Specification of online verifiable safety constraints for critical autonomous systems

Recent progress in the definition of decisional mechanisms has allowed computer-based systems to become more and more autonomous. For example, service robots can nowadays work in direct interaction with humans and carry out increasingly complex tasks. This transfer of responsibility poignantly raises the issue of system safety towards humans, the environment and the system itself. System surveillance by an independent safety monitor aims to enforce safe behaviour despite faults and uncertainties. Such a monitor must detect potentially dangerous situations in order to trigger safety actions aiming to bring the system towards a safe state. This thesis addresses the problem of identifying safety trigger conditions. A systematic process is proposed for the identification, starting from a HazOp/UML risk analysis. The proposed methodology also allows the identification of system states in which multiple safety actions might be executed concurrently, in order to be checked and, if necessary, corrected by a system expert. The methodology is applied to a robotic rollator.