Partialization - generalize a plan to find partial 
orderings

LPG - uses serial action graphs, then partializes them
to get parrallel plans.

Given serial plan, causal proof results in partialization,
causal links with start times and end times.

-Disjunctive temporal network

-Dispatch: any valid solution to the temporal network

-Find disjunctive temporal network such that the serial
plan is a dispatch of the network.

-Find dispatch with best properties

--------------------------

If multiple actions in the serial action graph provide
a precondition then there can be multiple temporal networks
for a given action graph (causal stucturs)

-Multiple causal structurs 
   disjunctive causal links
       Be greedy and select the causal structure
       that will have the right kind of dispatch.

-Multiple threat resolutions
   disjunctive orderings

-If disjunction present, finding dispatch can be costly

-Partialization can be NP-complete
-Greedy selection makes this much easier

LPG - far from optimal makespan.


LPG

flaws = unsupported conditions and mutex

uses serial action graphs to avoid mutex.
This allows larger neighborhoods to be searched,
(larger neighborhoods help avoid local max/min
problems of loacal search.

but where is the planning graph?

-PG is used to supply the list of potential repairs
to repair each flaw.

PG generates an envelope over which the neighborhood 
is generated.

uses PG based heuristics to choose which of repairs
to try.


To improve greedy search increase neighborhood size,
-skip some mutex to keep per-item cost down.

Because LPG is a greedy search, runtimes and solution
plans returned can be different from execution to 
execution.  Time reported are median times.