Scaling up parallel robot plans to improve plan execution time: an industrial use-case

We present a greedy approach to generate parallel plans for domains featuring a large number of objects which need to undergo the same operational procedure. The number of objects is such that not even a total-order (linear) plan can be obtained, thus calling for an alternative method. We propose a compositional approach based on obtaining template solutions for the same problem but over small batches of objects and then combining them into a parallel, sub-optimal solution for the original problem. The approach is showcased on a real-world industrial use-case provided by the Procter&Gamble company in the context of the EU project AIPlan4EU.