Assured resilience in autonomous systems – automotive and Industry 4.0

Given that techniques such as neural networks do not lend themselves to external scrutiny and validation, they are however not suitable for use in safety-critical systems like driverless transport systems or autonomous vehicles, thereby hindering the successful introduction of safe autonomous systems. Consequently, the ADA Center is researching new methods to sufficiently assure the resilience of AI procedures in high-performance embedded system platforms. To this end, we are developing monitoring mechanisms that demonstrably monitor all properties crucial to the safety of the autonomous system. In addition, we are developing a flexible resilience concept designed to be usable both for different learning methods, and for different sensor configurations. For instance, if partially explainable machine learning methods (such as Explainable AI) are deployed, this additional information will contribute to resilience assurance. Furthermore, the ADA Center is researching appropriate AI methods for the monitoring itself, for use as redundant paths, or for monitoring of non-explainable machine learning processes.