Despite recent progress in AI planning, many benchmarks remain challenging for current planners. In many domains, the performance of a planner can greatly be improved by discovering and exploiting information about the domain structure that is not explicitly encoded in the initial PDDL formulation. In this paper we present and compare two automated methods that learn relevant information from previous experience in a domain and use it to solve new problem instances. Our methods share a common four-step strategy. First, a domain is analyzed and structural information is extracted, then macro-operators are generated based on the previously discovered structure. A filtering and ranking procedure selects the most useful macro-operators. Finally, the selected macros are used to speed up future searches.We have successfully used such an approach in the fourth international planning competition IPC-4. Our system, Macro-FF, extends Hoffmann's state-of-the-art planner FF 2.3 with support for two kinds of macro-operators, and with engineering enhancements. We demonstrate the effectiveness of our ideas on benchmarks from international planning competitions. Our results indicate a large reduction in search effort in those complex domains where structural information can be inferred.

Large-scale, parallel clusters composed of commodity processors are increasingly available, enabling the use of vast processing capabilities and distributed RAM to solve hard search problems. We investigate Hash-Distributed A* (HDA*), a simple approach to parallel best-first search that asynchronously distributes and schedules work among processors based on a hash function of the search state. We use this approach to parallelize the A* algorithm in an optimal sequential version of the Fast Downward planner, as well as a 24-puzzle solver. The scaling behavior of HDA* is evaluated experimentally on a shared memory, multicore machine with 8 cores, a cluster of commodity machines using up to 64 cores, and large-scale high-performance clusters, using up to 2400 processors. We show that this approach scales well, allowing the effective utilization of large amounts of distributed memory to optimally solve problems which require terabytes of RAM. We also compare HDA* to Transposition-table Driven Scheduling (TDS), a hash-based parallelization of IDA*, and show that, in planning, HDA* significantly outperforms TDS. A simple hybrid which combines HDA* and TDS to exploit strengths of both algorithms is proposed and evaluated.

While there have been many papers published on path planning in grids, there has not been significant work on comparing existing approaches, and it is difficult to evaluate new work in comparison to existing work. After creating a public repository of grid-based path planning problems we created the grid-based planning competition (GPPC) to facilitate these comparisons. This article describes the motivation and design of the competition, as well as plans for the future of the competition.

Reconfiguration of radial distribution networks is the basis of supply restoration after faults and of load balancing and loss minimization. The ability to automatically reconfigure the network quickly and efficiently is a key feature of autonomous and self-healing networks, an important part of the future vision of Smart Grids. We address the reconfiguration problem for outage recovery, where the cost of the switching actions dominates the overall cost: when the network reverts to its normal configuration relatively quickly, the electricity loss and the load imbalance in a temporary suboptimal configuration are of minor importance. Finding optimal feeder configurations under most optimality criteria is a difficult optimization problem. All known complete optimal algorithms require an exponential time in the network size in the worst case, and cannot be guaranteed to scale up to arbitrarily large networks. Hence most works on reconfiguration use heuristic approaches that can deliver solutions but cannot guarantee optimality. These approaches include local search, such as tabu search, and evolutionary algorithms. We propose using optimal informed search algorithms in the A* family, introduce admissible heuristics for reconfiguration, and demonstrate empirically the efficiency of our approach. Combining A* with admissible cost lower bounds guarantees that reconfiguration plans are optimal in terms of switching action costs.

Pathfinding is important in many applications, including games, robotics and GPS itinerary planning. In games, most pathfinding methods rely on runtime search. Despite numerous enhancements introduced in recent years, runtime search has the disadvantage that, in bad cases, most parts of a map need to be explored, causing a time performance degradation. In this work we explore a significantly different approach to pathfinding, eliminating the need for runtime search. Optimal paths between all pairs of locations are pre-computed. Since straightforward ways to store pre-computed paths are prohibitively expensive even for maps of moderate size, pre-computed data are compressed, reducing the memory requirements dramatically. At runtime, pathfinding is very fast, as it requires visiting only the locations on an optimal path. In each location, a quick computation provides the next move along the optimal path. We demonstrate the effectiveness of this approach on Baldur's Gate game maps. The compression factor reaches two orders of magnitude, bringing the memory requirements down to reasonable values. Compared to A* search, the runtime speedup reaches and even exceeds two orders of magnitude. When averaged over paths of similar cost, the speedup reaches a value of 700 in our experiments.