90% Communication
Loss
Figure 4.9: Tracking Error of Follower
Spacecraft 3 with 90% Communication
Loss
Figure 4.10: Tracking Error of Follower Spacecraft 4 with 90%
Communication Loss
-
Chapter 4. ADS Stability Analysis with Synchronization Control
38
communication loss rate reaches 90%, which complies with the
results from the theorem.
It remains to be seen how close those results are to the
nonlinear system with adaptive
synchronization control.
4.2.3 Nonlinear System Simulation
The following figures show the simulation results of the full
nonlinear MSFF system with
adaptive synchronization control when the system experiences
communication loss. The
proposed communication loss management technique is once again
used, and the results
below are for 94% and 95% communication loss rates (The reason
for including these 2
cases is explained shortly):
Figure 4.11: Relative Trajectory of the Lin-
earized System with 94% Communication
Loss
Figure 4.12: Tracking Error of Follower
Spacecraft 1 with 94% Communication
Loss
Figure 4.13: Tracking Error of Follower
Spacecraft 2 with 94% Communication
Loss
Figure 4.14: Tracking Error of Follower
Spacecraft 3 with 94% Communication
Loss
-
Chapter 4. ADS Stability Analysis with Synchronization Control
39
Figure 4.15: Tracking Error of Follower Spacecraft 4 with 94%
Communication Loss
Figure 4.16: Relative Trajectory of the Lin-
earized System with 95% Communication
Loss
Figure 4.17: Tracking Error of Follower
Spacecraft 1 with 95% Communication
Loss
Figure 4.18: Tracking Error of Follower
Spacecraft 2 with 95% Communication
Loss
Figure 4.19: Tracking Error of Follower
Spacecraft 3 with 95% Communication
Loss
-
Chapter 4. ADS Stability Analysis with Synchronization Control
40
Figure 4.20: Tracking Error of Follower Spacecraft 4 with 95%
Communication Loss
As can be observed from the simulation results, the nonlinear
system behaves better
than the theoretical analysis indicated; while the theorem
applied to the linearized system
indicated the system will maintain stability for communication
loss rates lower than 90%,
the nonlinear system with adaptive synchronization control is
able to maintain stability
for communication loss rates lower than 95%. This shows the
limitations to using the
ADS theorem for analyzing communication loss in MSFF systems in
that the theorem
applies to linear systems, and the analysis on the linearized
MSFF system may not
match the results of the nonlinear system. This suggests than
until a nonlinear approach
to analyzing communication loss in MSFF systems is developed,
nonlinear simulations
should accompany any theoretical analysis conducted on the
linearized system to have
more confidence in the generated results.
Similar to the linearized systems, it can be observed that once
synchronization is
integrated into the adaptive controller, the system’s stability
becomes more sensitive to
communication rate. When the follower spacecraft only
communicated with the leader,
the nonlinear simulations from section 3.2.3 show that the
system remains stable for
communication loss rates lower than 99%, while when the follower
spacecraft are required
to also communicate with other follower spacecraft, the
stability of the system can be
maintained for communication loss rates lower than 95%. This
again suggests that MSFF
systems should only incorporate synchronization if the
communication loss rates expected
to be experienced by the system in space are less than the
limits provided by theoretical
and simulation analysis.
This chapter presents the theoretical analysis on MSFF systems
with synchronization
control that experience communication loss. Similar to the
approach taken in Chapter
-
Chapter 4. ADS Stability Analysis with Synchronization Control
41
3, the MSFF system is linearized, and the controller is
simplified as a constant feedback
control law. Simulation results are presented for the linearized
and the nonlinear system,
and contrasted with the theoretical analysis. Chapter 5 will
present the framework for
performing experimental work using multiple robots in formation
using the proposed
communication loss management technique.
-
Chapter 5
Robots in Formation Analysis and
Experiments
In addition to the theoretical and simulation work conducted for
this research project,
experimental work was also conducted to examine the effects of
the proposed communi-
cation loss management technique in a setup involving mobile
hardware. Due to equip-
ment and budget limitations at the (UTIAS), experimental work
with multiple satellites
was not feasible, and so other alternatives needed to be
considered. It was decided
that experiments involving multiple robots in formation would be
a feasible setup for
demonstrating the effectiveness of the proposed management
technique when the robots
experience communication loss. In this chapter, a breakdown of
the experimental setup
and equipment will provided, along with a system model and a
controller formulation.
Theoretical analysis on the multiple robots in formation (MRF)
will be conducted using
the ADS stability analysis theorem, and the results will be
compared with the outcome
from the experimental work.
5.1 Experimental Setup
5.1.1 UTIAS Resources
Using the facilities available at UTIAS, an experimental setup
was drafted to illustrate the
practical implementation of the proposed management technique
when communication
between the leader and the follower robots is lost. A 10-camera
Vicon motion capture
system (Fig. 5.1) is available and installed in one of the
laboratories with an area of about
42
-
Chapter 5. Robots in Formation Analysis and Experiments 43
Figure 5.1: A Vicon Camera
15m × 8m (Fig. 5.2). The Vicon system is capable of providing
position data of therobots in the x, y, and z direction, as well as
orientation data in terms of roll, pitch, and
yaw angle measurements. Since the robots will be moving in a 2-D
plane, only position
data in the x and y direction will be used, in addition to the
yaw angle measurements.
The data from the Vicon system is provided at 100Hz with
millimeter accuracy. Five
identical robots built from the iRobot Create (two-wheel
differential drive) platform are
also available and can be used to create a one-leader
four-follower formation scenario.
The robots are equipped with markers placed in different
configurations (allowing the
Vicon system to uniquely identify them), and are also equipped
with a mounting platform
containing a laptop computer running Player (Fig. 5.3); Player
is a software that provides
a network between the robots and the Vicon system, allowing for
data acquisition and
logging, and capable of performing motion control. While the
Vicon system is able to
log the position and orientation data of the robots, Player is
able to access this data and
perform any required calculations to determine the linear and
angular velocities to pass
to the robots. The follower robots will require to obtain the
data of the leader robot to
determine their control inputs, and thus loss of the leader’s
information from the Vicon
system constitutes communication loss between the leader and
follower robots.
-
Chapter 5. Robots in Formation Analysis and Experiments 44
Figure 5.2: Lab Area
Figure 5.3: An iRobot Create Roomba with Markers and a
Laptop
-
Chapter 5. Robots in Formation Analysis and Experiments 45
5.1.2 Test Scenario
The goal of creating this scenario is to run a simple MRF
experiment using the pro-
posed communication loss management technique where ADS
theoretical stability anal-
ysis could be performed, and show how well the theoretical
analysis compares with the
experimental results. To simplify the theoretical analysis, the
test scenario will involve
the leader robot only traveling in a straight line, without
rotating or changing directions;
this will eliminate the need to use angular dynamics in the
system model and will facili-
tate the application of the ADS stability analysis theorem.
While it is acknowledged that
this simplification does not result in a comprehensive test
scenario, it is satisfactory since
the objective is to have a proof of concept for the
effectiveness of the proposed packet loss
compensation technique, and illustrate the application of the
ADS theoretical analysis
on a real test case. Details of the experimental configuration
are presented in the final
section of this chapter. The objective of the control law will
be to maintain a desired
relative displacement between the leader robot and the follower
robots as the leader trav-
els forward with a constant linear velocity. Communication loss
can be induced between
the leader and follower robot (as will be demonstrated in
section 5.15) and the proposed
communication loss management technique will be used to
compensate for the lost data.
The results can then be contrasted to the theoretical analysis
of the system to determine
how effective the ADS stability analysis theorem is in
determining the maximum rate of
communication loss the system can withstand before becoming
unstable.
5.2 System Model
Many papers in the literature have studied the modeling of
mobile robots, which include
first order [14] [13] [12] and second order [31] [32] [11]
kinematics modeling. As stated
earlier, only a special case will be considered, where the
leader is traveling in a straight
line at a constant velocity. Thus, the system model can be
simplified as represented as
Newton’s second law [23]:
For a point mass traveling in a straight line, the system can be
modelled as follows:
Fx = mẍ
Fy = mÿ(5.1)
-
Chapter 5. Robots in Formation Analysis and Experiments 46
Figure 5.4: Simplified Robot Model
ẍ = Fxm
ÿ = Fym
(5.2)
which can be expressed in state-space form:
ẋ
ẏ
ẍ
ÿ
=
0 0 1 0
0 0 0 1
0 0 0 0
0 0 0 0
x
y
ẋ
ẏ
+
0 0
0 0
1 0
0 1
uxuy
(5.3)
where
ux =Fxm
uy =Fym
(5.4)
This state-space model will be used for the theoretical analysis
on MRF during periods
of communication loss.
5.3 Controller Formulation
5.3.1 Control Law
Due to the limitations of the equipment and sensors available at
UTIAS (no rate sensors,
no accelerometers), a simple and proved control law is desired
to keep the MRF system
in formation. As such, different papers discussing previous
experiments involving MRF
-
Chapter 5. Robots in Formation Analysis and Experiments 47
were studied, and the control law used in [42] was adopted the
purpose of conducting the
experiments. The control law is detailed as follows:
For an MRF system with i follower robots, let (rxi, ryi), θi,
and (vi, ωi) represent the
Cartesian position, orientation, and linear and angular speed of
the ith robot, then the
kinematic equations for the robots can be written as:
ṙxi = vi cos(θi) (5.5)
ṙyi = vi sin(θi) (5.6)
θ̇i = ωi (5.7)
The kinematic equations are feedback linearized for a fixed
point off the center of the
wheel axis denoted (xi, yi) where:
xi = rxi + d cos(θi) (5.8)
yi = ryi + d cos(θi) (5.9)
Let
viωi
= cos(θi) sin(θi)−1d
sin(θi)1d
cos(θi)
uxiuyi
(5.10)then:
ẋiẏi
= uxiuyi
(5.11)Letting (xdi ,y
di ) denote the desired xi and yi for the follower robots, a
simple control
law in the following form:
uxi = ẋdi − kxi(xi − xdi ) (5.12)
uyi = ẏdi − kyi(yi − ydi ) (5.13)
-
Chapter 5. Robots in Formation Analysis and Experiments 48
where kxi > 0 and kyi > 0 is used to guarantee that xi(t)
→ xdi (t) and yi(t) → ydi (t) ast→∞.
The main advantage of the control law above and the simplified
experimental case is
that only the velocity of the leader and the follower robots are
required for the calcula-
tions, and so the derivative of the data acquired from the Vicon
system need to be taken
once, rather than twice had the control law required
acceleration measurements. This,
however, is still a challenging task; due to the high frequency
at which the data are pro-
vided by the Vicon system (100Hz) and the noise associated with
those measurements,
calculating the slope between two consecutive data points in the
following form
x2 − x1t2 − t1
(5.14)
will not be a practical numerical differentiation method and
will yield unsatisfactory
calculations for the appropriate control forces (This has been
tested and confirmed).
Thus, a proper technique for determining the velocity of the
robots is needed before the
control law can be implemented.
5.3.2 Smooth Differentiation
Pavel Holoborodko is a mathematician who derived his own
noise-robust differentiation
techniques for different cases that are particularly beneficial
for carrying out experiments
with noisy data where differentiation is required [19]. His
differentiation scheme possesses
the following characteristics:
- Precise on low frequencies
- Smooth and guaranteed suppression of high frequencies
- Computationally favorable structure
The full derivation of the algorithm is available in [19]. For a
filter of length N , let
the coefficients be {ck} and the time step be h, then
f ′(x) ≈ 1h
M∑k=1
ck.(fk − f−k) (5.15)
where
ck =1
22m+1
2mm− k + 1
− 2mm− k − 1
(5.16)
-
Chapter 5. Robots in Formation Analysis and Experiments 49
m =N − 3
2(5.17)
M =N − 1
2(5.18)
The number sequence formed by the expression in the square
brackets belongs to a
self-dual family of strip graphs, and can be found in the
Encyclopedia of Integer Sequences
under the identity A050166 [46]. Table 5.1 displays some entries
from A050166 formatted
as a square array, which can be read diagonally.
Table 5.1: A050166 Formatted as a Square Array
1 1 1 1 1 1 1 1 1 1
2 4 6 8 10 12 14 16 18
5 14 27 44 65 90 119 152
14 48 110 208 350 544 798
42 165 429 910 1700 2907
132 572 1638 3808 7752
429 2002 6188 15504
1430 7072 23256
4862 15194
16796
Thus, for n = 2, Table 5.2 shows the differentiation filter
equations for different values
of N .
The general equation presented in Eq. 5.5 assumes that the data
is received at an
equally spaced interval h, however it was observed that the data
output from the Vicon
system is not always steady, and while the average output is set
for 100Hz, sometimes the
interval between a set of data and another set of data can take
up to 1/5th of a second.
Thus, the following form of the algorithm (which is designed for
irregularly spaced data)
is more applicable:
f ′(x) ≈M∑k=1
ck.fk − f−kxk − x−k
.2k (5.19)
-
Chapter 5. Robots in Formation Analysis and Experiments 50
Table 5.2: Differentiation Filter Example
N Smooth noise-robust differentiators (n=2, exact on 1, x,
x2)
52(f1 − f−1) + f2 − f−2
8h
75(f1 − f−1) + 4(f2 − f−2) + f3 − f−3
32h
914(f1 − f−1) + 14(f2 − f−2) + 6(f3 − f−3) + f4 − f−4
128h
1142(f1) + 48(f2 − f−2) + 27(f3 − f−3) + 8(f4 − f−4) + f5 −
f−5
512h
Another advantage of Eq. 5.19 is that in addition to being a
centered filter (which
is beneficial for offline calculations of derivatives) it can
also be used as a forward (one-
sided) filter, which is needed for real-time online calculations
of the derivative. One of
the disadvantages of using one-sided filters however is that it
takes longer to achieve the
same noise suppression as a centered filter [19], and thus more
data points are needed at
every calculation of the derivative to achieve valid results.
The algorithm described in
this subsection shall be used for online derivative calculations
during the experimental
validation of the proposed communication loss management
technique.
5.4 Theoretical Analysis
The MRF system can be analyzed using the ADS theorem to
determine the behaviour
of the system under periods of stochastic communication loss.
The space-state model of
the system is shown in Eq. 5.3, where the A m