Assuring Safety of AI-based Systems: Lessons Learned for a Driverless Regional Train Case Study

Artificial Intelligence (AI) offers great potential to enable the fully automated operation of trains.Mandatory novel functions to replace the tasks of a human train driver, such as obstacle detection on the tracks, can be realized using state-of-the-art Machine Learning (ML) approaches.However, the use of AI/ML to implement perception tasks in the railway context poses a new challenge: How to link AI/ML techniques with the requirements and approval processes that are applied in the railway domain in practical way?Within the safe.trAInproject we laid the foundation for the safe use of AI/ML to achieve the driverless operation of a regional train.Based on the requirements for the certification process in the railway domain, safe.trAIninvestigated methods to develop trustworthiness AI-based functions, taking data quality, robustness, uncertainty, and explainability aspects of the ML model into account.In addition, the project developed a safety argumentation strategy for an AI-based obstacle detection function of a driverless regional train.In this paper, we describe the challenges to assess an AI-based obstacle detection function according to the given regulation in the railway domain.Moreover, we describe our safety assurance strategy applied to our case study in the safe.trAInproject.