Proving properties about robotic systems with humans-in-the-loop relies on assumptions about human behaviour. Existing technologies require expertise not reasonably expected from psychologists and human-factors engineers, for instance. A user-needs analysis of industrial design techniques for human-robot interaction has identified a lack of standardised approach. We present RoboScene, a notation based on UML sequence diagrams that can be used to capture assumptions derived from human-factors artefacts, through novel constructs enabling consideration of stakeholders with different traits. We describe a tock-CSP semantics for RoboScene, and show how we can connect (mathematically) RoboScene diagrams to platform-independent software models. This is applied in the context of a Human-Centered Engineering process, demonstrated via an industrial case study.