►
From YouTube: SatPIC: 2014 Spring NuPIC Hackathon Demo
Description
Shashwat Kandadai & Andrew Morrison
A
So
this
is
andy,
I'm
joshua
and
the
goal
of
our
project
was
to
have
newpick
predict
the
orbit
of
a
satellite.
So
what
we
did
was
well
actually,
let
me
give
you
some
background
on
how
people
track
satellites.
A
Today,
it's
kind
of
the
same
way
it's
been
done
since,
like
the
60s
and
70s,
and
actually
like
pinpointing
a
satellite
out
in
space,
is
really
difficult,
so
people
like
norad
and
the
military,
they
set
up
massive
arrays
of
radars
that
are
like
gigantic
over
acres
and
they
just
like
beam
into
space
like
at
different
places
on
earth
and
they're
able
to
detect
things
like.
Oh
that's,
iss.
Oh
that's
a
rock!
Oh!
That
satellite
has
been
dead
for
like
the
last
20
years,
whatever
so
they
can.
A
A
A
A
Okay,
if
you
saw
this
over
here
with
this
velocity,
then
now
it
should
be
over
here
with
this
velocity
and
it
tends
to
be
pretty
accurate,
but
norad
has
to
keep
publishing
new
element
sets
for
the
satellites,
because
within
a
week
or
so
it
tends
to
start
drifting
because
actual
space
is
like
really
complicated
and
if
they
fly
around
and
like
it's,
it's
not
really
as
predictable
as
the
math
would
make.
It
seem.
A
So
that's
for
like
most
satellites
with
satellites
like
the
iss
they're
active
satellites,
so
they
can
like
move
up
and
down
and
they
have
they
have
little
jets
on
them.
They
like
scoot
around
and
do
weird
things,
so
their
position
is
even
more
unpredictable,
so
norad
actually
updates
the
iss
like
three
or
four
times
a
day,
so
they
keep
that
up
to
date.
Now,
of
course,
the
iss
is
like
super
advanced
and
modern
and
they
have
like
multiple
redundant
gps
like
things
on
them,
so
they
know
where
they
are.
A
But
this
is
the
like:
the
method
that
ham
radio
enthusiasts
use
to
talk
to
like
the
really
simple
cubesats
and
the
hamsats
and
stuff
like
that.
So
what
we
did
was
we
took
the
iss
and
we
took
a
bunch
of
these
element
sets
and
we
propagated
the
path
of
the
iss
over
many
different
element
sets.
A
So
we
could
get
a
very
smooth
and
as
accurate
as
possible
model
of
the
iss's
path
over
several
days,
and
we
gave
that
to
subitize
new
pig,
harness
and
basically
we're
looking
for
anomalies
and
hoping
that
lupic
would
learn
and
settle
down
and
kind
of
predict.
The
circular
or
elliptical
motion
of
the
satellite.
So
from
a
single
point
on
earth,
a
satellite
appears
to
like
kind
of
like
move
in
a
it
has
like
a
phase
shift.
It
doesn't
rise
and
set
at
the
same
place.
A
You
know,
oh,
you
know
it's
tuesday,
let's
drop
down
a
bit
or
it's
wednesday.
Let's
go
over
here,
so
hopefully
newpick
can
pick
up
on
that
as
well.
So
what
we
did
first
was
we
tried
to
get
the
anomaly
so
this
upper
graph
here
is
the
anomaly,
and
this
is
just
the
x
direction
of
the
satellite,
so
the
satellite
oscillates
in
the
x
direction
and
over
here
at
the
very
end
we
injected
an
anomaly.
A
A
A
You
can
see
that
it
doesn't
tend
to
vary
that
much.
So
we
asked
new
pig
what
it
thought
the
satellite
would
look
like
it's
not
as
smooth,
but
it
got
it
kind
of
got
the
idea
like
it's
still
going
in
a
circle.
I
think
it
takes
a
nose
dive
at
some
point
on
this
path,
but
this
is
just
one.
This
is
like
90
minutes
worth
of
the
iss's
flight,
so
not
bad.
I
think
the
next
thing
we
would
do
is
try
and
get
as
much
data
like
to
be
rendered
here.
A
If
I
show
you
over
here,
we
taught
new
pick
with
like
a
100
or
800.
Actually,
no,
we
only
took
the
first
800
of
these
after
we
injected
the
anomaly.
We
actually
taught
it
over
like
ten
thousand,
so
he
taught
it
for
a
while,
and
then
we
just
took
a
one
snapshot
of
that
and
rendered
it
rendered
the
predictions
for
that.
A
C
I
don't
know,
but
one
we've
done,
we've
been
looking
at
these
sine
wave
patterns
a
lot
and
one
thing
is
because
it's
a
memory
based
system.
It's
you
know
it's
looking
at
the
patterns
and
trying
to
figure
out
what's
happening
next,
you
probably
quantize
the
sine
wave,
and
you
know
you
discretize.
It.
C
The
other
possible
thing
is
that
I
don't
know
if
you
swarmed
over
this
set.