When Acyclicity is not Enough: Limitations of the Causal Graph

• Causal graphs are widely used in planning to capture the internal  structure of planning instances. In the past, causal graphs have been exploited to generate hierarchical plans, to compute heuristics, and  to identify classes of planning instances that are easy to solve. It  is generally believed that planning is easier when the causal graph is acyclic. In this paper we show that this is not true in the worst  case, proving that the problem of plan existence is PSPACE-complete  even when the causal graph is acyclic. Since the variables of the  planning instances in our reduction are propositional, this result  applies to STRIPS planning with negative pre-conditions. Having  established that planning is hard for acyclic causal graphs, we study  a subclass of planning instances with acyclic causal graphs whose  variables have strongly connected domain transition graphs. For this  class, we show that plan existence is easy, but that bounded plan  existence is hard, implying that optimal planning is significantly  harder than satisficing planning for this class.

Ämnesord

NATURVETENSKAP  -- Data- och informationsvetenskap -- Datavetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences -- Computer Sciences (hsv//eng)

Publikations- och innehållstyp

ref (ämneskategori)

kon (ämneskategori)