AutoDevSafeOps – Integrated development and operation of safe automotive systems

AutoDevSafeOps introduces a unique DevOps framework that embeds safety mechanisms throughout the entire lifecycle.
It enables modular over-the-air updates of safety-critical driving functions — including their safety processes — across vehicle and backend boundaries.

At the heart of the project lies a continuous DevOps cycle comprising these phases:

Plan, Create, Verify, Package, Release, Configure, and Monitor.

Each phase is supported by dedicated models, tools, and processes designed to ensure that safety-critical automotive software can be developed and updated quickly, reliably, and with certifiability in mind.

The approach is validated in four representative use cases addressing key challenges of automated driving:

1. Field-data based validation and optimization
   Runtime monitoring of STPA derived safety requirements
   Continuous monitoring of metrics for increasing safety performance
2. Hybrid model development cycle combining virtual verification & validation
   Multiple vECU components deployed in the Cloud for scenario-based simulation
   Deployment of software updates on emulated environment (QEMUs) for real-time scenario-based validation
3. OTA updates for automated cooperative automotive systems
   How to incrementally develop and update software exemplarily deployed on the @ADORe autonomous driving platform
4. End-to-end regulatory and forensic documentation along the supply chain
   Ensuring integrity of the software, verifying authenticity and origin

Over 80 cutting-edge technology bricks have been developed to address the project's use cases. Integrated into the proposed ADSO-process and supported by a robust, continuous and consistent safety assurance framework for the development of automated driving functions. Together, they cover the technical steps required for a comprehensive safety assessment process and thus form the basis for the continuous further development of safety-critical software with certifiability in mind.

Fraunhofer IKS leaded the work package on technological fundamentals and contributes methods to ensure reliability during continuous software updates.

At the core was a contract-based design language linking system-level safety requirements with component guarantees. These contracts were verified by runtime monitors that detect deviations and trigger mitigation strategies in real time. All concepts were evaluated in both digital-twin and real-vehicle environments.

IKS key contributions include:

• Designing modular ODD-oriented architectures that are resilient to software updates.
• Developing modular and incremental innovative ML-based functions, enabling flexibility and targeted updates.
• Implementing and integrating the Construction Zone Assist automated function on the  APIKS platform, using  synthetic field-data-based validation to enhance safety performance through continuous monitoring of SPIs.
• Developing runtime monitoring concepts for safety-critical systems across the entire development lifecycle.

Through these contributions, Fraunhofer IKS demonstrated how safety and adaptability can be continuously reconciled in complex automotive software ecosystems.

Through AutoDevSafeOps, Fraunhofer IKS advances its research focus on resilient and trusted cognitive systems – adaptive architectures that maintain safe operation even under uncertainty and dynamic conditions.

These architectures continuously assess their context, adapt behavior dynamically, and preserve safety throughout their lifecycle.

The project’s results highlight the principles needed for the safe and efficient evolution of software-defined vehicles:

1. A generic lifecycle model is essential to integrate continuous, consistent safety assurance.
2. AI-based, safety-critical and updatable systems require ongoing specification, monitoring, and fallback mechanisms.
3. Continuous assurance of AI functions maintains trustworthiness over the entire lifecycle.
4. Early, simulation-based validation ensures both efficiency and safety in development and updates.

Building on these insights, Fraunhofer IKS is transferring the developed methods, architectures, and toolchains into new industrial collaborations — supporting partners in establishing safe, adaptive, and certifiable DevOps processes for the next generation of intelligent, autonomous, and connected systems.

Beyond the automotive sector, these approaches also strengthen future applications in robotics, industrial automation, and aerospace, where resilience, trustworthiness, and continuous adaptability are equally essential.

The project was funded by the Federal Ministry of Education and Research.

These publications were produced as part of the AutoDevSafeOps project at Fraunhofer IKS:

• A. H. Shamseddin, “Increasing Edge-case Testing Coverage through Guided AI Synthetic Scenario Generation,” Master’s Thesis, Technische Universität München, 2025.
• M. Liu, “A Multi-Modal Framework for Reliable Uncertainty-Aware Out-of-Distribution Object Segmentation,” Master’s Thesis, Technische Universität München, 2024.
• G. Gerloni, “Monitoring Framework for Continuous Safety Assurance in Autonomous Driving: A Construction Zone Use Case,” Master’s Thesis, Freie Universität Bozen, 2024.
• A. Castella, “Active learning in ML-training for autonomous vehicles,” Master’s Thesis, Technische Universität München, 2024.
• F. Budic, “Design and Implementation of a Scenario-Based Testing Pipeline for Autonomous Driving,” Bachelor’s thesis, Technische Universität München, 2025.
• Y. Ahmed, “Probabilistic Deep Learning for Safety Monitoring in Autonomous Driving Systems,” Master’s thesis, Technische Universität München, 2025.
• N. Breit, “Impact and Analysis of the EU AI Act’s Transparency Requirements and Their Relation to Existing Standards,” Bachelor’s thesis, Technische Universität München, 2024.
• A. Salvi, G. Weiss, and M. Trapp, “Explaining Unreliable Perception in Automated Driving: A Fuzzy-based Monitoring Approach,” May 2025, doi: 0.48550/arXiv.2505.14407.
• F. F. Okumus, J.-V. Zacchi, M. Salfeld, M. Schweizer, N. Mata, and S. Kugele, “Runtime Monitor Synthesis for Automotive Software Architectures,” in Software Architecture - 19th European Conference, ECSA 2025, in LNCS. Springer, 2025.
• L. Mareis, “Optimizing Experimental Design for Causal Effect Estimation with Partial Measurements,” PCIC, 2024.
• J. P. C. de Araujo et al., “Applying Concept-Based Models for Enhanced Safety Argumentation,” in ISSRE, IEEE, 2024, pp. 272–283.
• Y. Ahmed, F. Roza, and N. Mata, “Temporal Multimodal Probabilistic Transformers for Safety Monitoring in Autonomous Driving Systems,” Proceedings of Machine Learning Research, vol. 266. pp. 535–554, 2025. [Online]. Available: https://www.scopus.com/inward/record.uri?eid=2-s2.0-105013955850&partnerID=40&md5=5741d137cd9e63b5d7f3bc074d7eaab6
• M. Liu, H. Dong, J. I. Kelly, O. Fink, and M. Trapp, “Extremely Simple Multimodal Outlier Synthesis for Out-of-Distribution Detection and Segmentation,” in The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025.