►
From YouTube: IETF-T2TRG-20220627-1300
Description
T2TRG meeting session at IETF
2022/06/27 1300
https://datatracker.ietf.org/meeting//proceedings/
A
A
C
D
C
A
A
A
The
second
main
topic
of
today
would
be
data
model
versioning.
So
that's
an
activity.
We
have
started
quite
some
time
ago
when
we
did
some
work
on
the
previous
weekly
call,
so
carson
will
be
giving
an
update
on
that
and
then
finally,
the
maybe
the
most
substantial
item
we
have
planned
for
today
is
this
work
on
sdf
extensions
for
relations
and
links.
A
So
this
is
a
bigger
work
item
that
we
have
been
kind
of
discussing
around
on
the
sdf
work,
but
while
the
sdf
work
has
been
very
much
focusing
on
finalizing
the
current
upcoming
rfc,
these
links
and
relations
have
been
something
kind
of
for
future
work.
So
that's
why
we
have
been
discussing
these
items
in
the
wishy
group
and
there
we
have
some
recent
development.
Vettel
larry
submitted
a
a
new
draft
on
that
mile
coaster
has
been
doing
quite
a
bit
work
on
that
topic
and
so
have
a
cursing
eye.
A
B
A
Okay,
thanks
carson.
So
if
there's
no
more
question
comments,
I
remember
jose
was
planning
to
join
and
she
was
actually
interested
on
the
first
two
agenda
items.
So
what
I
think
this
was
important
and
interested
on
the
first
agenda
item,
so
what
I
maybe
would
suggest
if
we
start
with
the
data
versioning
and
then
we
go
to
a
iot
standards
description.
B
Okay,
I
will
start
with
a
very
beautiful
set
of
slides
that
you
can
see
on
your
screens
and
if
you
actually
want
to
see
more,
then
you
have
to
go
to
the
top
left
corner.
Where
you
see
a
loop
and
you
can
click
on
the
loop
and
then
you
see
the
slide.
I
mean
it's
not
really
a
slide,
it's
just
the
wiki
page
and
you
can
scroll
through
that
and
and
see
what
I'm
talking
about.
B
Yet
so
I
would
start
with
that,
and
then
there
are
several
items
under
which
there
are
very
brief
nodes
that
we
probably
have
to
expand
to
turn
this
into
a
document
at
some
point,
but
right
now
we
we
just
do
it
at
the
wiki
level
and
there
are
first
of
all
a
set
of
reasons
for
versioning.
So
why
would
you
even
want
to
do
versioning?
What's
the
need,
and
so
this
is
about
versioning
of
the
data
models.
B
So,
for
instance,
the
the
requirements
for
models
might
change,
assume
that
there
is
some
new
regulation
going
in
effect
in
in
the
eu.
That
requires
all
iot
refrigerator.
To
do
something
specific,
then
you,
you
need
to
put
this
into
the
models
for
a
refrigerator,
and
then
you
have
to
be
able
to
create
a
new
version
of
the
model
or
a
new
model.
That's
really
just
the
same
thing
with
a
big
difference,
so
that's
one
one
of
the
considerations
we
have
to
think
about.
B
B
B
So
if,
if
you
have
a
refrigerator
and
and
freezer
and
you
are
applying
a
command
to
the
the
refrigerator
that
that
really
only
makes
sense
for
a
freezer,
then
some
something
weird
is
going
to
happen,
and
you
don't
want
that
to
be
possible.
So
you
have
to
limit
interability
but
of
course,
except
from
when
you
actually
need
to
do
that,
you
want
to
get
as
much
enterability
as
possible.
B
So
this
is
notes
we
we
have
copied
from
previous
discussions
and
the
the
point
today
was
mainly
to
point
to
this
document
and
maybe
get
some
discussion
going
here
on
the
wiki
page
by
by
adding
stuff
or
on
the
mailing
list.
If
there
really
needs
to
be
a
discussion
before
we
can
add
things-
and
we
want
to
develop
this
in
in
such
a
way
that
we
understand
that
data
model
versioning
enough
to
address
this
in
the
next
version
of
sdf.
B
E
Yeah,
I
have
a
question
more
general
question.
Carsten.
I
took
the
the
agenda
item
to
mean,
in
general
consideration
of
things
that
we
may
want
to
record
on
wishy
wiki
site,
and
I
don't
know
if
there
is
a
place
to
try
to
put
the
definitions,
as
we
were
discussing
recently
of
edge
and,
let's
say,
of
digital
twins,
to
at
least
to
put
a
few
paragraphs
together
as
a
consensus
of
our
current
understanding
of
what
those
things
mean
so
as
to
improve
the
understanding
and
hopefully,
consistency
in
using
the
terms
or
is
that
to
anyways.
B
B
A
B
B
A
E
E
I
mean
it
included
some
edits
that
I
got
from
uari
and
a
few
others
and
as
far
as
I'm
concerned,
it's
ready
to
publish,
I
posted
it
on
both
the
xml
and
pdf
versions
on
on
the
wiki
on
this
zone
that
carson
created.
So
if
somebody
wants
to
look
at
it's
on
the
github
and
the
document
title
this
morning,
I
had
a
coffee
thought:
the
official
title
of
the
paper
in
the
paper
itself.
E
I
have
also
changed
the
file
name
because
that's
what
shows
up
when
it's
posted-
and
I
and
it's
now-
draft
ttrg
iot
standards,
description,
version,
zero,
zero
and,
as
I
said,
I
render
checks
the
needs
and
stuff
like
that.
So
it's
wise,
I'm
concerned
it's
ready
to
submit
pending
the
approval
of
the
group
today.
A
A
So
maybe
a
question
of
the
group
has
well,
I
guess
many
of
you
have
not
seen
the
document
since
it's
not
in
the
data
tracker
yet,
but
if
you
had
a
chance
to
look
at
the
github
version,
do
you
have
any
comments
on
anything
to
be
tweaked
or
adjusted
before
we
publish
the
zero
zero
version
and
actually
milan?
Now
I
realized
one
small
thing
on
the
draft
name,
since
this
is
still
an
individual
person,
not
a
research
group
adopted
person,
you
could
add
after
draft.
You
know
your
own
last
name.
E
A
F
A
Okay,
here
in
london,
I
think
we
have
a
good
amount
of
time
for
the
final
topic.
So
thanks
a
lot
milan
yeah,
please
submit
the
draft.
If
you
want
you
can
you
can
check
with
the
file
name
and
those
issues
with
me
are
offline,
but
then,
once
you
have
submitted
the
draft,
you
get
the
notification
to
your
email
about
the
submission.
Then
you
can
forward
that
notification
to
the
rg
list
and
say,
like
you
know,
give
a
bit
of
background.
A
A
B
B
So
a
link
points
from
something
that
that
is
here
from
the
point
where
the
link
points
from
which
is
either
implicit
or
ex
made
explicit
by
using
an
anchor
to
there,
and
there
is
a
target
and
kind
of
the
object
of
the
link.
And
the
predicate
of
the
link
is
the
link
relation.
So
it
says
what
the
relationship
between
the
subject
and
and
the
object
is-
and
typically
we
also
can
attach
target
attributes
to
a
link.
B
So
we
can
say
a
little
bit
more
about
what
that
is,
that
you
will
actually
reach
when,
following
that
link.
So
you
might
have
several
links
that
just
this
differ
in
their
content
format,
or
something
like
that.
So
you
could
use
those
target
attributes
to
find
out
which
links
you
actually
want
to
follow.
B
So
that's
a
general
observation
about
links
and
links
have
been
in
use
in
this
way
for
for
quarter
century
now
and
more
so.
None
of
this
is
new
and
actually
is
a
standard
rfc808h
that
we
are
using,
which
is
called
web,
linking
that
that
is
underlying
much
of
what
we
are
doing
with
links
in
in
data
formats.
B
So
in
sdf
right
now
we
have
two
forms
of
that
and
I'm
pretty
sure
we
need
more
forms.
But
that's
what
we
have.
We
have
inclusion
by
reference
and
I
should
have
said
syntactic
inclusion
by
reference,
because
that's
really
what
we
have
right
now.
There
might
be
other
forms
of
inclusion,
the
sdf
ref
quality,
and
we
also
use
links
for
internal
pointers
within
the
model
tree
where
we
have,
for
instance,
sdf
required
petri.
H
B
At
least
great
okay,
so
that's
one
one
kind
of
link
where
the
the
fact
that
we
are
using
a
link
is
just
a
way
we
bring
together
together
the
various
parts
of
the
model.
How
we
talk
about
one
part
of
a
model
in
another
part
of
a
mod.
So
right
now
we
have
sdf
required
as
the
the
only
form
of
an
internal
pointer.
We
used
to
have
more
of
those,
but
we
were
able
to
get
rid
of
that,
but,
of
course,
yeah
the
the
way.
B
B
That
way,
so
you
actually
can
follow
the
link.
You
don't
follow
it
in
in
the
normal
course
of
things,
but
if
you're
building
software
that
that
uses
such
an
identifier,
you
may
want
to
reference
it
to
pull
some
some
information
about
those
identifiers
and
general
semantic
links,
things
that
we
would
be
doing
in
rdf.
These
are
also
related
to
these.
B
So
you
might
point
to
to
an
instance.
That
is
another
thing
or
the
the
affordance
of
a
thing.
So
you
might
be
able
to
point
into
a
thing
in
in
some
way
you
might
be
interchanging
these
links
and,
of
course,
you
might
still
be
using
some
vocabulary
while
doing
that.
So,
for
instance,
you
might
have
a
link
to
a
thing,
but
then
you
have
a
fragment
identifier
that
actually
is
taken
from
a
vocabulary,
so
there
may
even
be
some
some
construction
or
parsing
of
these
links.
B
B
B
And
finally,
we
have
links,
in
instance,
data
again
where,
where
we
really
want
to
model
what
kind
of
link
this
is
and
again,
if
you
have
a
refrigerator,
it
really
doesn't
make
a
lot
of
sense
to
to
use
a
link
that
is
meant
to
pair
it
with
a
freezer
to
instead
connect
it
to
the
garden
hose
or
something.
So
again,
we
probably
need
some
type
information
here,
and
so
this
may
just
be
data
or
maybe
actions
or
it
may
be
entire
things
or
sdf
objects.
B
B
Okay
and
of
course,
it
now
becomes
interesting
when
we
think
about
the
fact
that
that
sdf
really
was
designed
to
enable
translation.
So
you
have
some
ecosystem
specific
model
information,
and
you
want
to
be
able
to
translate
this
into
model
information
at
the
sdf
level.
So
this
means
you
need
to
be
able
to
translate
back
and
forth
and,
of
course,
every
ecosystem
environment
has
their
own
form
to
talk
about
links.
So
we
know
how
ipso
does
this
and
so
on.
B
Okay,
so
that
that's
essentially
what
I
I
wanted
to
point
out
and
of
course,
at
some
point
we
will
need
to
understand
the
processing
model
that
we
are
using
for.
Actually
operating
on
these
links,
we
need
to
have
good
examples
for
the
various
cases
and
and
how
they
differ
or
how
they
they
actually
may
be
similar
in
some
cases,
and
we
we
need
to
use
these
to
actually
understand
what
needs
to
be
in
the
model
or
in
the
meta
model.
So
what?
What
are
we
making
possible
to
put
into
the
model?
A
A
A
H
When
I
see
the
slides-
oh
yeah,
perfect,
it
works
because
I
have
this
audio
button
in
between
in
the
middle
of
my
screen,
but
it
seems
that
I
can
change
the
slides,
perfect
okay.
So
let's
go
here
so
we
have
been
working
with
this
sdf
relation
extension
to
this
sdf
and
basically
we
have
been
working
with
the
the
dtdl
relationship
and
trying
to
make
that
kind
of
similar
mapping
to
sdf.
H
H
The
draft
has
one
initial
design
and
there
are
quite
many
different
kinds
of
ways
to
use
the
relation
or
define
it
in
in
in
the
class
level,
and
one
thing
here
that
came
up
was
that,
okay,
so
basically
we
want
to
do
a
class
level
design
on
that,
but
in
some
sense
it
can
even
be
thought
about
instance.
So
it
must
be
something
between
maybe
class
kind
type
instance
there,
it's
somewhere
somewhere
in
the
between
possibly.
H
H
It
has
some
properties
with
sdf
property,
and
then
we
have
the
sdf
relation
field
here.
So
in
the
first
we
name
it
as
same
as
and
basically
defines
that
the
relation
type
is
it's
a
same
type
as
something
else,
and
here
we
make
the
target
to
another
ontology
where
we
have
this
temperature
sensor
defined.
So
basically,
we
have
similar
setup
here
in
this
system
and
the
second
relation
we
have
it's
contained
in.
So
basically
the
type
is
contained
in
and
the
target
is
the
room.
H
So
that
is
the
basic
basic
design
we
have
defined
here,
the
sdf
qualities
we
have
release
and
type.
What
kind
of
relation
we
have.
Then
we
have
a
target,
and
that
is
the
target
to
which
we
refer
to.
We
have
description.
Field
gives
the
basic
description
of
that
relation.
We
have
taken
here
max
items
and
mean
items
which
are
defined
in
in
the
dtdl
property
properties.
H
E
Has
this
low
question,
why
not
did
you
guys
explicitly
choose
to
avoid
it?
It's
quite
common
in
standards
like
haystacks
and
the
other
just
for
convenient
on
managing,
basically
operators
consoles
to
use
the
familiar
names
or
the
names
that
they
can
define
that
the
standard
otherwise
ignores
it.
Basically,
the
system
doesn't
interpret.
So
is
there
a
particular
reason
you
chose
not
to
have
that
in.
H
H
C
A
Yeah
yeah,
I
think,
like
a
perhaps
a
key
thing
here
would
be
like
how
do
we
handle
internationalization
in
sdf,
because
I
guess
that's
something
we
haven't
designed
yet,
and
I
guess
once
once
we
do
that
then,
whether
it's
a
new
quality
or
it's
a
different
way
to
use
the
existing
qualities
probably
would
be
the
way
to
go.
But
I
guess
that's
that's
a
full
sdf
level
discussion
and
not
just
for
the
relationship,
but
a
good
point.
H
Then
we
have
properties
that
can
be
defined
as
sdf
property.
The
one
thing
that
came
out
here
is
that
there
was
not
examples
or
any
existing
dtdl
files
that
could
have
been
used
so
that
we
could
have
made
better
mapping
and
testing
it,
how
it
actually
works,
unfortunately,
and
but
anyway,
this
is
quite
direct
mapping
between
the
sdf
quality
and
and
dtdl
qualities.
H
But
that
that
was
the
basic
design,
what
we
have
been
doing
here,
so
it's
a
relatively
short
draft,
but
then
we
have
this.
Sdf
relation
can
be
defined
different
ways
so,
and
we
had
some
alternatives
also
during
the
work
and
one
interesting
that
came
up
the
last
last
minute
was
how
about
defining
the
relation,
as
in
this
present
that
example.
H
This
is
in
conceptual
way.
This
is
a
little
bit
different
that
what
the
sdf
is
usually
using,
but
it
has,
I
guess,
pros
and
cons
and
how
to
do
it,
but
I
guess
the
most
important
thing
would
be
to
get
the
input.
How
do
people
see
what
kind
of
how
we
should
define
the
relation?
So
basically
we
have
one
one
alternative
now
in
the
draft
and
is
it
okay
or
not
what
should
be
fixed
and
so
on.
I
I
We
all
understand
for
links
that
karsten
articulated
earlier,
which
is
there's
a
anchor
and
a
target,
or
some
source
and
destination,
subject
and
object,
and
then
there's
a
predicate
that
consists
of
a
relation
and
then
additionally,
there's
target
attributes.
I
think
that
the
latest
link
draft
actually
says
well.
I
So
I
would,
I
would
say
that
I
I'm
definitely
in
favor
of
what's
in
there,
for
I've
changed
my
example
link
value,
but
also
target
attributes
or
link
attributes.
If
we
wanted
to
align
with
the
terminology
and
the
in
the
link
trapped,
I
think
that
would
be
okay.
It
would,
it
would
mean
the
same
thing.
It
would
just
have
a
different
name
that
would
be
more
aligned
with
the
the
web,
linking.
I
think
it
does
make
sense
to
align
with
web,
linking
especially
that
it
sort
of
seems
to
support
everything.
I
I
have
some
that
anyway,
I
don't
know
how
much
time
we
have
I'd
like
other
people,
to
be
able
to
make
comments,
and
I
I
have
some
material
I'd
like
to
share
as
well,
but
I
think
other
people's
comments
and
impressions
are
more
important.
So
I
think
we
only
have
a
few
minutes
left
anyway.
Those
are
my
comments.
Thank
you.
C
I
Until
they
have
a
system
thing
that
encloses
the
sub
system,
things
and
all
that
through
sdf
reps,
there's
too
many
layers
of
inheritance
in
the
model
right.
Everyone
would
agree
with
with
that,
as
the
model
gets
to
some
level
of
complexity,
and
I
don't
think
we
should
really
limit
the
the
level
of
complexity
that
sdf
provides
for.
So
I
think
it's
important
to
be
able
to
say
here
are
my
building
blocks
for
system
models
and
at
different
times
right.
You
know.
I
I
This
object
connects
to
this
other
object
in
this
case,
so
you
might
want
to
have
to
say
that
this
thing
is
part
of
another
thing,
using
a
semantic
link
as
opposed
to
making
it
part
of
it
using
stf
breath
you
know,
and
so
you
can
have
a
loose,
a
more
loose,
loosely
coupled
sort
of
relationship
that
still
implies
some
kind
of,
or
you
can
say
this
thing
is
connected
to
this
other
thing,
or
you
could
say
this
this
this.
This
object
acts
on
some
properties
of
this
other
object.
I
So
basically,
what
I'm
saying
is
you
make
an
rdf-like
graph,
but
basically
is
how
you
construct
the
system
and
then,
when
you
go
to
make
a
implementation
of
the
system,
whether
you're
running
it
out
in
python
code
or
c
plus
plus
or
you
know,
a
database
schema
where
you're
going
to
make
rows
of
a
data
in
a
database
to
represent
all
this
stuff
for
a
graph
database
where
you're
going
to
make
notes,
etc
right.
So
how
are
you
going
to
implement
it?
I
The
system
model
has
all
the
information
you
need
in
sdf
to
implement
it
so
use
all
the
sdf
tools
to
sort
of
take
all
these
linkages
and
do
whatever
you
need
to
in
the
target
system.
In
my
target
system,
I
use
lightweight
mtm
objects,
and
so
the
target
system
just
uses
an
object
link
to
connect
one
to
the
other,
and
then
the
objects
are
so
simply
designed
that
that
you
know
they
know
what
to
do
when
they're
linked.
I
I
E
I
Say
much
about
it:
it's
just
how
the
zones
connect
and
what
the
surfaces
look
like
and
how
the
surfaces
connect
to
each
other,
but
but
but
those
are
all
described
in
other
pieces
of
the
model.
The
the
model
of
the
floor
is
mainly
just
a
thing
that
holds
other
things
like
a
container,
and
so
I
want
to
be
able
to
build
models
like
that
using
stf.
I
A
Okay,
so
what
I'm
hearing
like
what
you're
saying
is
like
yeah?
We
need
this
kind
of
capability
where
we
can
define
relations
beyond
the
simple
hierarchy
that
we
can
do
with
sdf
today,
and
then
there
are
a
lot
of
nuances.
They're
like
what
are
you
actually
modeling,
if
it's
kind
of
semantic
relation
or
or
something
else-
and
I
guess
that's-
that's
kind
of
interesting
part
like
do
we
use.
A
Do
we
use
sdf
relationships,
let's
say
whatever
we
shall
call
the
keyword.
You
use
the
same
thing
for
all
of
these
use
cases,
or
is
there
need
for
multiple
different
kind
of
links
like
going
back
to
what?
What
carson
pointed
out
like
in
terminology
should
the
instance
links
instances
of
of
data?
Should
that
use
the
same
construct
or
not?
A
And
to
be
honest,
I
I
don't
have
a
good
answer
yet.
So,
of
course
I
see
for
the
instance
of
data.
I
see
two
different
methods
to
model
that
one
would
be
something
like
a
property
that
you
can
read
and
write
runtime
or
another
one
would
be
basically
using
a
property
using
a
new
programmatic
type
link.
A
B
There
will
be
a
link
to
the
light
and
there
are
several
aspects
of
of
describing
that.
First
of
all,
there
is
a
property.
The
property
has
a
name.
This
property
has
writability
and-
and
maybe
if
we
add
security,
stuff
or
authorization
stuff
later,
it
will
have
some
authorization
mechanisms
to
it
and
so
on,
and
the
other
part
is
that
the
the
value
of
that
property
actually
is
meant
to
lead
to
the
affordances
of
the
light.
B
So
again,
bringing
the
light
switch
together
with
the
garden
hose
may
be
very
funny,
but
it
may
not
be
the
the
kind
of
value
you
want.
The
property
of
that
light
switch
to
take
on.
So
there
is
a
description
of
the
data
type
that
this
property
actually
can
take
on.
That
includes
information
about
the
target
of
of
that
link.
B
So
we
we
have
the
the
modeling
of
the
property
that
that
tells
us.
How
do
we
set
this
stuff
and
we
also
have
the
modeling
of
the
data
that
tells
us.
So
what
do
we
set
here-
and
the
third
thing
of
course,
is
trying
to
explain
now
that
the
light
search
actually
has
that
link?
What
does
it
actually
mean?
B
So
is
the
light
switch
going
to
do
anything
with
that
link,
or
is
it
just
stored
there
in
a
property
for
somebody
else
to
do
something
with
that
link,
so
that
also
needs
to
be
described.
So
if
we
have
some
some
on
off
mechanism
as
part
of
that
that
light
switch,
we
may
want
to
now.
I'm
missing
a
word
link
there
to
that
link
so
to
to
make
sure.
Well,
okay,
the
the
light
switch
has
three
different
knobs.
B
B
So
if
I
look
at
the
example
that
that's
on
the
screen
right
now,
what
this
tells
me
if
I
translate
this
into
8288
web
linking
so
this
would
be
a
bi-directional
link.
These
do
not
exist,
but
but
that's
not
a
problem
I
mean
we
can
always
invent
that,
so
the
the
the
type
really
would
be
the
the
link
relation
and
the
party
one
would
be.
B
That's
really
not
discussed
in
this
model,
so
you
need
to
understand
next
relationships
to
actually
learn
something
from
this
that
you
probably
then
can
use
in
your
real
world
modeling
of
the
two
arms,
but
that
is
not
creating
or
modifying
an
affordance
of
of
the
digital
side
of
that
arm.
The
fact
that
they
are
physically
next
to
each
other.
It
has
no
bearing
on
on
what
you
actually
can
do
with
these
things,
or
maybe
it
does,
but
that
would
have
to
be
defined.
B
So,
of
course,
the
the
underlying
question
behind
the
slide.
Six
is
what
other
kinds
of
physical
properties
can
we
describe
that
maybe
doesn't
have
anything
to
do
with
links
at
all.
So
I
I
have
elevation
or
something
that
that
I
can
describe
for
something
that
doesn't
need
a
link,
but
it's
of
the
same
kind
of
information
that
the
relations
on
this
page
have.
A
Yes,
actually,
the
elevation
is
kind
of
an
interesting
example,
because
I
guess
you
could
easily
define
that
with
this
kind
of
construct.
That,
like
the
type
of
the
link,
would
I
mean,
maybe
the
current
slide
is
actually
a
bad
example
better.
Can
you
flip
an
example
earlier
example,
because
this
is
owning.
A
A
So
if
you
think
about
that
elevation
example
you
had-
and
if
you
would
need
to
this
this
let's
say-
and
I
I
assume
that
your
example
was
an
elevation
of
it.
You
know
a
device
would
be.
You
know
x
meters
awesome,
so
you
could.
What
one
way
to
express
that
would
be
the
sdf
relation
called
elevation.
The
relation
type
would
be
a
terminal
ontology
that
is
about
elevation,
and
the
target
would
actually
instead
be
another
definition.
Just
a
row
value
you
know,
42..
B
B
That
is
of
the
same
general
kind,
where
the
the
information
doesn't
describe
an
affordance.
That's
all
we
can
do
in
sdf
right
now
describe
references,
but
it's
actually
describing
a
property
that
that
is
yeah
modulating
all
these
affordances.
But
it's
not
really
changing
anything.
What
you
can
do
with
your
your
water
kettle,
you
can
buy
water,
but
the
the
knobs
don't
change
when
you
use
it
at
different
elevations.
A
Indeed-
and
I
think
them
because
I
guess
there's
so
many
different
ways
to
use
links-
we
probably
do
advance
the
discussion.
We
need
to
focus
on
a
specific
case
to
begin
with,
so
what
I
think,
what
we
could
perhaps
focus
for
a
while.
Is
this
bit
the
fact
that
what
I'm
thinking
is
probably
the
closest
thing
that,
for
example,
logan
and
jerry
need
based
on
the
discussions
we
had
earlier?
A
Is
this
having
let's
say
it's
more
semantic
information
on
those
definitions,
whether
it's
affordance
or
thing,
or
an
object
and
being
able
to
say,
for
example,
hey
this?
This
object
is,
is
kind
of
the
same
thing
as
this
temperature
sensor
or
in
that
ontology,
and
that's
kind
of
what
the
same
as
relation
is
here,
trying
to
express.
B
So
one
is
the.
The
object
is
the
same
as
the
concept
of
a
temperature
sensor,
which
is
a
bit
weird,
but
but
maybe
you
actually
mean
that
and
the
other
one
is
the.
The
object
is
contained
in
the
concept
of
a
room,
which
also
is
not
very
clear
to
me,
I
mean,
usually
things
are
contained
in
specific
rooms
and
not
in
the
concept
of
a
room.
H
B
A
A
But
actually
maybe
it's
as
a
cherry
accident,
I
saw
you
wrote
on
the
chat
to
kind
of
start
with
a
motivating
use
case.
Maybe
that
kind
of
approach
is
is
here
also
useful
and
maybe,
for
example,
look
at
one
of
the
use
cases
that
you
have.
We
could
see
like
if
we
can
model
those
those
use
cases
with
the
right
kind
of
information
here.
So
do
you
want
to
have
yeah
your
stud
thanks.
D
A
Well,
I
guess
maybe
we
could
do
a
bit
of
both
like
if
you
have
one
quick
and
easy.
We
could,
you
know,
discuss
right
now
and
then
we
could
maybe
have
a
look
at
snow.
Now
we
have
15
minutes
left
so
something
very
simple.
We
could
use
right
now
and
then
maybe
for
a
future
session.
We
could
look
at
a
bit
more
detail.
D
D
D
D
But
we
can
send
this
excuse
me.
We
can
send
that
command
to
several
different
devices.
That's
able
to
fulfill
it,
so
maybe
the
blind
say
yeah.
I
can
open
and
you
know,
but
I'm
going
to
raise
the
temperature.
If
I
do
that
the
door
says
well,
I
I
can
open
and
I'll
lower
the
temperature.
The
ai
needs
to
be
able
to
make
a
generic
assessment
and
then
send
that
command
to
a
specific
object.
A
Okay,
thanks
jerry,
I
mean
maybe
like
we'll-
be
the
kind
of
key
for
your
use
case,
the
key
pieces
of
information
that
you
would
need
to
find.
For
example,
I
mean
you
would
need
to
be
able
to
find
blinds,
for
example,
so
you
need
in
your
system
a
way
to
say:
hey,
give
me
sdf
definitions
of
things
that
are
of
a
kind
blind
and
the
blind
being
a
term
used
in
a
vocabulary
that
your
system
knows
about.
D
A
D
A
So
here
what
I
would
expect
like
that
you
would,
if
you
look
at
the
design
on
this
slide,
probably
you
would
do
something
like
look
for
things
that
have
a
relation
I
mean
it
could
be
same
as
or
some
other
term.
That
is
how
you
would
describe
what
kind
of
things
you're
looking
for
and
then
the
target
would
match.
A
Do
the
exact
thing:
that's
a
blind
type,
and
then,
of
course,
that
only
works
if
it's
a
single
ontology
for
both
the
types
and
targets
used
within
your
domain
or
if
the
ondologies
are
mapped
together
across
multiple
domains.
That
at
least
would
give
you
that
piece
of
information,
then
you
would
still
use
sdf
as
the
generic
way
to
describe
capabilities,
but
you
have
additional
semantic
hints
on
your
system
to
look
for.
D
A
A
A
B
D
D
D
D
D
You
know
without
having
a
discussion,
so
you
know
what
I'd
really
like
to
do
is
be
able
to.
You
know,
not
bother
the
group
with
a
specific
concern
of
mine.
I'd
be
able
I'd
like
to
be
able
to
just
go
through
the
documentation
and
such
and
say.
Oh,
you
know
what
I
need
to
be
using
sdf
relation
for
this.
A
A
A
C
I
I
don't
have
any
issue
with
this
design.
In
fact,
I
think
this
design
can
actually
do
the
two-arm
three-arm
thing
all
you
have
to
do
is
just
refactor
it,
it's
really
no
different.
So
I
I
made
an
example
where
I
have
an
ontology
I
defined
using
this
this
sdf
relation
pattern.
Can
I
share
a
slide
quickly?
Sure?
I
C
G
I
I
A
I
It's
on
my
github
also,
but
it
would
be
nice
if
I
could
share
my
screen
what
I
was
going
to
show.
You
was
an
sdf
file
that
defines
two
components
of
an
ontology
and
two
items
and
their
relations
they're,
both
their
attributes
and
their
relations
to
one
another.
So
it's
a
piece
of
equipment
that
has
components
within
it
and
but
instead
of
defining
that,
as
you
know,
because
there
are
so
many
layers
in
this
instead
of
defining
any
of
that,
as
you
know,
sdf
inclusion,
it's
all
defined
through
the
relation
types
you.
C
C
I
I
I
think
to
align
with
what
more
people
are
expecting
to
see
from
the
web
linking
world
or
just
won't,
won't
change
anything
but
yeah,
and
so
sdf
defines
these
relations
and
then
what
I
wish
I
could
show
what
I've
done,
but
I
don't
really
want
to
define
the
instances
in
sdf
in
this
case
for
this
system.
This
is
like
the
other
system
I'm
working
on.
I
I
I
define
sdf
all
the
way
to
the
instance
modeling
level
and
then
make
instances
from
that,
but
here
I
just
want
to
use
sdf
to
define
the
ontology
and
then
when,
when
the
graph
instances
are
made
they're
just
made
in
plain
jsons,
so
I
convert
the
sdf
graph
with
all
of
its
relations
to
a
schema
that
allows
you
to
check
and
generate
a
plain
json,
but
also
we
this
is.
This
is
really
part
of
my
regular
day
job.
I
We
also
generated
an
sql
database
and-
and
I'm
going
to
do
all
the
schema
for
that
also
from
this,
so
it's
sort
of
a
pattern
that
shows
the
relations
of
sdf
objects
to
other
sdf
objects,
but
I
think
it's,
I
don't
think
it
needs
any
new
features.
In
other
words,
I
think
what
you're
defining
there
is
is
functionally
complete
and
we
just
need
to
test
it.
I
think
what
we
should
do
going
forward.
A
A
I
A
I
A
Yep,
okay,
so
sounds
very
good.
So
let's
take
that
as
an
accident,
I
said
michael,
let's
work
on
the
different
use
cases.
You
know
what
cherry
would
have
for
you,
what
they
have
from
us
and
make
examples
of
sdf
definitions
using
this
pattern
and
and
see
how
it
plays
out
and
then
come
back
to
that
and
work
our
way
from
there
and
and
I
guess
it
would
be-
I'm
sorry
go
ahead.
A
A
A
Okay-
here
you
know
in
that
case
those
and
the
next
steps.
So
basically,
let's
check
with
the
with
nicos
and
michael
richardson
like
if
that's
the
sdf
github
is
an
appropriate
thing
to
do.
I
will
post
it
there.
We
can
post
a
link
to
the
rtlist
for
continued
discussion
and
specifically
on
the
draft,
and
meanwhile
we
work
on
the
examples
and
come
to
those
I
think
after
summer
vacations,
maybe
in
the
next
weekly
call
and
take
closer
look
what
we
have
achieved.
A
Good,
so
with
that
thanks
a
lot
everyone
for
joining
the
we
wish
you
call
today.
I
think
we
made
some
good
progress
here
and
then
see
you
soon
again,
perhaps
at
the
ietf
meetings
during
the
affiliate
ietf
personally,
I
won't
be
on
site,
but
I'm
planning
to
share
a
few
lessons
remotely,
but
then
latest
in
the
thinking
rc
sessions
after
summer
vacations
towards
the
end
of
august.