Robust Integration of Planning, Execution, Recovery and Testing to Win the RoboCup Logistics League

David Beikircher, Marco De Bortoli, Leo Fürbaß, Thomas Kernbauer, Peter Kohout, Dominik Lampel, Anna Masiero, Stefan Moser, Martin Nagele, Gerald Steinbauer-Wagner

Abstract: This paper presents the evolution of the software architecture developed by the GRIPS team from Graz University of Technology for the RoboCup Logistics League (RCLL) competition, to the point of winning the 2023 RoboCup Logistics League. The RCLL simulates a flexible production environment, and the team’s software addresses the challenges of automation, planning, and execution in this dynamic setting. The software architecture consists of three main layers: planning and dispatching, mid-level execution, and low-level control. Key features include temporal planning for optimization, Behavior Trees for fine-grained control, and robust recovery strategies. Testing methods encompass real field testing and simulations.