<s>
Partial-order	B-Application
planning	I-Application
is	O
an	O
approach	O
to	O
automated	B-Application
planning	I-Application
that	O
maintains	O
a	O
partial	O
ordering	O
between	O
actions	O
and	O
only	O
commits	O
ordering	O
between	O
actions	O
when	O
forced	O
to	O
,	O
that	O
is	O
,	O
ordering	O
of	O
actions	O
is	O
partial	O
.	O
</s>
<s>
Partial-order	B-Application
planning	I-Application
relies	O
upon	O
the	O
Principle	O
of	O
Least	O
Commitment	O
for	O
its	O
efficiency	O
.	O
</s>
<s>
This	O
is	O
a	O
partial	O
plan	O
because	O
the	O
order	O
for	O
finding	O
eggs	O
,	O
flour	O
and	O
milk	O
is	O
not	O
specified	O
,	O
the	O
agent	O
can	O
wander	O
around	O
the	O
store	O
reactively	B-Application
accumulating	O
all	O
the	O
items	O
on	O
its	O
shopping	O
list	O
until	O
the	O
list	O
is	O
complete	O
.	O
</s>
<s>
A	O
partial-order	O
planner	O
is	O
an	O
algorithm	O
or	O
program	B-Application
which	O
will	O
construct	O
a	O
partial-order	O
plan	O
and	O
search	O
for	O
a	O
solution	O
.	O
</s>
<s>
The	O
problem	O
can	O
be	O
interpreted	O
as	O
a	O
search	B-Application
problem	I-Application
where	O
the	O
set	O
of	O
possible	O
partial-order	O
plans	O
is	O
the	O
search	O
space	O
.	O
</s>
<s>
As	O
seen	O
in	O
the	O
algorithm	O
presented	O
above	O
,	O
partial-order	B-Application
planning	I-Application
can	O
encounter	O
certain	O
threats	O
,	O
meaning	O
orderings	O
that	O
threaten	O
to	O
break	O
connected	O
actions	O
,	O
thus	O
potentially	O
destroying	O
the	O
entire	O
plan	O
.	O
</s>
<s>
Partial-order	B-Application
planning	I-Application
algorithms	O
are	O
known	O
for	O
being	O
both	O
sound	O
and	O
complete	O
,	O
with	O
sound	O
being	O
defined	O
as	O
the	O
total	O
ordering	O
of	O
the	O
algorithm	O
,	O
and	O
complete	O
being	O
defined	O
as	O
the	O
capability	O
to	O
find	O
a	O
solution	O
,	O
given	O
that	O
a	O
solution	O
does	O
in	O
fact	O
exist	O
.	O
</s>
<s>
Partial-order	B-Application
planning	I-Application
is	O
the	O
opposite	O
of	O
total-order	O
planning	O
,	O
in	O
which	O
actions	O
are	O
sequenced	O
all	O
at	O
once	O
and	O
for	O
the	O
entirety	O
of	O
the	O
task	O
at	O
hand	O
.	O
</s>
<s>
Anthony	O
Barret	O
and	O
Daniel	O
Weld	O
have	O
argued	O
in	O
their	O
1993	O
book	O
,	O
that	O
partial-order	B-Application
planning	I-Application
is	O
superior	O
to	O
total-order	O
planning	O
,	O
as	O
it	O
is	O
faster	O
and	O
thus	O
more	O
efficient	O
.	O
</s>
<s>
They	O
tested	O
this	O
theory	O
using	O
Korf	O
’s	O
taxonomy	O
of	O
subgoal	O
collections	O
,	O
in	O
which	O
they	O
found	O
that	O
partial-order	B-Application
planning	I-Application
performs	O
better	O
because	O
it	O
produces	O
more	O
trivial	O
serializability	O
than	O
total-order	O
planning	O
.	O
</s>
<s>
They	O
found	O
that	O
partial-order	B-Application
planning	I-Application
is	O
more	O
adept	O
at	O
finding	O
the	O
quickest	O
path	O
,	O
and	O
is	O
therefore	O
the	O
more	O
efficient	O
of	O
these	O
two	O
main	O
types	O
of	O
planning	O
.	O
</s>
<s>
Partial-order	O
plans	O
are	O
known	O
to	O
easily	O
and	O
optimally	O
solve	O
the	O
Sussman	B-Application
anomaly	I-Application
.	O
</s>
<s>
This	O
was	O
a	O
result	O
of	O
partial-order	B-Application
planning	I-Application
that	O
solidified	O
its	O
place	O
as	O
an	O
efficient	O
planning	O
system	O
.	O
</s>
<s>
This	O
higher	O
per-node	O
cost	O
occurs	O
because	O
the	O
algorithm	O
for	O
partial-order	B-Application
planning	I-Application
is	O
more	O
complex	O
than	O
others	O
.	O
</s>
<s>
This	O
has	O
important	O
artificial	B-Application
intelligence	I-Application
implications	O
.	O
</s>
