KIT | KIT-Bibliothek | Impressum | Datenschutz

Data-Driven Automotive Development: Federated Reinforcement Learning for Calibration and Control

Rudolf, Thomas 1; Schürmann, Tobias ORCID iD icon 1; Skull, Matteo; Schwab, Stefan 1; Hohmann, Sören 2
1 FZI Forschungszentrum Informatik (FZI)
2 Institut für Regelungs- und Steuerungssysteme (IRS), Karlsruher Institut für Technologie (KIT)

Abstract (englisch):

The importance of data-driven methods in automotive development continuously increases. In this area, reinforcement learning methods show great potential, but the required data from system interaction can be expensive to produce during the traditional development process. In the automotive industry, data collection is additionally constrained by privacy aspects with regard to intellectual property interests or customer data. Suitable reinforcement learning approaches need to overcome these challenges for effective and efficient learning. One possible solution is the utilization of federated learning that enables learning on distributed data through model aggregation. Therefore, we investigate the federated reinforcement learning methodology and propose a concept for a continuous automotive development process. The concept contributes separated training loops for the development and for the field operation. Furthermore, we present a customization and verification procedure within the aggregation step. The approach is exemplary shown for an electric motor current control.


Originalveröffentlichung
DOI: 10.1007/978-3-658-37009-1_26
Dimensions
Zitationen: 2
Zugehörige Institution(en) am KIT Institut für Regelungs- und Steuerungssysteme (IRS)
Publikationstyp Proceedingsbeitrag
Publikationsdatum 14.03.2022
Sprache Englisch
Identifikator ISBN: 978-3-658-37008-4
KITopen-ID: 1000143695
Erschienen in 22. Internationales Stuttgarter Symposium : Automobil- und Motorentechnik. Bd.: 1. Hrsg.: M. Bargende
Veranstaltung 22. Internationales Stuttgarter Symposium : Automobil- und Motorentechnik (2022), Online, 15.03.2022 – 16.03.2022
Auflage 1. Aufl.
Verlag Springer Fachmedien Wiesbaden
Seiten 369–384
Vorab online veröffentlicht am 13.03.2022
Schlagwörter Data-Driven Development, Federated Reinforcement Learning, Model-based Verification
Nachgewiesen in Dimensions
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page