Using Available Memory to Transform Graphplan's Search

We present a major variant of the Graphplan algorithm that employs available memory to transform the depth-first nature of Graphplan's search into an iterative state space view in which heuristics can be used to traverse the search space. When the planner, PEGG, is set to conduct exhaustive search, it produces guaranteed optimal parallel plans 2 to 90 times faster than a version of Graphplan enhanced with CSP speedup methods. By heuristically pruning this search space PEGG produces plans comparable to Graphplan's in makespan, at speeds approaching state-of-the-art heuristic serial planners.