Entwicklung eines DRL-Agenten zur Reihenfolgeoptimierung für Hochregallager mit Shuttle-Fahrzeugen

Abstract

Due to increasing dynamics and heterogeneity in production, the demands on intralogistics and especially on storage systems have increased. Storage systems must be flexible and enable a high throughput. These requirements are fulfilled by shuttle systems. To be able to increase the throughput of shuttle systems on a software basis, concepts have been developed that use Deep Reinforcement Learning (DRL) to minimise the blockages that arise, e.g. when changing gears or when several withdrawals are made in one gear, by changing the retrieval sequence. These concepts only consider a very small number of storage locations. Real shuttle systems sometimes have several thousand storage locations per level. Therefore, this paper develops a DRL concept that adapts the retrieval sequence in a real shuttle system to minimise blockades and increase throughput.