XV111
PREFACE
The
approach
is a
balanced combination
of
mathematics
—
linear systems
and
probability
theory
— in
order
to
understand
how a
state estimator should
be
designed, with
the
necessary
tools
from
statistics
in
order
to
interpret
the
results.
The use of
statistical techniques
has
been
somewhat neglected
in the
engineering literature pertaining
to
state estimation,
but it is
necessary
for
(the nontrivial task
of)
interpreting stochastic data
and
answering
the
question whether
a
design
can be
accepted
as
"good."
This
is
particularly important
for
practicing engineers
and is
presented
in
sufficient
detail based
on our
belief (and extensive experience with real systems)
that
it
should
be an
integral part
of
advanced engineering education.
1
The
material covers
the
topics usually taught
in
control-oriented EE/systems
and
aeronauti-
cal
engineering programs.
The
relevance extends
to
other areas dealing with control
in
mechan-
ical
or
chemical engineering. Recently,
the
state estimation techniques have been gaining wider
attention
due to
their applicability
to
such
fields
as
robotics, computer vision
for
autonomous
navigation,
and
image
feature
extraction with application
to
medical diagnosis. While
the
course
is
mainly directed toward
the
M.S. students,
it is
also part
of the
Ph.D. program
at the
University
of
Connecticut, with
the
intent
of
providing
the
students with
the
knowledge
to
tackle real-world
problems, whether
by
using existing algorithms
or by
developing
new
ones.
2
The
presentation
of the
material stresses
the
algorithms, their properties
and the
under-
standing
of the
assumptions behind them.
We do not
subscribe
to the
philosophy
of
"Give
me
the
facts
and
don't bother
me
with
details."
Consequently,
proofs
are
given
to the
extent that
they
are
relevant
to
understanding
the
results. This
is
intended
to be a
modest step
in
bridging
the
much talked about "gap between theory
and
practice"
— it
will illustrate
to
students
the
usefulness
of
state estimation
for the
real world
and
provide
to
engineers
and
scientists working
in
industry
or
laboratories
a
broader
understanding
of the
algorithms used
in
practice.
It
might
also avoid
the
situation summarized
by a
participant
at one of the
continuing education courses
taught
by the
first
author
as
follows:
"Although
I
studied Kalman
filters
when
I
worked
to-
ward
my
Ph.D.
(at one of the
major
U.S. universities),
I did not
expect that they worked with
real
data."
This happens when, because
of the
theorems,
the
students cannot
see the
forest
of
applications.
Tuning
of a KF — the
choice
of its
design parameters
— is an
art.
One of the
contributions
of
this text
is to
make
it
less
of a
black magic technique
and
more
of a
systematic approach
by
connecting
the
filter
parameters
to
physical system parameters, namely,
the
object motion
uncertainty
— its
predictability
— and the
sensor measurement uncertainty
— the
sensor errors.
This
is
particularly important when
KFs are
used
as
modules
in an
adaptive estimator, like
the
Interacting Multiple Model (IMM) estimator. Another contribution
is
providing guidelines
as to
what estimator should
be
used
in
specific
problems, namely, when
an
(adaptive)
IMM
estimator
is
preferable
over
a KF.
The
Text
and
Hyperlinked
Viewgraph
Format
The
format
of
this text
is
also unique
—
textgraph
— in
that
it
dares
to
attempt
to
accomplish
two
goals
in one
format:
to
serve
as a
self-contained concise text, without excess verbosity, and,
at
the
same time,
to
enable
the
lecturer
to use the
pages
of
this text
as
viewgraphs
for
lectures.
To
this purpose,
a
double-spaced version
of all the
pages
of
this text (with appropriate page breaks)
1
The
authors
disclaim
any
responsibility
for
severe
damage
readers
might
suffer
when
falling
asleep
face
forward
while
reading
this
book.
2
As Ben
Franklin
said,
the
goal
in his
life
was "to
rise
to an
(sic)
happy
mediocrity
and
secure
competency."
Our
objective
is to
provide
the
tools
for
future
successful
applications.