Presented at the Scheduling and Planning Applications Workshop at the 31st International Conference on Automated Planning and Scheduling PlanRob (August 2021)
Stefan-Octavian Bezrucav; Malte Kaiser and Burkhard Corves (RWTH Aachen University)
AI task planning approaches are increasingly used in projects with flexible processes, where it must be autonomously deliberated about the selection and scheduling of the actions for the involved actors. In order to implement AI task planning in such cases, the considered scenario must be modelled as a planning problem and a specific planning system for solving it needs to be chosen. However, multiple models for such a planning problem are possible and there exist numerous planning systems. The main challenge at this point is to choose those pairs (planning problem – planning system) that deliver the best results for that specific scenario.
In this paper, an industrial scenario derived from the use-cases of an EU-Project Sharework is targeted. In this scenario mobile manipulators, humans and Autonomous Grounded Vehicles (AGVs), share the working area in order to manipulate items and to execute tasks. After the presentation of a first model for the planning problem for this scenario and its limitation, three further alternatives are introduced. We evaluate the effects on the relevant metrics for such scenarios of these three different modelling variants for the required task planning problems, in combination with four state-of-the-art AI planning systems. The results suggest, that the most suitable combination is the one of a distance-based variant and the TFD planning system.