The environments and the goals that an autonomous system needs to be able to understand and act upon are not fully known and clearly identifiable at design time, especially for those systems that interact with the physical world. Still, for applications that require a certain level of assurance, requirements, which include the definition of how a system is supposed to interact with its context, need to be defined, analyzed, refined towards an implementation that can be shown to deliver certain guarantees. We first broadly discuss the importance of the early stages of the design process where requirements are defined, and we then discuss the formalisms needed to capture requirements and refine them towards implementation for autonomous systems.