Decoupled State Space Search – Dissertation Abstract

Decoupled State Space Search is a recent approach to exploiting problem structure in classical planning. The particular structure needed is a star topology, with a single center component interacting with multiple leaf components. All interaction of the leaves with the rest of the problem has to be via the center. Given this kind of problem decomposition, we have showed that search on this reformulated state space can be exponentially more efficient than standard search. However, there do also exist cases in which decoupled search has to spend exponentially more effort to solve a task. We want to tackle this issue by combining decoupled search with different known search enhancement techniques, such as partial-order reduction, symmetry reduction, or dominance pruning. Presumably, these can be nicely combined with our new approach, such that we can prevent the exponential blowup. Decoupled search is not restricted to classical planning, though. Its principles apply to all kinds of (heuristic) search problems, like, e. g., in Model Checking.