A New Algorithm for Handling Actions in Flexible Graphplan

Flexible Graphplan has attracted considerable interest by its extending Graphplan to capture subtlety of most real problems. But its operators are instantiated by full expansion way that may cause an explosion of the number of actions. Corin R. Anderson has proved that factored expansion often produces a much smaller domain model than full expansion does. In this paper, we will incorporate factored expansion into Flexible Graphplan by the approach that split a flexible action into several flexible components, one for each satisfaction degree. Because these components are not independent, we redefine the action mutex of Flexible Graphplan and handle it in our new algorithm FE-FGP. Our work will generate flexible planning graph with less action nodes and improve the efficiency of plan extraction greatly.