Fault Injecting Co-simulations for Safety

摘要

Robotic systems, either operating alone or in fleets, in order to be trusted, have to provide certain safety guarantees to ensure no harm comes to neither humans, environment, other robots or equipment. Co-simulation tools enable the integration of models from different modelling applications and can be adopted to ensure the safe behaviour of such systems using simulation. Different parameters of the included models can be explored, including safety aspects, by simulating different sets of scenarios. Of particular interest are those scenarios in which different system components fail at different times, potentially leading to safety violations. The core contribution of this paper is a simple but powerful fault injection mechanism, that enables developers to seamlessly inject faults in their co-simulations. We present an initial evaluation of the proposed mechanism through a water-tank case-study, to discuss its applicability and benefits for system evaluation during the development stage. Thereafter, we discuss the value of this work in the context of autonomous agricultural robot fleets, and outline our next steps in this direction.