Daniel Gnad

Beyond Forks: Finding and Ranking Star Factorings for Decoupled Search

Star-topology decoupling is a recent search reduction method for forward state space search. The idea basically is to automatically identify a star factoring, then search only over the center component in the star, avoiding interleavings across leaf components. The framework can handle complex star topologies, yet prior work on decoupled search considered only factoring strategies identifying fork and inverted-fork topologies. Here, we introduce factoring strategies able to detect general star topologies, thereby extending the reach of decoupled search to new factorings and to new domains, sometimes resulting in significant performance improvements. Furthermore, we introduce a predictive portfolio method that reliably selects the most suitable factoring for a given planning task, leading to superior overall performance.

Star-topology decoupled state space search

Abstract State space search is a basic method for analyzing reachability in discrete transition systems. To tackle large compactly described transition systems – the state space explosion – a wealth of techniques (e.g., partial-order reduction) have been developed that reduce the search space without affecting the existence of (optimal) solution paths. Focusing on classical AI planning, where the compact description is in terms of a vector of state variables, an initial state, a goal condition, and a set of actions, we add another technique, that we baptize star-topology decoupling, into this arsenal. A star topology partitions the state variables into components so that a single center component directly interacts with several leaf components, but the leaves interact only via the center. Many applications explicitly come with such structure; any classical planning task can be viewed in this way by selecting the center as a subset of state variables separating connected leaf components. Our key observation is that, given such a star topology, the leaves are conditionally independent given the center, in the sense that, given a fixed path of transitions by the center, the possible center-compliant paths are independent across the leaves. Our decoupled search hence branches over center transitions only, and maintains the center-compliant paths for each leaf separately. As we show, this method has exponential separations to all previous search reduction techniques, i.e., examples where it results in exponentially less effort. One can, in principle, prune duplicates in a way so that the decoupled state space can never be larger than the original one. Standard search algorithms remain applicable using simple transformations. Our experiments exhibit large improvements on standard AI planning benchmarks with a pronounced star topology. 1

Unchaining the Power of Partial Delete Relaxation, Part II: Finding Plans with Red-Black State Space Search

Red-black relaxation in classical planning allows to interpolate between delete-relaxed and real planning. Yet the traditional use of relaxations to generate heuristics restricts relaxation usage to tractable fragments. How to actually tap into the red-black relaxation’s interpolation power? Prior work has devised red-black state space search (RBS) for intractable red-black planning, and has explored two uses: proving unsolvability, generating seed plans for plan repair. Here, we explore the generation of plans directly through RBS. We design two enhancements to this end: (A) use a known tractable fragment where possible, use RBS for the intractable parts; (B) check RBS state transitions for realizability, spawn relaxation refinements where the check fails. We show the potential merits of both techniques on IPC benchmarks.

Star-Topology Decoupling in SPIN

Star-topology decoupling is a state space search method recently introduced in AI Planning. It decomposes the input model into components whose interaction structure has a star shape. The decoupled search algorithm enumerates transition paths only for the center component, maintaining the leaf-component state space separately for each leaf. This is a form of partial-order reduction, avoiding interleavings across leaf components. It can, and often does, have exponential advantages over stubborn set pruning and unfolding. AI Planning relates closely to model checking of safety properties, so the question arises whether decoupled search can be successful in model checking as well. We introduce a first implementation of star-topology decoupling in SPIN, where the center maintains global variables while the leaves maintain local ones. Preliminary results on several case studies attest to the potential of the approach.