An Optimization Framework for Interdependent Planning Goals

This paper describes an approach for optimizing over interdependent planning goals. Most planning systems allow only simple, static dependencies to be defined among goals where these dependencies remain constant between different problems. However, in many domains, goals are related through detailed utility models that may significantly change from problem to problem. For instance in one problem, a particular goal’s utility may increase if other related goals can be achieved. In another problem, this utility increase may differ or actually decrease if the same combination of goals is achieved. To address these types of problem situations, we have implemented a methodology for representing and utilizing information about interdependent goals and their related utilities using the ASPEN planning and scheduling system. We show through experimental results that this approach significantly increases overall plan quality versus a standard approach that treats goal utilities independently.