Constraint Based Planning by Stochastic Search

Planning by incomplete stochastic search ooers a promising alternative to classical, complete planning methods. The success of this approach is documented by recent performance results obtained by transforming planning tasks into propositional sat-issability problems and using existing eecient local search methods to solve them. On the other hand, we argue that these results can still be improved if local search is employed directly to planning problems , i.e., without making the detour via propositional logic. To this end we introduce a general-purpose planning structure which is amenable to stochastic search methods. The core of this structure is formed by constraints, which represent a planning problem and at the same time determine the objective function to be minimized by local search. Adapting the planning structure on the basis of a pre-processing by domain analysis may furthermore signiicantly reduce the number of constraints and so speed up the following search.