Partial Order Planning
The process of computing complete and acyclic Partially Ordered Plan for a given Planning Task.
Some subgoals need to be done in sequence and others dont.
Any algorithm that can place two actions into a Plan without specifying which comes first.
Idea:
- use DAG that supports multiple paths from initial states to goal states
- nodes = actions
- edges = Precondition, Add List, Delete List (Effect)
Acyclic → partial ordering
Advanatges:
- problems can be decomposed
- efficient by least-commitment strategy
- causal links will pinpoint unworkaple plans early
Clobbering conflicts can be resolved by:
We are doing problem description level search and not state space level search. This is some form of thightening.