On the Parallelization of Greedy Regression Tables (original) (raw)

This paper presents PGRT, a parallel version of a best first planner based on the Greedy Regression Tables approach. The parallelization method of PGRT distributes the task of extracting applicable actions to a given state among the available processors. Although the number of operators limits the scalability of PGRT, it has proven to be quite efficient for low scale parallelization. A modified Operator Reordering method has been used in order to achieve further increase in the efficiency of the parallel algorithm. We illustrate the speedup of PGRT on a variety of hard logistics problems, adopted from the AIPS-98 planning competition.