►
Description
Akarsh Simha
Adapting plate-solving for visual astronomy
Plate solving has become increasingly a standard component of the workflow used by amateur astrophotographers today. Thanks to plate solving, imagers can still hit the exact field they want without relying on the mount's raw pointing, or building sophisticated pointing models. The pointing needs of visual observers are much more relaxed, but the deeper observations beyond NGC/IC would still greatly benefit from arcminute-grade pointing, so that we can be certain that the faint object that takes minutes of observing to make an appearance is bang in the center of our fields. However, existing "Digital Setting Circle" (DSC) systems with encoders struggle to achieve accuracy higher than tens of arcminutes, without pointing models. Issues like the leveling of the Dobsonian, the roundness of the ground board, flexure in various aspects of the Dobsonian design and so on have a significant impact on accuracy. Can we instead use plate-solving, through a finder scope, to point dobs for visual use? This talk will go into the challenges of adapting plate-solving to visual astronomy, and my attempts at addressing some of these challenges. A field demonstration of a very crude prototype that I have built over the pandemic will feature at the end. The emphasis is that with more engineering effort, I predict that this can be turned into a system that is much more accurate and robust to scope mounting errors than DSCs.
A
All
right
welcome
everyone
to
the
november
sjaa
imaging
sig.
Tonight
we
have
a
treat.
We
have
akers
simha
who's,
going
to
tell
us
about
a
project
of
his
useful
to
visual
observers,
but
taken
from
technology
related
to
us,
imagers
and
just
I'll
say
us
a
few
sentences
about
akers.
I
met
akrish
because
he
is
a
contributor
to
k-stars.
In
fact,
for
one
year
he
ran
the
k-stars
project
right
before
the
current
maintainer
jason
mutlock
started.
A
B
Thank
you
hi,
so
hi
everyone,
I'm
sort
of
an
outsider
here,
because
I,
I
think
those
of
you
who
follow
tac
or
elsewhere
know
that
I'm
a
visual
observer.
I
do.
I
have
dabbled
in
imaging,
but
I'm
sort
of
it's
it's
not
as
exciting
to
me
as
visual
observing
and
I
don't
have
the
equipment
to
take
good
images.
B
B
You
should
just
be
doing
that
with
your
visual
scopes
and
I
never.
I
never
thought
about
it
seriously
enough
until
the
pandemic
hit,
and
I
had
some
spare
time
on
my
hands
and
so
on.
So
I
started
saying
you
know
what
I
have
nothing
else
to
do.
Let
me
try
and
build
this
system,
so
this
is
essentially
my
pandemic
project
and
it's
it's
still
very
prototypish,
but
I'm
going
to
mostly
focus
on
the
technical
ideas
that
went
into
this
project.
B
So
the
idea
here
in
in
a
sentence
is
to
try
and
use
plate
solving
to
point
visual
dobsonian
telescopes.
So
so
so
why
do
we
need?
Why
do
I
need
to
do
something?
This
complicated
well
clear,
dark
weekend
nights
are
precious
and
you
need
everything
for
visual
astronomy
to
work
out
and
there's
the
question
of
what
do
you
enjoy?
Do
you
enjoy
finding
objects?
Do
you
enjoy
observing
them
or
do
you
enjoy
both
for
some
people?
B
B
If
I
have
a
machine
star
hop
for
me,
I
will
be
much
happier
and
in
some
sense
what
plate
solving
is
giving
me
is
a
way
of
centering
the
field
that
I
want
without
me,
looking
through
the
finder
scope
and
trying
to
figure
out
star
patterns,
but
instead
have
the
algorithms
take
care
of
that
for
me
right.
So
that's
my
motivation.
B
B
When
I
bought
this
telescope,
it
had
no
encoders
on
it
and
for
many
years
I
thought
I
wanted
to
put
encoders
on
it,
but
I
didn't
have
the
mechanical
skills
to
do
it.
You
need
to
make
a
specific
kind
of
shaft,
and
you
know,
and
and
after
you
do
all
of
this
and
mount
encoders
on
your
telescope-
you
get
okay,
pointing
you
might
hit
an
object
within
15
to
20
arc
minutes
if
you're
lucky.
B
At
least
this
is
my
impression,
I'm
not
heavily
used
encoder
systems
the
one
time
I
use
them.
You
know
it
doesn't
it
doesn't
hit.
You
right
at
the
center
and
if
you
want
it
to
hit
really
accurately
you
have
to
have
your
telescope
leveled
and
you
need
to
build
a
pointing
model
so
that
so
that
it
points
correctly.
I
think
this
is
a
problem
with
imaging
mounts
as
well
in
some
sense,
so
so
the
advantage
with
plate
solving
is
in
principle.
B
You
don't
need
to
retrofit
anything.
It's
just
replacing
your
eye
at
the
finder
scope
or
in
the
eyepiece,
with
with
a
camera
right
and
in
my
prototype
I
think
I
get
great
accuracy.
Unfortunately,
the
demo
that
I
recorded
has
problems.
I
I
think
I
know
what
the
problems
are.
The
accuracy
there
is
not
good,
but
but
in
general,
I've
gotten
something
like
five
arc
minutes.
B
So
I,
my
dissatisfaction
with
encoders,
mostly
comes
from
the
fact
that
I'm
usually
observing
art,
galaxies
or
hicks
and
groups
and
such
things
and
these
things
are
tiny
and
very
faint,
and
you
know
when
you
center
one
of
these
arp
galaxies
in
an
18
inch
telescope,
which
is
what
I
have
you.
When
you
put
your
eye
at
the
eyepiece,
you
usually
see
nothing
and
you
need
to
wait
for
your
eye
to
adapt
to
the
field
and
you
know
gather
enough
signal
or
whatever
to
start
seeing
things.
B
B
So
that's
the
context
for
why
I'm
doing
this
so,
like
I
said
the
dream
I
have
is
that
you
know
there
might
there
there
probably
exists
an
electronic
finder
scope
that
somebody
will
build
and
you
replace
the
existing
finder
scope
on
your
telescope.
You
know
with
with
this
electronic
fiber
scope,
and
you
suddenly
have
push
to
capabilities.
B
B
B
So
I
think
this
group
probably
doesn't
need
to
see
the
slide
at
all.
You
probably
know
what
plate
solving
is,
but
in
case
there's
somebody
in
the
audience
that
doesn't
it's
essentially
both
the
con.
It's
it's
this
concept
that
was
introduced
in
this
paper,
astrometry.net
by
dustin
lang
and
others.
Basically,
it's
a
it's.
B
This
idea
that,
given
a
photograph
of
any
region
of
the
sky,
you
should
be
able
to
tell
me
give
me
a
mapping
from
every
pixel
on
that
photograph
to
the
ra
deck
coordinates
corresponding
to
that
pixel
so
and
they
they
just
didn't,
propose
this
idea
or
an
algorithm.
They
actually
implemented
it,
and
so
they
have
open
source
software
that
can
find
the
location,
orientation
and
scale
of
any
picture
of
the
sky.
B
So
once
you
have
the
location,
orientation
and
scale,
you
can
map
every
pixel
position
to
an
r,
a
deck,
a
pair
of
coordinates
and
out
of
curiosity,
the
paper
is
out
there.
It
does
so
by
matching
star
patterns
triangles
and
quadrilaterals,
and
it
compares
them
against
the
database.
You
can
download
the
software.
You
can
also
try
it
online
by
uploading,
an
image
and
you
amateur
astrophotographers
are
already
using
it.
Heavily
visual
observers
are
not
so
that's
what
I
hope
to
change
with
with
this
prototype
the
other
great
thing
about
plate
solving.
B
Is
it's
never
seen
a
false
positive?
It
might
say
I
don't
know
what
to
do
this
with
this
field
and
bailout,
but
if
it's
safe
that
this
field
is
so,
and
so
it's
almost
certainly
right
and
I've
seen
it
jitter
around,
but
not
by
much
right,
so
maybe
like
two
or
two
arc
minutes
is,
is
how
much
jitter
I've
seen
at
most
and
that's
probably
because
of
smudging
of
stars
and
things
like
that
and
lens
modeling
and
whatever.
B
So
I
have
a
demo
here.
I
think
I
will
skip
this
because
most
of
you
are
familiar.
If,
if
somebody
wants
to
see
it
I'll
bring
it
back,
so
what's
the
system
I'm
trying
to
build
in
conception?
So
this
is
not
my
implementation
but
ideas
on
how
we
might
bring
plate
solving
to
visual
astronomy
just
to
go
back.
Keep
in
this
keep
in
mind
that
every
every
every
one
of
these
visual
telescopes
has
a
finer
scope.
That's
on
it!
I
guess
you
can't
see
my
finderscope.
B
It's
not
mounted
up
there,
but
you've
seen
what
a
telescope
with
the
finderscope
is.
Like
you
know,
people
have
a
telrad
and
a
finder
scope
and
such
things-
and
the
idea
here
all
through
is
that
I'm
trying
to
replace
that
finder
scope
with
something
electronic.
So,
let's
first
talk
about
plate
solving
assisted,
go
to
as
used
by
you
folks,
imagers
right.
You
have
the
ability
and
the
stability
of
tracking
to
put
to
put
your
camera
in
your
main
prime
focus
or
whatever,
and
take
pictures
through
your
telescope.
B
So
if
you
have,
if
you
are
working
in
that
paradigm,
you
have
go-to,
you
have
great
tracking,
you
have
a
stable
mount,
then
what
you
can
do
is
very
simple:
you
take
a
picture
you
plate
solve
it
and
you
find
the
plate
coordinates
this
plate
center
coordinates.
You
have
some
target
in
mind,
let's
say
orion
nebula
or
whatever
you
know
it's
ra
index
from
a
catalog.
B
You
find
the
error
in
the
array
in
deck
and
if
the
error
is
larger
than
your
tolerance,
then
that
means
you're
not
on
the
object.
So
then
you
correct
the
scope
position.
So
this
way
you
don't
need
to
build
a
pointing
model.
If
your
scopes
go
to
doesn't
hit,
you
can
just
use
plate
solving
to
fix
it,
and
I
believe,
k,
stars
and
ecos
already.
B
Now,
if
the
main
telescope
is
a
visual
telescope
like
mine,
this
is
my
telescope
here
and
you
can
see
as
I
wiggle
it,
that
its
mount
is
anything
but
stable.
These
dobsonians,
if
you
try
to
do
imaging
with
them
at
best
you're
doing
lucky
imaging
well,
I
have
tried
to
put,
I
have
tried
to
put
a
planetary
camera
on
this
and
take
and
do
some
lucky
imaging
and
it
is
a
nightmare.
I
mean
I
might
shoot
100
frames
and
have
three
good
frames
right.
So
that's
that's.
B
B
My
platform
is
actually
designed
for
30
degrees,
north
latitude,
so
actually
I
have
eight
degrees
of
polar
alignment
difference
and
I
still
use
it.
It
works
well
enough
for
visual.
You
know
so
so,
basically,
taking
images
through
the
main
telescope
is
is
a
is
a
nightmare
plus.
You
also
know
that
plate
solving
smaller
fields
is
much
harder
because
there
are
just
so
many
smaller
fields
in
the
sky
that
you
have
to
search
through
your
search.
Space
is
much
larger,
so
plate
plate
solving
wide
field.
B
Images
is
much
faster
and
the
focal
length
on
an
18
inch.
Telescope
like
this
is
2000.
Millimeters
is
my
focal
length.
So
if
I
put
a
typical
camera
there,
I'm
going
to
have
a
13
arc
minute
or
six
arc
minute
field,
depending
on
you
know
whether
I
have
a
barlow
or
not,
and
so
it's
it's
going
to
take
forever
to
solve
those
tiny
fields
as
well.
B
The
best
part
is
that
this
alignment
doesn't
actually
you
don't
actually
need
to
optically
align
the
finderscope
with
the
main
telescope.
So
one
of
the
main
ideas
here
is
that,
if,
if
you
figure
out
what
is
the
degree
of
misalignment
between
your
finder
scope
and
your
main
telescope,
you
just
memorize
that
misalignment
and
apply
that
every
time
you're
done,
you
don't
actually
need
to
physically
and
optically
align
the
finder
scope.
B
So
I
have
a
one
star
alignment
process
which
works
as
follows:
I
center
a
known
star
in
the
main
telescope
and
then
I
plate
solve
through
the
finder
scope.
The
plates
the
plate.
Solving
algorithm
tells
me
here's
the
center
coordinates
of
my
plate.
This
is
the
rn
deck,
but
I
know
which
star
I
have
centered,
so
I
know
the
actual
array
deck
in
my
main
telescope.
B
So
I
just
ask
the
you
know
I
just
compute
with
the
plate
solution
that
I
have
found.
What
is
the
pixel
offset
on
the
plate
x,
comma
y,
corresponding
to
the
star
that
is
centered
in
the
telescope?
So
here's
a
graphic
that
introduced
that
that
explains
this.
So
the
optic
axis
of
the
finder
scope
is
pointing
somewhere,
it's
not
aligned
with
the
main
telescope.
B
That's
the
center
of
this
plate,
the
actual
star
that
I've
aligned
in
the
finder
scope
is
at
some
offset
x,
comma
y
and
my
one
star
alignment
procedure
essentially
memorize.
That's
that
offset
so
if
you're
working
with
digital
setting
circles,
which
is
another
word
for
encoders,
you
actually
need
to
align
two
stars
or
three
stars.
So
even
the
alignment
process
here
is
simpler.
You
just
need
to
align
one
star,
okay,
yeah
and,
like
I
said,
no
need
to
optically
align
the
finer
scope
in
the
main
telescope.
B
If,
if
this
confuses
you
think
of
it,
this
way
that,
if
you
think
about
telescope
axis
and
finder
scope
axis
as
vectors,
then
the
difference
in
those
two
vectors
which
is
the
offset
that
rotates
rigidly
with
your
telescope
system,
assuming
that
there's
no
flexion
in
your
telescope
system
and
so
on.
You
know,
assuming
that
everything
rotates
rigidly
this
offset
is
going
to
be
it's
going
to
remain
fixed,
no
matter
where
you
point
the
telescope
right.
So
so
that's
why
just
knowing
this
offset
is
good
enough.
B
Okay.
So
how
does
plate
solving
work
if
you're
trying
to
position
a
visual
telescope
by
plate
solving
through
the
finder
scope?
So
by
the
way?
Just
to
give
you
context,
my
final
scope
is
270
millimeters
in
focal
length
and
with
my
zw
asi
290
mc
camera,
which
is
overkill
for
this
purpose.
But
this
is
the
camera
I
had.
I
get
a
field
of
view
of
about
one
and
a
half
degrees.
B
So
that's
how
wide
my
fields
are
field
is
and
if
I'm
not
tracking
properly
for
a
few
seconds,
you
know
it.
It
really
doesn't
matter
for
for
what
I'm
doing
here
yeah.
So
that's
just
to
set
up
how
the
finder
scope
addresses
the
problems
that
I
mentioned
with
plate
solving
through
the
actual
main
telescope
so
anyway,
so
how
does
plate
solving
through
a
finder
scope?
Work
it's
very
similar
to
the
previous
flowchart.
I
showed
you
take
a
picture
through
the
finder
scope.
B
Now
you
plate
solve
it
and
you
compute,
the
radix
coordinates
corresponding
to
the
plate,
coordinates
that
you
memorized
x,
comma
y.
During
the
alignment
process.
You
don't
look
up.
The
center
coordinates.
Instead,
you
look
at
those
offset
coordinates
and
then
you
ask
how
different
is
that
from
the
target,
our
deck?
You
get
a
you
know
offset
in
rna
and
an
offset
index,
and
if
we
are
not,
you
know
with
intolerance,
then
then
we
correct
the
scope
position.
This
is
now
assuming
that
you
have
a
go-to.
I
guess
this
is
similar.
B
If
you
would
like
to
conceive
of
something
it's
like
this
celestron
product.
That
is
meant
for
selectron
mounts.
I
don't
know
what
it's
called.
I
think
it's
part
of
it's
one
of
the
incarnations
of
their
star
sense
system,
which
is
a
small
finder
scope
that
sits
on
top
of
your
telescope
and
guides
your
celestron
mount.
So
it's
sort
of
like
that,
and
you
have
to
do
a
one-star
alignment
even
there
to
bring
to
to
compute
this
offset.
B
B
Okay.
So
now,
visual
visual
astronomers
have
even
more
problems
than
this.
We
don't
have
go
to
a
lot
of
us,
so
I
track
my
telescope
in
an
equatorial
platform
and
if
you
don't
know
what
an
equatorial
platform
is,
it's
it's
a
very
crude
device
that
it's
a
platform
for
your
telescope
that
tracks
the
telescope
for
maybe
an
hour
along
one
axis,
and
it
does
so
by
rocking
the
telescope
around
the
pole
and
yeah
and
and
maybe
you'll
get
an
hour
of
tracking,
and
then
you
reset
it.
B
B
Okay,
so
how
do
we
adapt
place
solving
to
push
to
this
is
surprisingly
a
lot
more
complicated
than
you
would
imagine
that
it
is
so.
In
fact,
we
don't
even
really
need
to
we
need,
we
don't
need
go-to
or
even
tracking
in
principle.
I
still
need
tracking
for
my
setup
and
I
don't
think
that's
a
I
think.
B
B
So
that's
not
the
problem,
you
don't
need,
go
to.
You,
don't
need
tracking
to
do
the
plate
solving
part.
The
problem
is
that
plate
solving
takes
time
and
the
best
I
have
managed
with
the
entire
stack
so
far
is
once
all
per
three
seconds.
That's
about
maybe
a
min,
maybe
a
second
and
a
half
for
the
exposure
about
half
a
second
to
transfer
the
data
from
the
camera
to
my
laptop
and
about
half
a
second
to
solve
it.
B
So
the
solving
itself
is
extremely
fast
because
I
I
know
it's
scale
hint
during
the
alignment
process.
I
actually
figure
out
what
is
the
scale
arc?
Second
per
pixel
of
my
camera,
and
once
I
give
the
solver
the
scale,
it's
not
doing
a
blind
solve
and
that's
extremely
fast,
so
I
can
get
a
solve
every
three
seconds
or
so,
but
for
pushing
a
telescope
we
need
extremely
fast
feedback.
We
want,
you
know,
10
frames
per
second,
at
least
right,
so
you
know
how
you're
moving.
B
Of
course,
if
it's
a
motor-
and
you
say-
oh
you're,
15
degrees
off
the
motor,
the
encoders
in
the
motor
itself
know
what
15
degrees
is.
So
you
can,
you
can
run
the
scope,
15
degrees,
but
I
have
no
sense
for
what
15
degrees
is.
So
I
want
to
be
seeing
some
feedback
saying
I'm
getting
closer
to
my
target
or
not,
and
so
I
need
faster
feedback
and
plate
solving.
B
So
the
solution
I've
come
up
with
and
I
suspect
that
celestron
does
something
very
similar
is
to
combine
plate
solving
with
something
else.
That
gives
me
fast
feedback
and
that
is
encoders
or
imus
and
I'll
explain
what
an
imu
is
now
I
started
this
entire
thing,
with
the
premise
of
avoiding
encoders
right
and
and
and
saying
why
I
don't
like
encoders
well,
there
might
still
be
advantages
to
adding
plate
solving
to
a
telescope
that
already
has
encoders,
because
you
could
bring
the
accuracy.
B
You
know
you
could
increase
the
accuracy
and
also
avoid
doing
this
two-star
alignment
because
plate
solving
tells
you
where
you
are
so
you
know
it
has
advantages,
but
I
wanted
to
avoid
encoders.
So
I
went
the
imu
route
and
what
is
an
imu
imu
stands
for
inertial
motion
units,
these
mems
devices,
micro,
electron
mechanical
devices,
these
days
they're
everywhere.
B
You
know
they're
in
your
phones,
they're
in
your
cars,
they're
they're,
just
pretty
ubiquitous
now
and
what
they
are
is
a
three-axis
accelerometer,
three-axis
gyroscope
and
a
three
axis
magnetometer
x,
y
and
z.
Basically,
three
axes
all
combined
into
a
chip,
and
so
if
you
are
a
tinkerer
who
works
with
arduinos,
you
can
just
buy
this
arduino
nine
axis
imu
shield,
which
is
what
I
use,
and
it
can
give
you
approximate
orientation
now
what
these
sensors
are
good
at
is
they
can
sense,
changes
in
motion?
B
They
are
very
noisy.
They
are.
They
are
very
inaccurate
and
people
who
have
tried
to
use
them
to
point
their
telescopes,
their
threads
on
tsc
and
so
on.
They
just
don't
really
work
very
well,
maybe
you'll
get
a
degree
or
two
degrees.
You'll
be
a
degree
or
two
degrees
off
which
is
a
complete
deal,
breaker,
so
yeah
these
these
devices
are
fast.
They
are
cheap,
they
are
small,
but
they
only
provide
rough
positioning.
B
So
that's
the
problem
with
them,
but
hey
they
are
fast.
So
that's
good
enough
for
me.
You
know
I
can
combine
it
with
blade,
solving
and
get
something
that
is
accurate
and
fast,
because
the
fast
comes
from
the
imus
and
the
accurate
comes
from
the
plate
solving
right.
So
that's
what
I'm
trying
to
do
here
yeah.
So
I
already
mentioned
that
combined
plate
solvent
which
is
slow
but
accurate
positioning
with
imu
output,
which
is
fast
but
rough
positioning
right,
but
the
caveat
is
now
there
are
more
things
to
align.
B
The
imu
has
an
x
y
and
z
axis.
That
is
printed
on
the
circuit
board.
If
you
like,
depending
on
how
the
chip
is
oriented,
and
now
I
have
a
problem,
I
have
to
figure
how
to
align
this
imu
with
my
telescope
access.
So
in
some
sense
there
are
three
axis
now,
if
you
like.
There's
the
telescope
access
not
shown
here.
I've
just
marked
an
arrow,
there's
the
finder
scope
axis,
and
then
you
can
conceive
of
an
imu
access,
even
though
really
we
should
be
talking
about
frames
and
not
axes
but
yeah.
B
B
B
If
I,
if
that
metal
happens,
to
be
iron
and
has
a
little
bit
of
stray
magnetic
field
that
can
throw
off
its
sense
of
north
right-
and
this
is
a
problem-
that's
very
challenging
to
solve-
and
the
way
I
solve
it
is
to
assume
that
the
north
is
unreliable.
So
I
only
think
of
I
only
use
the
imu
with
an
arbitrary
as
the
mud
offset.
So
that
means
I
need
to
do
two
pieces
of
calibration
one.
B
I
need
to
figure
out
how
my
imu
is
mounted
on
the
telescope,
and
the
second
is,
I
need
to
figure
out
which
way
is
not
because
the
imu
is
not
as
unreliable
well.
The
great
news
is,
this
does
not
add
more
manual
alignment
steps
because
we
can
use
plate
solving
to
keep
calibrating
these
sensors,
like
I
like
I
mentioned
so
the
algorithm
there
that
I'm
using
is
pretty
involved,
so
I
won't
get
into
the
details.
B
I'm
just
going
to
show
a
caricature
of
it
in
in
the
coming
slides
yeah,
like
I
said,
lots
of
complicated
math,
but
it's
working
so
now.
Let
me
explain
how
I
do
the
calibration,
so
there
are
two
things
to
calibrate,
which
means
I
need
two
data
points
with
with
to
calibrate
the
imu
access
to
the
finder
scope
access.
B
So
the
first
data
point
comes
for
free,
along
with
this
one
star
alignment
that
I
have
to
do
between
the
telescope
and
the
final
scope
axis.
So
the
alignment
procedure
is
as
follows.
I
center
a
known
star
I
capture
and
solve
through
the
finder
scope.
I
compute
this
pixel
offset
x,
comma
y,
based
on
the
alignment
stars
array
index,
and
that
gives
me
one
calibration
point
for
it.
Sorry,
it
also
gives
me
one
calibration
point
for
the
imu,
because
I
read
the
imu.
B
That
tells
me
one
piece
of
in
one
piece
of
information
amongst
these
two
calibration
informations
that
I
need
the
the
second
thing.
The
the
second
alignment
step
is
very
simple.
I
just
need
to
move
the
telescope.
Just
literally
move
the
telescope
right
and
once
I
move
the
telescope
it
triggers.
I
have
plate
solving,
usually
running
in
a
loop,
so
it
triggers
a
capture
and
solve
and
it
it
computes
the
alphas
of
my
telescope,
and
I
know
what
the
imu
reading
is.
B
So
I
have
the
second
calibration
point
just
by
moving
my
telescope.
So
the
first
calibration
point
tells
me
how
the
imu
is
rotated
with
respect
to
the
telescope
axis.
The
second
calibration
point
tells
me,
which
way
is
not
in
in
the
imu's
frame
of
reference
and
from
these
two
calibration
points
I
have,
I
have
a
full
way
to
map
the
imu's
output
to
the
positioning
of
my
telescope.
B
Now
the
thing
is
this
is
unreliable,
and
I
know
this
because
these
sensors
drift
over
time.
They
have
noise,
they
have
all
sorts
of
problems,
and
you
know
the
stream
magnetic
field
can
come
and
go
and
and
so
on.
So
this
calibration
af
keeps
running
in
a
loop
every
time
I
have
a
play
solver.
I
have
new
calibration
data
point
and
I
can
add
it
and
recalibrate
the
imu.
So
so
this
is.
B
B
This
is
I've
gone
over
this
a
bunch
of
times
and
I'll
come
back
to
this
pause
image
part
in
a
moment,
but
now
what
I
do
is
because
I
have
just
recalibrated
the
imu
freshly
at
every
plate
solve.
I
am
able
to
use
the
fast
trust
of
imu
over
short
moments
and
or
short
periods
of
time
and
use
the
imu
to
give
guiding
instructions
in
altitude
and
azimuth.
B
So
we
have
a
fast
servo
that
involves
the
imu
and
we
have
a
slow
servo,
because
every
time,
the
every
time
the
scope
is
stationary,
I
am
able
to
take
an
exposure,
make
a
plate
solve
and
recalibrate
the
imu
and
present
a
new
offset
to
the
user.
So
so
there's
a
slow
but
accurate,
servo
and
there's
a
fast
but
inaccurate
servo
and
the
two
work
together
to
tell
me
how
to
push
my
telescope.
So
I
get
fast
feedback
from
the
imu,
but
the
imu
is
constantly
corrected
by
the
plate
solve
whenever
possible.
B
So
this
is
the
system
and
right
and
the
the
problem.
The
problem
that
comes
up
here
is
imu's.
At
least
the
ones
I
am
using
are
not
really
sensitive
to
small
movements
and
jerky
movements,
so
once
my
offset
is
starting
to
look
like
this
point,
zero
degrees
in
an
altitude
point
two
degrees
in
azimuth.
B
The
imu
is
sort
of
useless.
So
at
this
point
I
want
to
present
the
user
with
something
very
intuitive
on
how
to
do
this
last
mile
and
the
the
way
I
do.
That
is
that
I
have
a.
I
have
a
display
right
next
to
my
telescope
here
right
next
to
my
eyepiece,
and
with
this
display
I
present
the
user
with
the
correctly
oriented
pause,
image,
pauses,
palomar,
optical
sky
surveys.
Digital
is
dss.
You
might
know
it
as
dss,
it's
basically
there.
Okay,
you
can
think
of
it.
B
As
this
there
exists
a
software
and
a
database
that
I
can
query
hey,
give
me
a
photograph
of
the
sky
around
this
area
and
deck
and
it'll
do
it
right
with
with
with
north
orientation
guaranteed.
So
basically,
what
I
do
is
in
that
display.
That's
right.
Next
to
my
eyepiece
is
present
a
picture
of
what
I
am
expected
to
see
in
the
eyepiece
from
the
digitized
sky
survey
correctly
oriented,
and
that
way
the
user
can
do
the
last
mile
in
the
eyepiece.
B
B
This
is
very
intuitive.
You
just
move
the
telescope
such
that
you
bring
the
object
from
six
o'clock
to
the
center
and
yeah.
So
the
last
mile
is
is
really
straightforward
like
this,
and
so
that's
the
entire
system.
So
it's
a
it's.
It's
blade
solving
to
calibrate
the
imus.
I
am
used
for
fast
feedback
to
get
to
the
last
mile
and
the
last
mile
is
an
eyepiece
with
a
reference
image
guiding
you
in
exactly
which
direction
and
where
the
object
is
right.
B
So
in
practice
this
system
can
such
a
system
can
be
built.
In
my
opinion,
it
it.
I
think
all
that
is
left
is
engineering
work,
and
so
this
is
sort
of
my
north
star
system
and
it'll
have
this
sort
of
user
installation
where
the
user
just
removes
their
existing
finder
scope
puts
in
this
electronic
finder
scope,
which
is
fully
integrated.
B
It
has
its
own
computer
running
plate
solving
like
a
raspberry,
pi
or
whatever,
and
they
mount
an
hdmi
tablet
or
a
display
next
to
the
eyepiece
and
connect
wirelessly
to
the
system
and
in
in
in
the
ideal
user
experience.
You
know
this.
The
combination
of
fast
and
slow
loops
are
so
seamless
that
iterations
are
not
needed
right
now.
I
actually
need
to
iterate
about
four
or
five
times
to
get
to
the
object,
so
I'll
move
my
telescope
and
then
I'll
wait.
Let
go
of
it.
B
Wait
for
a
plate
solved
and
the
plate
fills
recalibrates,
my
imu,
so
my
offset
might
be
looking
like
0.5
degrees
and
then
it'll
say:
hey,
it's
not
0.5
degrees,
it's
actually
one
degree,
because
the
imus
drifted,
and
so
I
have
to
do
about
four
or
five
iterations
to
narrow
down
on
the
object.
I'm
not
doing
any
star
hopping,
I'm
just
letting
the
plate
solving
recalibrate
my
imu,
so
hopefully
we
can
engineer
a
system
where
this
doesn't.
This
is
not
necessary
and
the
fast
and
slow
loops
calibrate
so
quickly
that
it's
it.
B
These
things
matter
how
solid
your
mount
is,
whereas
here
in
this
system,
all
that
matters
is
thermal
drift
and
the
flexure
difference
between
the
two
optic
axis,
the
one
of
the
finder
scope
and
of
the
telescope.
So
the
the
pointing
model
here
assuming
that
thermal
drift
can
be
solved.
It
only
has
to
learn
flexion
and
I
think
that's
that,
so
I
I
believe
that
you
might
be
able
to
increase
the
model
of
this.
The
accuracy
of
this
pointing
model
with
just
a
few
data
points.
B
D
B
We
have
this
mess
of
wires
going
to
my
laptop,
because
my
prototype
involving
the
raspberry
pi,
which
is
a
single
board.
Computer,
did
not
quite
work
out
on
the
field
and
needs
some
re-engineering,
but
in
principle
all
of
these
wires
could
be
eliminated
except
a
few
go
into
the
secondary
cage
which
I'll
talk
about,
and
you
could
mount
a
board
computer
right
here
and,
of
course
you
will
need
some
way
to
power.
It
so
there'll
be
some
power
supply
wires
and
now,
let's
take
a
look
at
the
secondary
cage
of
my
telescope.
B
Currently,
it
is
showing
nothing,
but
it
it's
red
filtered
and
it
has
a
touch
panel
overlay
on
top
of
the
red
filter.
So
you
can
still
use
it
like
it's
a
touch
screen
and
an
hdmi
cable
runs
down
to
my
computer
from
here.
This
is
to
show
a
reference
image
pause
image
for
what
I'm
seeing
in
the
eyepiece.
B
Now
I
actually
have
here
a
track
pad
and
that
trackpad
can
be
eliminated.
It's
it's
there
because
I
have
a
driver
issue
with
multi-touch
for
the
panel.
That's
paste
it
on
top
of
this
touchscreen.
If
we
used
an
amoled
display
which
doesn't
have
a
backlight
like
ed
allen
does
on
his
telescope,
then
we
should
be
able
to
just
use
the
native
touch
screen
of
the
display,
rather
than
have
all
of
these
complicated
workarounds
to
red
filter,
a
regular
touch
screen
lcd
touch
screen.
B
My
telescope
rides
on
an
equatorial
platform
which
is
seen
here
as
a
black
item
and,
and
that
makes
the
mathematics
harder,
but,
unlike
with
dses,
you
do
not
need
to
re
re-align
re-perform
a
two-star
alignment.
Every
time
you
reset
your
platform
because
because
the
only
because
the
finder
scope
is
attached
to
the
telescope,
so
any
offset,
the
outside
of
that
system
doesn't
affect
the
performance
of
the
system.
The
so
the
platform
only
complicates
the
math.
B
B
B
B
So
you
don't
you
see
it's
not
yet
giving
me
guiding
instructions
that
label
that
just
says
guide
has
no
instructions
on
how
to
push
my
telescope
as
yet,
and
that's
because
the
imu
calibration
remember
requests
two
data
points,
so
I
need
to
generate
the
second
one
by
going
to
my
telescope
and
moving
it
in
altitude
by
about
10
degrees.
You
know
this
is
so
that
it
can
fix
the
magnetic
magnetometer
or
so
that
it
can
fix
north
without
relying
on
the
magnetometer.
B
So
now
that
it's
done
that
you
see
it,
it
just
takes
another
solve,
and
now
I
have
guiding
instructions.
So
all
I
needed
to
do
to
get
that.
Second
calibration
point
was
to
move
my
telescope
so
now
and
he
here's
the
part
that
needs
a
lot
of
improvement,
but
there
it's
possible
to
improve
it.
I'm
going
to
use
the
imus
to
push
my
telescope
so.
B
B
So
if,
as
you
can
see
in
this
display,
I'm
not
far,
the
object
is
just
below
in
my
ips
field
of
view
at
like
six
o'clock
and
remember
this
is
oriented
correctly
and
that
circle
actually
does
represent
the
fov
of
my
current
eyepiece,
which
is
a
six
millimeter
eyepiece
on
this
telescope.
So
I'm
just
going
to
bring
it
in
from
six
o'clock
and
there's
the
object.
It's
right
there
and
now
let
me
center
it
to
show
you
that
you
know
this.
The
solving
is
sensitive
to
that
kind
of
difference.
B
So
I
have
now
centered
it
in
my
eyepiece
and
once
the
solver
hits
again,
it
should
show
it
pretty
close
to
center
yeah.
I
think
that's
the
amount
of
error.
I
have
right
now
what
you're
seeing
so
so
I
wish
that
you
know
I
I
would
have
hit
much
closer
to
center,
but
I
think
I
know
what's
going
on,
I
changed.
B
I
changed
the
sling
that
holds
my
primary
mirror
to
what's
called
a
glitter
sling
and
I
think,
there's
a
problem
where
my
mirror
and
collimation
sort
of
shift.
Every
time
I
move
the
telescope
around,
because
it's
on
a
platform.
So
the
anyway,
so
basically
my
I
think
I
I
have
a
problem
where
my
mirror
is
physically
moving
as
I
jerk
the
telescope
and
it
doesn't
go
back
to
center,
so
actually
I'm
losing
collimation.
B
This
is
because
I
have
a
sling
mount
and
I
think
that's
the
primary
source
of
that
error,
but
the
error
that
you
saw
at
the
end,
so
you
what
you
should
have
seen
if
we
had
perfect
positioning,
is
that
that
object.
M77
should
have
been
right
in
the
center
of
that
display
in
the
center
of
that
circle,
but
the
error
that
instead,
it
was
at
the
edge
of
the
circle.
B
B
So
what
I'd
like
to
try
next
and
I'm
looking
for
suggestions
and
contributions
and
this
entire
system
is
sort
of
non-commercial.
I
don't
plan
to
make
anything
out
of
it.
I
have
to
go
through
some
bureaucracy
with
my
employer
to
release
the
software
as
open
source,
so
I
can't
share
the
code
or
the
software
yet,
but
I'm
happy
to
share
all
the
other
ideas.
Like
you
know:
hardware,
mathematics,
yeah,
just
just
ask
for
it.
B
So
so
the
things
that
I'd
like
to
try
is
transition
to
a
shorter
focal
length,
so
I'm
currently
solving
with
a
270
millimeter
focal
length
and
but
the
the
problem
is
that
my
limiting
factor
that
is
giving
me
that
final
error
is
flexure
and
imperfections
in
my
mount
or
imperfections
in
the
way.
My
telescope
preserves
alignment
and
not
how
many
not
the
pixel
size
of
my
image
right,
not
as
not
the
resolution
of
my
imaging.
B
So
if
I
go
to
a
shorter
focal
length,
I
might
be
able
to
get
away
without
tracking,
because
the
star
smudging
is
not
going
to
be
detected
by
the
algorithm
and
I
might
be
able
to
get
shorter
exposure
since
brighter
stars
are
likely
available
in
this
larger
fob
that
I
will
get
and
in
fact
one
of
my
friends
who
I've
been
talking
to
harshad
who's,
the
author
of
this
app
called
sky.
He
reports,
five
art
minutes
accuracy
with
just
the
phone
camera
right.
So
it's
pretty
impressive.
B
B
Okay,
yeah,
so
this
thing
has
a
lot
of
flexure
because
this
body
of
the
camera-
and
all
of
that
is
really
heavy.
So
that's
something
I
want
to
move
away
from,
so
that
so
that
there's
not
as
much
flexure
and
another
great
thing
is,
you
probably
use
zw.
Some
of
you
probably
use
zwo
asi
290
cameras
for
imaging
the
same
sensor.
B
I
forget
his
last
name.
I
was
talking
to
a
person
named
peter
from
albuquerque
at
the
oki
tech
star
party,
which
is
where
all
of
this
video
was
shot,
and
he
was
telling
me
that
yeah
they
are
seeing
pretty
good
star
density
with
just
this
camera.
So
I'm
very
curious
and
I
bought
the
board.
I
just
haven't-
had
the
time
to
go
and
put
it
in
and
try
it
out.
B
So
that's
where
I
am
what
I'd
like
to
try
next
and
that's
pretty
much
the
end
of
my
talk,
I'm
open
for
questions
comments.
I
have
a
few
extra
slides
with
pictures
in
case
somebody
wants
to
see
maybe
I'll
stop
on
this
picture
of
the
software
ui.
This
is
the
prototype
software
ui
I
built
in
python
to
orchestrate
this
entire
process.
Yeah.
That's
that's
my
talk.
E
B
That
is
a
very
interesting
question.
I
should
be
using
kelman
filters,
but
I
am
not
the
reason
is.
I
haven't
figured
out
how
to
use
kelvin
filters,
so
I
I
don't
have
background
in
the
in
in
that
domain
of
you
know,
sort
of
probabilistic,
modeling
or
whatever
you
would
call
it
I'd
like
to
acquire
background
in
it.
B
So
the
normal
sphere
that
we
know
is
has
a
two-dimensional
surface,
but
just
imagine
one
extra
dimension.
So
it's
a
three-dimensional
sphere.
The
the
the
possible
possibilities
of
rotations
which
are
quaternions,
lie
on
that
space,
and
I
I
don't
know
how
to
formulate
this
kelman
filter.
Mathematics
in
that
domain
and
whatever
literature
I
could
find
on,
it
was
too
hard
for
me
to
understand
so
right
now,
I'm
just
recalibrating
with
no
memory
right.
Every
time.
B
I
get
a
point
that
where
I
can
figure
the
orientation
I
use
that
every
time
I
get
a
point
where
I
can
also
figure
the
magnetic
not
which
is
it.
I
need
a
big
shift
in
altitude
for
that.
I
I
just
I
just
use
it.
So
it's
that
point
estimates
and
they
don't
have
any
memory
and
that's
something
I
love
to
fix.
E
E
The
next
position,
where
your
your
airplane
will
be,
but
given
that
that
drifts
over
time
and
is
not,
is
a
fast
but
not
precise,
every
once
in
a
while,
they
correct
it
with
gps
data,
which
is
essentially
the
same
thing
that
you
are
doing,
but
you
are
not,
of
course,
based
on
gps,
but
based
on
plate
solving.
So
maybe
you
can
find
it
some
precook
the
code
in
the
ardupilot
community.
B
B
B
B
A
B
A
Yeah,
how
about
voice
about
audio
like
wearing
an
ear,
a
white?
You
know
a
bluetooth,
headset
and
the
thing
somehow
pinging
you
without
being
too
annoying,
because
I
would
imagine
that
don't
you
want
to
keep
your
eyes
in
the
telescope
while
you're
doing
all
this
and
while
you're
moving
it
around,
and
it
seems
to
me
like,
if
you're
moving
away
and
looking
at
your
display
and
then
coming
back
you're
being
that's.
B
B
So
right
now
I
haven't
had
any
issues
using
the
display,
because
it's
like
I
mentioned
in
that
video
that
I
shot,
I
put
an
actual
plexiglas
red
filter
on
it.
You
know
anything
that,
in
my
experience,
unless
the
display
is
very
good
at
doing
a
zero
intensity
black
right
any,
if
you
don't
put
a
plexiglass
filter
on
the
display
and
you're,
just
using
red,
gelatin
or
you're,
trying
to
use
the
display's
internal
settings
or
like
digitally
say
that
you
know
display
only
red
pixels.
B
My
experience
is,
unless
the
display
is
very
good,
it's
going
to
still
leak
a
lot
of
white
light
that
will
hamper
your
adaptation,
and
so
I
went
through
this
extra
length
to
put
this
plexiglas
screen,
and
now
it
doesn't
bother
my
adaptation
to
look
at
that.
So
so
I
haven't
had
so
much
of
a
problem
with
it,
but
you're
right.
I
it's
it's
much
nicer.
If
I
could
just
keep
my
eye
on
the
eyepiece
all
through
yeah
I'll
think
about
that.
F
B
C
B
The
problem
is
whatever
I
described
earlier
right:
the
the
stability
of
the
tracking
end
of
this
mount.
It's
so
shaky
that
you
will
have
very
smudge
stars.
That's
the
problem
number
one
problem
number
two:
unless
the
camera
is
extremely
sensitive,
which
you
know
night
vision
devices
are
so
that
I
think,
is
a
possibility.
B
The
second
problem,
though,
is
that
solving
tiny
fields
takes
longer,
so
you
have
to
make
sure
that
it's
not
a
blind
solve
and
even
then
you're,
just
searching
through
more
fields,
deeper,
star
indices.
One
of
the
advantages
that
I
have
is,
I
don't
know
how
how
much
you're
tinkered
with
plate
solving.
But
there
are
these
huge
index
files
that
are
based
off
various
catalogs
right.
B
One
of
the
advantages
that
I
have
with
the
wide
field,
finder
scope
plate
solving
is,
is,
I
can
load
the
entire
index
and
hold
it
in
memory.
So
it's
in
my
cache,
the
very
first,
the
the
very
second
time
I
hit
plate
solving
it's
no
longer
reading
from
the
ssd
it's
just
reading
from
memory,
so
that
improves
the
speed
of
my
solving
the
problem
with
the
large
star
industries.
B
Of
course,
this
finder
scope
method
also
has
its
own
share
of
complications,
which
is
flexure
and
really
and
thermal
drift,
and
I
have
really
no
proper
way
of
modif
modeling
this,
but
I
think
night,
vision,
eyepiece,
is
an
interesting
idea.
If
you
have
a
camera
that
you
can
hook
into
your
eyepiece
that
or
into
your
focuser,
that
has
enough
of
a
sensitivity
that
you
can
run
very
short
exposure.