Safety-critical autonomous robotic systems should be able to reliably execute their missions despite changes in the operating environment and system failures. Due to complex hierarchical architecture, efficient and correct implementation of fault tolerance is a challenging engineering issue. The states of the robots and progress of mission execution should be monitored at different layers and mission planning and execution should be adapted accordingly. In this paper, we formalise the main properties of fault-tolerant mission execution in Event-B. It allows us to derive a complex hierarchical architecture of fault tolerance mechanisms in a systematic correct-by-construction manner. The approach enhances design of robotic systems with mechanisms for monitoring mission progress and adaptation triggered whenever the progress is hindered by failures.