Towards Safe Neural Networks for Autonomous Mobile Systems: Literature Review and Research Agenda

Ziegler P, Hilbinger J, Franke J, Reitelshöfer S (2026)

Publication Language: English

Publication Type: Conference contribution, Conference Contribution

Publication year: 2026

Publisher: Springer

Series: Lecture Notes in Production Engineering

City/Town: Cham

Pages Range: 503-512

Conference Proceedings Title: Production at the Leading Edge of Technology. Proceedings of the 15th Congress of the German Academic Association for Production Technology (WGP), Hannover, September 2025

Event location: Hannover DE

ISBN: 9783032195234

DOI: 10.1007/978-3-032-19524-1_54

Abstract

Autonomous driving has achieved notable success in structured environments like highways and urban roads, where deterministic control systems and well-defined safety protocols ensure reliable operation. In controlled settings such as warehouses, fully automated vehicles rely on static sensor fusion techniques to navigate in predictable conditions. However, off-highway domains, such as agriculture and construction, present substantial challenges due to variable terrain, unpredictable obstacles, and environmental factors including dust, rain, and vegetation. These complexities necessitate advanced perception systems that utilize sophisticated data-processing pipelines, incorporating neural networks and multimodal sensors. Despite their functional potential, neural network architectures raise concerns in safety–critical applications, necessitating innovative solutions to ensure robust operation. In this paper, we address these challenges with a systematic literature review that explores the evolving landscape of safety–critical AI applications, focusing on promising approaches to enhance AI-based perception for off-highway autonomous mobile systems (AMS). Our findings highlight the role of mid-level sensor fusion, Out-of-Distribution (OoD)-detection, and explainable AI (XAI) techniques in improving reliability and functional safety. Drawing on these insights, we propose a research agenda for developing a reliable perception system tailored to off-highway applications. This agenda emphasizes the reliance on multiple research streams for handling diverse environmental conditions and enabling real-time integration of different validation architectures.

Authors with CRIS profile

Patrick Ziegler Lehrstuhl für Fertigungsautomatisierung und Produktionssystematik (FAPS) Jörg Franke Lehrstuhl für Fertigungsautomatisierung und Produktionssystematik (FAPS) Sebastian Reitelshöfer Lehrstuhl für Fertigungsautomatisierung und Produktionssystematik (FAPS)

How to cite

APA:

Ziegler, P., Hilbinger, J., Franke, J., & Reitelshöfer, S. (2026). Towards Safe Neural Networks for Autonomous Mobile Systems: Literature Review and Research Agenda. In Ludger Overmeyer, Bernd-Arno Behrens (Eds.), Production at the Leading Edge of Technology. Proceedings of the 15th Congress of the German Academic Association for Production Technology (WGP), Hannover, September 2025 (pp. 503-512). Hannover, DE: Cham: Springer.

MLA:

Ziegler, Patrick, et al. "Towards Safe Neural Networks for Autonomous Mobile Systems: Literature Review and Research Agenda." Proceedings of the 15th Congress of the German Academic Association for Production Technology (WGP), Hannover Ed. Ludger Overmeyer, Bernd-Arno Behrens, Cham: Springer, 2026. 503-512.

BibTeX: Download