Graph Pruning and Symmetry Breaking on Grid Maps
Daniel Harabor
My research proposes to speed up grid-based pathfinding by identifying and eliminating symmetric path segments from the search space. Two paths are said to be symmetric if they are identical save for the order in which the individual moves (or steps) occur. To deal with path symmetries I decompose an arbitrary grid map into a set of empty rectangles and remove from each rectangle all interior nodes and possibly some from along the perimeter. A series of macro edges are then added between selected pairs of remaining nodes in order to facilitate provably optimal traversal through each rectangle. The new algorithm, Rectangular Symmetry Reduction (RSR), can speed up A* search by up to 38 times on a range of uniform cost maps taken from the literature. In addition to being fast and optimal, RSR requires no significant extra memory and is largely orthogonal all existing speedup techniques. When compared to the state of the art, RSR often shows significant improvement across a range of benchmarks.