Artificial Intelligence is currently limited to non‑safety‑critical railway scenarios because the sector’s standards and regulations are extremely strict. To raise the trustworthiness of data science and AI algorithms and achieve compliance, three pillars must be addressed: robustness, Operational Design Domain (ODD) and explainability.
Robustness is the ability of an AI system to maintain its performance level under any circumstances, as defined by ISO24029.
ODD, defined by DIN DKE SPEC 99004, explicitly specifies the operating conditions a system is intended to handle, making its behavior bounded and verifiable.
Explainability demands that an AI system present the key factors influencing its outcomes in a form that humans can understand.
These three research areas are already mature outside the railway sector, with algorithms and methods ready for direct adoption.
A system‑level view is required to keep all trust requirements continuously interacting within a safe MLOps environment, thereby gaining acceptance from regulators, operators and the public.
Beyond safeguarding safety‑critical use cases, establishing deep trust aligns with worldwide regulatory frameworks, unlocking AI’s full potential and accelerating its adoption in mission‑critical domains.
Review