*** Answers to questions & miscillaneous points ***

-Ginsberg's Approximate Planning uses a set subtraction operation to
remove subsets of a candidate set that are known to be incorrect.  A
plan is approximately correct when (# of ways to be wrong)/ (# of ways 
to be right) sufficiently approaches zero.

-Marr stated that any representation allows certain ideas to be
represented more easily, while other become more difficult or even
impossible to express.  Using our representation, we can represent
"all sequences that do not start with A7," but we can't represent "all 
sequences that do not contain A7."  

-Assuming all refinement strategies are monotonic and a minimum
solution take k steps, then k refinements are needed to get a minimal
candidate of length k.

-The basic refinement planning template can search an infinite depth
if no solution exists.  The more flexoble Split & Prune method can get 
lost in an infinite search even if a solution exists. (Ex.  After 1
refinement of 1-way rocket domain, performing a search on [0*Fly()<oo] 
will get stuck even though solutions exist in paths that begin with a
loading action)


*** Refinement Planning con't ***

Forward State-space Refinement
	-Partial plan's prefix is extended by adding applicable
actions (action whose preconditions are met)
	-Always uses completely specified states of the world

Backward State-space Refinement
	 -Suffix extended by adding relevant actions (action that
support conditions of regressed state)
	-Since it is goal directed, it usually leads to lower
branching

Plan-space Refinement
	   -Actions are added that give some needed precondition, or
existing actions are used to give the precondition.  The latter is
called simple establishment
       -All intervening steps must preserve the condition
(de-clobbering)
		-new steps can be added to support conditional effects 
(confrontation)
		-placement of step is moved forwards/backwards
(promotion/demotion)
	-(optional book-keeping) Add causal links to preserve the
truth and falsity of establishment.  This adds systemacity.

Least commitment - wanting to have the largest candidate set
possible.  This appears counterintuitive to progressiveness, but the
higher the commitment, the greater the chance of backtracking.