A Preliminary Study on the Relaxation of Numeric Features in Planning

Modern AI planners use different strategies to simplify the complexity of current planning problems and turn them more affordable. In this paper, we present a new approach that divides the planning search into two consecutive stages. First, a sequential plan is generated by relaxing the numeric features of actions, thus ignoring the scheduling part of the problem. Second, this plan is parallelised while the satisfaction of numeric constraints is guaranteed, thus including the scheduling part of the problem. In our experiments, we have studied/identified classes of problems where this division is more appropriate and effective w.r.t. the plan quality. Keywords. Planning, Scheduling, Multiobjective planning, Resource optimisation

• Type: Conference
• Year: 2005
• Where: CCIA
• Authors: Antonio Garrido, Eva Onaindia, Donato Hernández
• Keywords: CCIA 2005, Consecutive Stages, Current Planning Problems, Modern AI Planners