Improving Applicability of Planning in the RoboCup Logistics League Using Macro-actions Refinement

Marco De Bortoli, Lukás Chrpa, Martin Gebser, Gerald Steinbauer-Wagner

Abstract: This paper focuses on improving the action plans obtained through the use of sequential macro-actions in temporal planning. Macro-actions are a way to address the high complexity of temporal planning in challenging domains. Investigating the Robocup Logistics League (RCLL), a testbed for logistics scenarios in the area of Industry 4.0, we introduce a method to unfold the macro-actions of an obtained abstract plan back into their original atomic actions in an improved plan. This allows to extract potentially better solutions in terms of makespan from the Simple Temporal Network (STN) representing the abstract plan. The proposed method is evaluated on a macro-based modeling of the RCLL domain and is shown to yield improved plans over those obtained using either the original atomic actions or the macro-actions without refinement.