- Titel
- Generalisierungseigenschaften eines Multi- Agenten-Reinforcement-Learning-Ansatzes zur Steuerung von Portalrobotern
- Zitierfähige Url:
- https://nbn-resolving.org/urn:nbn:de:bsz:14-qucosa2-997474
- Übersetzter Titel (EN)
- Generalisation capabilities of a Multi-Agent Reinforcement Learning Approach for Controlling Gantry Robots
- Konferenz
- 21. ASIM-Fachtagung Simulation in Produktion und Logistik. Dresden, Germany, 24. bis 26. September 2025
- Quellenangabe
- Simulation in Produktion und Logistik 2025
Herausgeber: Sebastian Rank
Herausgeber: Mathias Kühn
Herausgeber: Thorsten Schmidt
Erscheinungsort: Dresden
Verlag: Technische Universität Dresden, Professur für Technische Logistik
Erscheinungsjahr: 2025
ISBN: 978-3-86780-806-4 - Erstveröffentlichung
- 2025
- DOI
- https://doi.org/10.25368/2025.266
- Abstract (EN)
- The transformation of production and logistics through Industry 4.0 and digitalisation presents new challenges, including shorter product lifecycles, increased product variety, and more dynamic production conditions. Reinforcement Learning (RL) is considered to offer promising solutions to these challenges. This study investigates the generalisation capabilities of a Multi-Agent Reinforcement Learning (MARL) approach for controlling gantry robots in flow shops. A decentralised Deep Q-Learning algorithm is employed, and its generalisation is evaluated through simulation experiments. The results demonstrate that the MARLbased control strategy can maintain stable performance across a variety of process changes without the need of retraining. However, more efficient strategies are still achievable with scenario-specific training. The study highlights the importance of selecting generalisable hyperparameters and suggests future research into transfer learning to improve or potentially avoid the need for retraining. The results contribute to the development of adaptive, autonomous control systems in dynamic Industry 4.0 environments.
- Freie Schlagwörter (DE)
- Simulation, Künstliche Intelligenz, Modellierung, Optimierung, Produktlebenszyklen, Multi-Agent Reinforcement Lernansatzes (MARL)
- Freie Schlagwörter (EN)
- Simulation, Artificial Intelligence, Modelling, Optimisation, product lifecycles, Multi-Agent Reinforcement Learning (MARL)
- Klassifikation (DDC)
- 620
- Klassifikation (RVK)
- ZM 9500
- Publizierende Institution
- Technische Universität Dresden, Dresden
- Version / Begutachtungsstatus
- publizierte Version / Verlagsversion
- URN Qucosa
- urn:nbn:de:bsz:14-qucosa2-997474
- Veröffentlichungsdatum Qucosa
- 21.10.2025
- Dokumenttyp
- Konferenzbeitrag
- Sprache des Dokumentes
- Deutsch
- Lizenz / Rechtehinweis
CC BY-NC-SA 4.0