complete-tutorial-ml97-4ths

Description

Verallfurther&to$this%MotivICML/COLItTin1997ThanksLEARNINGofinu.ePLANNINGManesuelaecienM.qualitVappliedelosoCommenComputerISciencenotDepartmenoftFCarnegieCSD,MellonLearningUniveersitLearningyinductivPittsburghofPdecisionAof15213-3891itsphone:ground(412)h.268-8464'fax:Do(412)jo,268-5576foremail:[e-mail]://www.cs.cmconu.edu/c~$mmLearningv/1CMUcopVplanningeloso,eyCSD,dataCMUthe'wledge&guide$pro%hShortoTbasedutorialdata.Description:proLearningbintheoreticalPlanningcPlanningCSD,is$aondecisiontmakingDanielproersidadcessI,whichelphinreferencesve)olvtheesdotheThegenerationtutorialofailablesequencesandoftheactionsfreethatvac.hievVe'aOutlinesetPlanningofPlanninggoalstofrom4aeloso,givventhestate.vTheplanner'smainerformance.issuesiningovbolvondedeinclassication.theaddressescomputationalacquisitionplanningknotasktoare:tlyHoawmakingtocessacquirereacandsolutionsrepresengotdtheyplanningonactioninputmoLearningdel?planningHovideswfortoothgenerateandplansresearcin2acomputationallyeloso,tractableCMUw&a%y?tsHothiswcumentocreatetoplansBorraofUnivgoCarlosoIdMadrid,qualithisy?organizingandtutorial ...

Informations

• Publié par: Nufor
• Nombre de lectures: 9
• Langue: English

Extrait

eloso,
'
planner's
&
explanations
$
reac
%
tutorial.
ICML/COL
ation:
T
addresses
1997
planning
LEARNING
to
in
the
PLANNING
du
Man
V
uela
b
M.
tly
V
y
eloso
and
Computer
ts
Science
I
Departmen
exhaustiv
t
the
Carnegie
eel
Mellon
CMU
Univ
Applied
ersit
the
y
in
Pittsburgh
e
P
kno
A
making
15213-3891
go
phone:
input
(412)
for
268-8464
2
fax:
&
(412)
cumen
268-5576
Univ
email:
his
v
of
[e-mail]
at
u.edu

h
v
ttp://www.cs.cm
after
u.edu/
tact
~

mm
%
v/

1
Conclusion
c
v

v
V
erformance.
eloso,
go
CSD,
ond
CMU
classication.
'
acquisition
&
to
$
a
%
cess
Short
solutions
T
d
utorial
on
Description:
Learning
Learning
vides
in
oth
Planning
researc
Planning

is
CMU
a
%
decision
this
making

pro
Borra
cess
Carlos
whic
Madrid,
h
organizing
in
A
v
(certainly
olv
is
es
end
the
cumen
generation
author
of
is
sequences
for
of
con
actions
tutorial.
that
to
ac
[e-mail]
hiev
3
e
eloso,
a
&
set

of
and
goals

from
Planning
a
c
giv
CSD,
en
e
state.
o
The
erall
main
p
issues
Learning
in
planning
v
es
olv
ey
ed
inductiv
in
data
the
It
computational
the
planning
of
task
wledge
are:
ecien
Ho
guide
w
decision
to
pro
acquire
to
and
h
represen
of
t
o
the
qualit
planning
based
action
its
mo
data.
del?
in
Ho
pro
w
ground
to
b
generate
applied
plans
theoretical
in
h.
a
c
computationally
V
tractable
CSD,
w
'
a
$
y?
Commen
Ho
on
w
Do
to
t
create
Thanks
plans
Daniel
of
jo,
go
ersidad
o
I
d
I,
qualit
for
y?
help
and
this
nally

Ho
list
w
references
to
not
scale
e)
up
included
to
the
real-w
of
orld
do
problems?
t.
Mac
The
hine
of
learning
tutorial
approac
a
hes
ailable
can
further
b
and
e
tacts
applied
the
to
F
planning
free
to
con
automate
v
the
u.e
pro
.
cess
c
of
V
in
CSD,
terpreting
'
the
$
planning
Outline
exp
Motiv
erience
Planning
in
Learning
to
Planning
reusable
Learning
task
to
kno

wledge
4
in

order
eloso,
to
CMU
improMac
'
qualit
&
learning
$
V
%
trend:
Planning
a
in
eloso,
v
Sp
olv
to
es:
%
Motiv
c
ation
w

to
Giv
hing
en
Handco
kno
wledge
wledge
in
ab
and
out

a
erview
task
ulate
domain
to

a
Giv
of
en
w
a
6
problem
&
sp
Motiv
ecied
engineering
as:
domain
>
strategies.
an
qualit
initial
hes:
conguration
planning
of
wledge:
the
>
state
user's
of
CMU
the
Motiv
\w

orld"
bined
>
solving
a
CMU
set
plans
of
tractable
goals

to
create
b
o
e

ac
scale
hiev
orld
ed


CMU
Find
%
a
Exp
solution
CSD,
to
Kno
the
hes:
problem,
and
i.e.,
wledge.
a
con
way
Dene
to
pro
transform
plans.
the
learning
initial
Automate
conguration
of
in
erience
to
task
a
con
new
y
state
recen
of
bine
the
7
w
eloso,
orld
&
where
utorial
the

goal
planning
statemen
erview
t
hes
is
planning
true.
transfer
5
erience
c


w
V
generate
eloso,
in
CSD,
computationally
CMU
w
'
y?
&
Ho
$
to
%
plans
Man
go
y
d
issues
y?
to
Ho
resolv
to
e..
up
.
real-w
Motiv
problems?
ation
c
A
V
few
CSD,
are:
'

$
What
Reac
kno
Planning
wledge
ertise
denes
ation
the
eloso,
task

domain?
wledge

approac
Ho
>
w
de
to
rene
represen
kno
t
>
the
ecify
planning
trol
action
>
mo
kno
del?
to

duce
What
y
is

the
hine
(sucien
approac
t)
>
initial
the
state
terpretation
of
the
the
exp
w
in
orld?
reusable

kno
What
domain,
are
trol,
the
qualit
(prioritized)
.
goals?
Most

t
Ho
Com
w
with
to
input.
acquire
c
domain
V
kno
CSD,
wledge
'
ecien
$
tly
T
from
Goals
exp
ation
ert
Ov
h
of
uman
algorithms
planners?
Ov

of
Whic
approac
h
com
algorithm
with
to
Accum
use
and
to
problem
generate
exp
the
8
solution

plan?
Ho
TASK KNOWLEDGE PLANNER
ANALYZE PLANS GENERATED