Safety Case


Safety and Verifiability – Michael Fisher

Research Challenges (T4RC):

  • 1

    New RAI Systems must be Designed for Verifiability. Ad-hoc development of RAI solutions often leads to opaque systems with little clarity over where (in which embedded component) key decisions are made
  • 2

    New Verification Techniques for Nuclear Systems. A new aspect of autonomous systems is that key decisions will now be made by software, rather than by an operator. Consequently, it is important to be able to assess not just what any RAI system will do, but why it chose to do this
  • 3

    RAI must be designed for Long-term Autonomy In nuclear scenarios. The ability of these systems to self-monitor, diagnose, and reconfigure, will be essential to ensure they can remain effective for long periods without close support
  • 4

    Comprehensive V&V. In assessing nuclear RAI it must be ensured that formal verification, simulation, and testing (both virtual and field) are linked together in a coherent way and across sub-components, full systems, and multi-agent systems
  • 5

    Evolving Standards/Regulations relating to nuclear RAI. As existing certification/standards approaches typically do not work well for novel RAI solutions, stakeholders must be engaged with to provide input/advice/techniques to support the evolution of these standards/regulations