Graphplan Backward Search (Direct Search I)
Objective: Find a sub-graph of the plangraph that corresponds to a valid plan.
Method: Start from the goal propositions at the last level
Select actions to support the goals so that no two are mutex (choice)
Recurse on the preconditions of the selected actions
(recursion ends at the initial state)
(When backtracking over the goals at a level, memoize them)
Optimizations: Adaptation of DVO, FC, EBL, DDB etc… [Kambhampati, IJCAI99]