►
From YouTube: Numenta Research Meeting - May 31, 2019 Part 1
Description
Orientation and object composition and reference frames.
Broadcasted live on Twitch -- Watch live at https://www.twitch.tv/rhyolight_
A
C
C
A
B
We
talked
about
orientation,
we're
talking
about
something
physical
that
is,
and
that
is
located
in
a
space
like
a
good
cell
space
or
XYZ
space,
and
at
some
point
where
this
thing
is
physically
located
and
then
at
that
point
any
movement
you
can
make
by
staying
on.
That
point
is
to
change
an
orientation,
so
I
I
think
so
so
you
can
say
that's
how
I
view
it
I
don't
know
if
that's
consistent,
I
think
it's
consistent
when
you're
different.
B
So
you
could
say
the
point
did
the
point
of
my
head
and
the
space
of
my
head
and
my
eye
is
at
one
point.
So
my
eye
doesn't
move
relative.
My
head,
it's
a
1-point
oath
of
my
head,
but
it's
changed
determine
moves.
It's
changed.
The
eyeball.
It's
changing
its
orientation
in
my
head,
I
would
qualify.
I
could
change
my
eyeball
relative
to
the
rule
I.
B
B
C
Yeah
I
mean
what
you're
saying
makes
sense.
I
guess
one
of
the
one
of
the
things
to
settle
on
or
consider
is:
do
we
ever
want
to
use
the
word
orientation
to
describe
this?
No,
no
okay
and
I
sort
of
I
sort
of
I
sort
of
tried
to
give
you
guys,
but
the
way
I
drew.
This
was
intended
to
try
to
make
you
well.
C
C
These
orientations
are
always
in
an
external
reference
frame,
but
they're
not
they're,
not
in
the
reference
frame
of
the
sensor.
Just
because
that
wouldn't
be
meaningless.
Yeah
yeah,
you
can
say
the
same
thing
about
the
location
of
the
sensor.
Location
of
the
sensor
is
always
an
external
reference
right
yeah.
They
have
known
their
tongue
now
kind
of
a
unifying
statement.
We
can
make
novel
about
movements,
movements
Kennedy
in
the
reference
frame
of
the
sensor.
C
C
B
B
C
B
B
I
think
I
think.
The
point
is
that
you
can't
a
sensor
cannot
move
when
it's
on
reference
frame.
You
can
say
I'm
going
to
move
to
a
point
that
was
there
now,
but
but
the
new
point
is
in
the
reference
in
the
room.
Is
that
the
three
feet
in
front
of
a
point
moves
with
me?
I
can't
move
I
can't
occupy
that
space
because
it's
always
running
a
movement
in
the
sensors.
B
C
B
First
copy,
as
in
the
sensors
reference
reference
copy,
says:
you're
gonna
move
this
it's
as
you're
gonna
move
in
certain
rated.
Some
do
these
motion
commands,
but
I
kind
of
be
movin.
It
sort
of
like
the
brain
doesn't
really
have
any
idea
what
it
is
they
purchase.
It
just
says:
there's
some
thing
that
I'm
gonna
send
out
and
it's
gonna
change
something,
and
so
it
doesn't
know
it's
walking
and
doesn't
do
a
90.
B
It
could
be
virtual,
like
you're
mathematicians
thinking,
so
the
idea
that
you
just
send
out
some
signals
is
it
and
then
what's
my
new
location,
that's
how
I
view
it
I,
don't
feel
like.
Oh
I'm
walking
three
feet
forward.
It's
just
like
I'm,
sending
out
a
signal
and
what's
my
new
location
in
this
front
or
what's
my
new
sensation
as
I
move
I'm,
not
sure
we
should
argue
about
this.
B
Point
location
support
is
not
really
useful,
calculating
anything
in
the
reference
frame
of
the
movements
in
the
reference
name
of
the
sensor.
The
whole
point
is
we
have
to
figure
out
where
the
sensor
is
going
to
be
after
it
moves
and
that
where
it's
going
to
be
has
to
be
in
a
referencing
of
something
else,
so
that
their
only
value
in
moving
is
to
know
your
new
location
in
some
other
reference
frame.
B
B
A
B
C
A
B
Viewpoint
into
this
three
feet
forward:
alright
I
get
I,
guess
I,
don't
even
think
I
have
to
convert
I,
just
say
I'm
going
to
do
something
and
the
result
of
what
I
do
is
going
to
be.
You
know
I
have
to
understand
this.
The
similarity
turn
this
one
on
that
one
I
don't
have
to
think
about.
It
is
an
allosteric
reference
frame.
It's
just
a
behavior,
whatever
baver
I'm
doing
is
going
to
result
in
something.
B
B
Here's
a
familiar
five.
Well,
it's
a
new
location
after
I.
Do
that
where
am
I
alright,
so
I
think
as
long
as
you
know,
suddenly
don't
develop
our
theories
about
that
using
talking
about
the
reference
frame
with
the
sensors,
as
only
focus
on
this
issue,
I
think
we're
fine.
We
can.
We
can
even
agree
to
disagree,
so
I
think
about
this
I,
don't
think
anything
impacts.
What
I
was
arguing.
B
I
was
arguing
that
in
both
these
cases,
the
two
left
panels
we're
talking
about
something
physical
thing
in
this
case
doesn't
matter
if
it's
the
eyeball
or
prism,
and
that
thing
has
a
location
and
orientation
relative
to
some
other
reference
track
and
orientation.
What
happens
is
what
happens
when
you
move
and
not
change
your
location?
B
A
A
B
Because
I
guess
this
before
and
you've
answered
it
before
and
I'm,
just
so
finding
confusing
you
know
this
work
you've
been
doing
recently
markets,
okay,
they're
trying
to
show
while
head
Direction
cells,
look
like
this
one-dimensional
thing
and
the
animal
can
move
in
three
dimensions.
So,
oh
well,
we
can
explain
their
operation
and
head
Direction
cells
by
having
some
common
sector
or
probably
right.
There
got
read
I'm
sure
yeah.
What's
your
preferred
time,
I
call
I
go
back
and
forth
between
gravity
vector
and
reference
vector.
Okay,.
B
That's
the
more
general
okay,
so
it's
a
reference
vector.
It's
still
not
clear
to
me.
That
seems
like
okay,
so
we
have
a.
We
have
a
we
happen
to
represent
3d
orientation.
Now,
head
Direction
cells
represent
one
of
those
dimensions,
and
so
now
there's
two
of
the
other
dimensions
are
still
sort
of
wrapped
up
this
reference
vector
and.
B
B
B
At
the
moment,
we
have
the
same
problem
with
grid
cells.
In
that
sense,
it's
also
two-dimensional
point,
and
actually
we
don't
even
quite
understand
how
they
do
that,
but
but
but
they're
clearly
two-dimensional,
and
so
the
question
is
one
I
did
get
three
dimensions
out
of
that
and,
and
then
I
was
saying
well
I
believe
that
it
seems
to
me
that
the
possibility
that
this
reference
factor
could
be
putting
a
roll
of
both
of
those
cases.
B
The
combination
of
a
reference
vector
and
so
I
started
thinking
like
or
what
happened.
The
grid
cells
as
the
animal
was
moving
over
the
the
ball
I
can
little
animations.
You
had
yeah,
okay,
imagine
a
plan,
so
the
head
Direction
itself.
Well,
you
can
always
imagine
head.
Direction
cells
are
already
based
on
sort
of
this
pole
work
to
the
bottom
to
the
top
of
the
animal
right.
So
groups
like
that
so.
B
Same
but
not
the
same
thing
out
the
grid
cells
I.
Don't
you
thought
about
that?
But
to
me
this
gets
there's
a
just,
a
key
key
problem
here:
it's
how
we
get
these
three-dimensional
locations
vector
is
the
three
dimensional
orientation
vectors
and
how
they
work
together,
because
we
don't
have
organs,
and
so
you
know,
maybe
the
best
thing
we
can
have
is
something
like
this
reference
vector.
That's
so
ties
them
together.
We
can't
order
them
to
talk
them
together.
C
B
C
Basically,
the
reference
vector
is
a
set
of
cells.
The
topologically
there's
a
bump
somewhere
on
the
surface
of
this
sphere,
there's
a
bump
of
activity
center
on
the
surface
of
the
sphere
yeah-
and
this
is
like
in
the
context
of
an
animal
and
with
gravity.
This
is
just
the
direction
of
gravity
or
the
direction
of
Earth
or
something
like
that
yeah
and
then
otherwise,
you
get
the
equivalent
of
head
Direction
cells,
which
is
a
bomb
somewhere
in
a
ring.
How.
B
B
Fold
this
is
a
two
dimensional
sheet
of
cells
right,
so
I
can
either
vision
on
the
two-dimensional
sheet
continue
trying
to
pinch
ourselves.
We
call
it
quits
on
module
and
and
what
you're
saying
is,
okay,
that
you
can
move
in
a
direction
on
that
and
the
wraparound
dr.
Torres
yeah.
It's
not
clear!
Why
that's
a
taurus
and
honest!
It's
not
totally
obvious
to
me
working
out,
but
why
would
this
look
like
if
I
slammed
it
out?
B
C
B
C
B
B
B
Well,
if
I
can
make
my
connections
any
way,
I
want,
then
I
couldn't
lay
this
out
right.
It's
sort
of
like
taking
a
map
of
the
earth
and
putting
it
on
one
of
those.
You
know
projections
it
gets
all
distorted,
but
it
doesn't
really
matter
as
long
as
I.
You
know
you
even
have
the
right
sort
of
connections.
At
every
point
you
could
say
your
your
spacing
can
get
wider
the
edge
it
goes
out
like
this,
so
you'd
have
to
learn
it
because
you're
right,
it
would
have
to
be
learned.
It's
a
bizarre.
B
A
B
Go
calculus
try
this
out
and
just
make
the
connections
work.
Then
you
could
put
the
cells
anywhere
in
the
brain
and
doesn't
matter
where
they
are
just
make
the
connections
work.
Doesn't
it
just
doesn't
strike
me
as
right?
It
gives
you
know
could
be,
but
I
don't
think
we're
getting
out
of.
There
is
something
else
going
about
feels
like
it
something
else
going
on.
B
Yeah
I
mean
auditioning
that
we
see
these
one-dimensional
head,
Direction
cells
everywhere
right
and
we
even
have
sort
of
a
penis
to
the
cortex.
You
can
think
of
it.
You
know
orientation,
so
maybe
I
don't
know,
but
we
know
one
never
reported
any
kind
of
you
know
cell.
Like
you
know,
some
properties
like
that
I
would
say
that
if
anything
they're
all
this
does
exist
and
it
has
been
reported.
C
B
B
B
Yeah
basic,
it's
it's
three
rings
that
are
filled
with
fluid
and
and
as
you
move
in
any
as
you
change
your
orientation
at
all,
the
fluid
moves
into
different
rings
differently,
depending
on
just
purely
the
plano
on
it's
only
a
solute
those
in
there.
So
it's
it's
literally
three,
two,
so
water
or
fluid.
So.
C
B
B
B
B
You
know
it's
funny
work
if
I
think
about
head
Direction
cells.
I,
don't
know
this
in
like
in
the
erotic,
cortex
or
some
book
killing
wherever
they
find
them,
that
they
actually
arrange
any
kind
of
ring,
or
you
know
if
I
think
about
orientation
cells
in
the
cortex
they're,
not
an
arrangement
right,
they're,
just
sort
of
a
continuous
transition.
B
It's
more
like
one
dimensional,
you
know
repeating
pattern.
Will
we
draw
them
like
this?
Is
there
any
evidence
that
actually
are
any
kind
of
I
mean?
How
would
we
think
the
real
head
direct
themselves
and
in
the
brain
look
like
I
mean
we
know
that
good
results,
like
you
could
say,
oh
like
that
any
paper
you
can
just
go
so
one
tell
over
once,
all
over
once
all
over
you
see
what
they
respond
to.
They
done.
B
B
C
B
It's
still
surprising
I
mean
you
think
would
be
something
someone
would
want
him
to
do.
This
I
mean
it
direct
from
cells
been
known
for
a
long
time
and
people
studying
people
still
studying
on
them.
Things
that
you
just
think
like
someone
who
discovered
that
I
mean
it's
a
clue
right,
it's
a
clue
to
how
the
whole
system
works.
If
you
know
how
its
arranged
just
like
with
grid
cells,
it
to
be
clue
what
you
see,
how
they
actually
organs
down,
it's
a
important
part
of
the
puzzle.
B
B
C
Yeah
I
mean
a
couple:
they
they're
hard
to
interpret
it,
but
yeah
I
would
say
the
same
person.
Cape
Jeffrey's
lab
is
specializing
in
that
with
rats.
Various
types
of
that
and
another
lab
that
all
in
offski
lab
is
doing
it
with
bats,
and
so
the
Jeffrey
results
can
be
divided
into
ones
where
ya
rise,
walking
at
various
incline
planes
or
finding
at
walls
or
a
set
of
results
where
the
rat
is
actually
walking
through,
like
basically
a
jungle
gym
type
like
big
lattice
baby.
C
You
could
look
at
it
and
interpret
it
in
a
lot
of
ways.
The
grid
is
sort
of
messed
up
by
walking
diagonally
it's.
If
you
look
at
the
pictures
of
the
firing
fields,
it's
hard
to
form
conclusions,
they
sit
there.
Still
this
kind
of
hexagonal
they're
not
really
cleanly
hexagonal
with
what
was
going
on
on
the
floor.
So
it
changes.
Yeah,
yes,
but
really,
if
you
go
in
with
a
bias
like
you
expected,
the
cities
would
be
the
same.
You're
gonna
conclude
it's
about
the
same.
B
Gonna
be
different.
It's
hard
to
reach
a
conclusion,
at
least
from
looking
at
the
pictures
later.
One
game,
I
just
was
trying
to
use
that
as
an
analogy.
I
said:
oh
good,
so
for
the
same
said
directions
then
I
can
say:
okay!
Well,
then
they're
only
going
to
change
when
the
animal
is
moving
in
its
plaintive
body
plane
or
like
wherever
planets
on
its
moving.
So
that
would.
B
B
So,
even
if
its
orientation
changes,
but
it's
it's
personally,
not
rotating
around
its
own
pole,
then
that's
the
key
thing.
So
then
the
analogy
here:
ok,
ok,
well,
the
pitch!
Those
are
all
the
change.
Am
I
walking
forward
in
my
plane
or
walking
backwards,
or
you
know
whatever,
but
therefore
what
I
could
be
moving
through
the
room
at
some
at
some
angle
or
something
like
that
and
that
wouldn't
change?
That's
weird!
It's
it's!
Just
like
the
grits,
the
head
Direction
cells,
a
little
bit
weird.
B
B
A
C
Then
a
slightly
similar
set
of
results
rather
than
another
one
where
the
RAM
is
still
moving
on
a
plane,
but
this
time
the
plane
is
the
wall,
so
I'm
not
going
to
have
full
3d
movement,
yet
I,
don't
know
the
best
way
to
draw
this.
Let's
let's
say
there
are
two
two
areas
for
the
rat
to
walk.
This
is
the
wall.
This
is
the
floor.
There's
a
90
degree
angle
here,
yeah
they're,
a
good
cell
phone
is,
like
you
know,
hexagonal
fields
on
the
floor,
just
call
that
a
hex
gun
on
the
wall.
C
Field-
it's
not
it,
but
if
some
of
the
in
their
other
older
results
where
the
rats
were
less
experienced
with
all
this
climbing
and
stuff,
often
they
weren't
fields,
they
were
big
pillars.
They
were
big,
they
were
big,
they
looked
the
whole
whole
section
of
wall,
we'd
had
a
good
sell
firing
and
another
one
where
if
the
round
was
climbing
up,
it's
hard
to
I
was
going
to
draw.
If
the
wall
has
a
set
of
pegs
coming
out
of
it
mm-hmm
yeah,
then
they
found
that
the
fields
were
something
like
this.
B
That
there
might
be
a
cell
over
the
whole
wall
is
something
treating
that's
kind
of
what
I
was
getting
at
here.
Imagine
again,
I'm
following
the
analogy
of
the
head
directs
itself
stuff,
you
could
report
it
so
and,
and
so
the
head
Direction
cells
are,
you
know
they
asked
weirdly
in
three
dimensions.
That's
that's
one
of
the
things.
They
only
look
normal
in
two-dimensional
space,
but
the
real
behavior
is
weirder
than
that.
So
it's
like
okay,
let's
not
get
hung
up
on
the
normal.
B
B
B
We
now
know
the
cells
are
weird
they're,
not
like
we
thought
they
were.
The
way
you
think
of
they
are
is
a
is
a
projection
from
three-dimensional
space
of
some
sort,
and
then
maybe
the
green
cells
are
also
weird
but
not
like.
We
think
they
are
and
they're
also
maybe
the
projection
from
T
domestic
space,
and
maybe
you
have
to
look
at
grid
cells
and
orientation
cells
at
the
same
time
to
know
anything
if
it's
like,
maybe
I'm,
just
making
this
up.
But
this
is
how
admit
that
it's
like?
B
B
C
Trying
third
set
of
results
that
made
us
most
uncomfortable
us
being
me
and
Mirko
and
ela
is
now
if
the
rat
is
climbing
through
like
3d
space,
because
it's
climbing
through
basically
this
big
toy
structure
and
I'm
not
going
to
be
able
to
draw
yeah
yeah
lots
of
these
interconnections
anyway,
with
the
constant
lattice,
the
animal
can
walk
in
basically
different,
three
or
four
different
directions.
From
any
point
there,
the
grin
cells
have
these
3d
firing
fields.
These
3d
blobs
various
points
in
space.
C
B
That
so
you
know,
so
that
brings
up
a
thought,
maybe
the
whole
point
of
the
grid
cells.
In
my
mind,
the
entire
point
of
them
is
that
you
need
to
be
able
to
do
a
map,
and
the
point
of
the
map
is
a
particular
behavior.
When
you
actually
could
behavior,
you
need
to
know
the
next.
The
next
point
of
map-
that's
all.
It
really
is
my
my
thing
about
the
uniform
field.
B
B
B
B
And
I
would
still,
but
I
could
any
point
of
that
I
can
look
around
see
the
entire
room,
and
so
for
me
to
make
predictions
about
the
entire
room.
I
have
to
know
my
location
in
a
very
linear
of
3d
space.
The
route
may
not
be
like
that.
You
know
it's
poor
vision.
It's
got
its
whiskers,
and
so
it
may
not
be
able
to
look
at
the
entire
environment
and
look
around.
B
B
Yeah,
it's
a
complex
set
of
hallways
to
get
out
right.
Well,
I'm,
going
through
those
hallways
I,
actually
don't
know
where
I
am
in
this
building.
Yes,
some
generally
D,
but
but
I've
lost
the
back
of
the
building.
So
all
I
really
learned
is
oh
there's
a
hallway.
I
know
this
hallway
I
need
to
get
to
that
point
in
this
way.
Oh
now,
there's
that
all
what
I
need,
so
the
rat
could
be
like
that
rat
climbing
around
on
these
little
things
might
be
like
me,
going.
B
It's
like
I,
don't
really
know
where
I
am
in
the
building.
I
can't
say
what's
above
me,
that's
a
good
analogy,
so
we
have
to
be
little
bit
careful
about
that.
We're
a
human
if
I
was
suspended
to
the
point
to
this
room,
I
would
and
therefore
I
expect.
My
yeah
I
would
expect
not
to
have
a
blobby
or
on
grid
cells,
good
cells,
but
we.
B
B
C
B
C
B
B
B
B
Can
just
say
that's
one
more
thing:
yep
is
the
question.
So
I
was
working
through
this
sort
of
the
composition,
optic
problem
here
and
and
saying
something
we
have
no
origins
in
the
origins,
but
I
have
some.
This
is
all
Gd
I
have
a
I
have
a
reference
frame
for
this
object,
I'm
rushing
straight
for
this
object.
B
B
So,
like
a
movement
in
this
direction,
here
changes
things
differently
than
in
here,
but
the
question
I
have
for
you
is
the
following:
if
I,
if
I
could
tell,
if
I
can
figure
out,
the
orientation
of
the
freshmen
came
to
this
rep
from
this
being
room,
I
could
figure
out
the
orientation
a
to
to
the
parent
object
and
I
had
something
of
similar
July
for
displacement
cells
which
don't
work
at
this
situation.
Is
that
sufficient.
C
B
B
B
C
B
I'm
at
this
point,
I
can
say
what
point
I
went
in
P
and
what
point
am
I
right
and
if
I,
if
I,
knew
that
and
I
know
the
orientation
that
that
seems
to
be
sufficient,
because
then
what
I
could
do
then
I
could
just
say:
okay,
let's
take
this
I
think
my
my
old
screen
here
is
okay,
this
porn
that
screen
and
this
point
of
the
screen
of
every
Co
lined
and
now
I
have
to
rotate
it
like
that.
I'm
done,
I
specified
the
position
of
these.
That
would
be
sufficiently
if
I
could
use.
B
B
B
Position,
integration
and
I
start
moving
and
I
know
my
orientation
says
a
to
P
different
orientations:
hey
I
could
continually
update
where
I
am
I
would
know
where
I
am
on
and
where
I
am
and
P
there's
no
problem.
The
problem
we
have
is
defining
the
relationship
being
without
specifying
a
point
that
seems
to
be
the
problem
like
it's
a
fact.
If
anything
down
this
becomes
the
origin,
then
it's
like
okay
I'm,
going
to
locate
angle
into
the
peopIe
you
specify
at
this
point.
That
point
is
a
common
point.
B
Out
myself,
but
but
that's
not
what
we
do,
I
mean
I
I
have
to
be
able
to
so
to
say:
okay
right
now,
I'm
over
here
on
P
and
oh,
let
me
look
up.
Where
would
I
be
on
a
and
I
can't
look
that
up
I
can't
I
can't
we
don't
know
how
to
like
I
told
you
this
common
point.
I
can
navigate
from
here
or,
as
you
put
two
points
like
an
every
day,
know
that
but
I
can't
just
say:
okay,
how
would
I
represent?
B
How
are
you
represent
a
if
I,
how
could
I
represent
a
Belgians
P
if
I
just
picked
some
random
point
P
and
say:
where
does
it
do
that?
That's
the
problem,
so
the
problem
is
sufficiently.
We
have
enough
information
here.
I
can
specify
a
point
of
two
points
at
point
orientation,
two
points,
but
that's
not
what
I
want,
because
I
don't
want
to
figure
out
where
I
am
I'm
over
here.
A
A
A
B
B
A
B
B
B
B
Some
generalization
here,
right,
I,
would
be
able
to
say
I,
don't
have
to
learn
the
transform
from
every
point
here
every
point
here.
If
that's
what
you're
suggesting
okay
man,
that
your
work
but
I
suddenly
don't
want
to
learn
that
I
have
to
somehow
generalize,
say:
here's
a
simple
representation
for
the
position
of
this.
For
this
and
now
I
can
go
from
this
neuron
to
that
and
a
generalized
by
I.
Guess
if
I
learned
it
exhaustively,
I
get
it
I
can
see
how
it
would
work,
but
we
don't
want
to
learn.
C
C
B
A
B
A
B
Orientation
all
right,
that's
what
we
were
trying
to
get
to
a
displacement.
Positive
orientation
is
pretty
simple
and
can
easily
be
imagined.
So
imagine,
I
have
a
special
on
lines
of
orientation.
So
now
I
have
a
set
of
cells
where
what
self-active
is
basically
representing
the
rope.
The
relative
rotation
of
these
guys?
Okay,
so
I-
have
that
let's
say
Heather
so
I
have
a
cell
I
mean
I,
have
ten
different
orientations,
I
attended
themselves,
it
can
be
active.
It
says:
oh
I,.
A
B
B
A
C
I
mean
basically
I
kept
running
with
the
game
cell
trick:
yeah,
hey,
let's
connect
grid
cells
that
fire
together,
yeah
and
when
they
fire
together
in
this
reference
frame
and
that
reference
frame
I
could
say
this
much
better,
but
I
extrapolated
this
idea
to
use
the
same
exact
trick
for
orientation,
but
it's
limited
and
that
be
mini.
But.
B
B
B
Stick
with
one
displacement:
module
and
one
when
the
space
is
practitioner,
one
displacement,
location,
lock
and
that's
it
I'm
not
worrying
about
the
background
and
say:
okay
I'm,
just
gonna
use
those
one
nodules
and
and
now
I'm
going
to
try
to
figure
out
my
location
and
orientation
using
those
tomorrow.
Yeah.
B
B
A
C
This
whole
60
degree
thing
doesn't
depend
on
how
many
cells
there
are
here
like.
If,
if
suppose,
you
were
actually
representing
the
actual
coordinates
of
a
bump
like
in
the
floating
point,
if
you
could
do
that
and
get
the
actual
coordinates
of
like
a
bump
in
their
orientation
floating
point,
yeah
a
60
degree
thing
will
stop
the
the
limitation.
B
C
C
C
B
B
And
then
one
reason
they
didn't
discover
a
right
away,
because
the
fields
are
too
far
apart.
The
propeller
predator-filled
was
big,
and,
and
so
they
didn't
had
to
go
to
a
very
big
environment
before
they
saw
them
said.
I
was,
in
my
mind,
thinking
about
this
problem
like
this.
Let's
say
that
I
could
fit
all
in
one
rhombus
area
so
that
there's
no
repeating
cells
in
this
room.
You
know
there's
no
repeating
locations.
This
is
one
cell
active
and
each
point
in
this
blob
of
cells
here
and
nothing.
B
Anyone
else,
because
that's
because
they're
repeating
promises,
that's
how
I
was
starting
to
think
about
this,
and
in
that
case
then
he's
just
at
any
point
in
time.
Any
point
this
space
there's
a
bunch
of
cells
active,
but
there
is
no
60
degrees
because
I
don't
get
to
the
next
firing
field.
You
know
I'm
saying
it's
just
that's
what
I
was
thinking
about
it
just
on
the
blog
ourselves.
He's
bought
just
the
only
response
in
this
area
and
I'm
selling
sponsor
this
area,
and
this
area
happened
in.
C
B
A
A
B
A
B
That's
it,
but
that's
not
that
much
now
what
if,
okay,
this
is
another
way,
there's
a
weapon.
What
have
we
had
to
have
a
pretty
coarse
coating
here,
so
I
try
to
do
some
and
I'm
just
going
through
a
thought,
experiment
and
now.
I'm
gonna
have
a
whole
bunch
of
modules,
but
I'm
still
no
tiling
in
any
module.
I'm
still
gonna
be
my
modules
are
gonna,
be
you
know,
I'm
going
to
go
to
an
extra
could
sell
the
orientation
module
orientation
module,
but
never
I'm
still
not
doing
any
sort
of
repetitive
fields
with
you.
B
B
B
A
B
B
B
B
A
B
B
B
Well,
this
is
an
interesting
idea.
There's
a
Moroccan.
This
is
actually
another
competent
there's
another
possibility.
I've
considered
it
a
lot.
The
idea
they
or
direction
cells
might
have
the
same
trick
that
we
think
itself.
That
is
that
you
could
have
a
whole
bunch
of
different
small
head
Direction
modules,
because
the
directions
also
seem
to
be
all
over
the
place.
They
they
see,
there's
lots
of
them,
so
maybe
a
whole
bunch
of
different
modules.
Each
one
is
anchored
differently
on
a
particular
environment,
let's
say
by
sensor
infant.
B
So
now,
if
you
ran
out,
if
you
read
out
bunch
of
cells
from
a
bunch
of
these
little
modules,
what
you'd
end
up
with
is
a
representation
of
orientation.
That's
specific
to
any
mine
would
be
this
is
it
would
exhibit
path
integration,
but
it
also
say
this:
is
your
orientation
in
this
particular
environment
and
I
know
how
to
define
X
orientation
but
I
would
also
know
the
environment.
I
know
what
rule
are
yeah,
those
guys
anchor
it
differently,
so
you
couldn't
have.
B
B
B
And
I
don't
have
to
do
a
lot
of
repeats
in
here,
so
they
did
a
small.
You
know,
could
I
get
this
to
work
and
I
by
combining
the
unique
orientation
and
you
credit
cells
and
lots
of
them
can
I
get
it
to
work
in
some
way.
You
would
be
able
to
integrate
correctly
until
you
recognize,
which
which
models
would
not
work.