Online Parallelization in Conformant Planning

Presently, the main idea of most state-of-the-art conformant planners is to transform a conformant planning problem into a search problem in the space of brief states. However, directly searching a parallel plan solution in the space of brief states has an Achilles´ heel: the space of brief states whose elements are sets of possible world states, is always too large. It´s alomst impossible for the planner to branch on all possible subsets of parallel actions because the branching factor will exponentially increasing. In this paper, we design a system called CFFP on the basis of an efficient conformant planner CFF (conformant fast-forward). CFFP generates parallel plans by using recursion online parallelization of partial plans. Although this method can´t generate optimal parallel plan, empirical results show that it can generate good plan close to the optimal parallel planning by small cost compared to CFF.