A novel tool for delivering learning content to autistic children in a manner that keeps the children engaged, is tailored towards their strengths, and improves learning gains"; this is the promise of using robots to enhance interventions for autistic children. While robots have been found to pique the children's interest and improve engagement in interventions, designing robots to sustain long-term engagement that leads to learning is difficult. The children are very different from each other in how autism affects the development of their cognitive, language, and intellectual ability, which needs to be taken into account for the child-robot interaction. How this can and should be done is still an open question -one that will be addressed in this dissertation.In the first part of this dissertation, I describe our research in developing a novel robot-assisted intervention for teaching the basics of emotion recognition to autistic children. Our research was part of the EU-project DE-ENIGMA. A key research question that we address is how we can design a robot-assisted intervention to engage autistic children in learning. To address this broad research question, we report a descriptive study where autistic children interacted with our initial prototype of the "DE-ENIGMA robot-assisted intervention". In this study, we report on how the children's individual differences are correlated with how the children interacted within the intervention. Furthermore, we conducted a literature study to assess the user needs and user requirements that are relevant for developing robot-assisted interventions for autistic children, and describe how we developed the DE-ENIGMA intervention.The second part of this dissertation is related to a concept that is central to autism: predictability. Autistic children are believed to generally favour predictable environments, and contemporary Bayesian accounts of autism place the inability to effectively deal with unpredictability at the core of the condition. Because robots are programmable, we could, in theory, program them to be highly predictable. By doing so, we could address this need for predictability and possibly improve the engagement of autistic children to the intervention. In fact, the highly predictable nature of robots is a commonly used argument for why robots may be promising tools for those working with autistic children. Predictability, however, is poorly defined in current literature and lacks an operationalisation that we can use for manipulating a robot's predictability. We therefore provide a novel formalisation and operationalisation of predictability, and how it relates to human-robot interaction, based on the predictive processing framework. Furthermore, we report on two experimental studies where we investigated the effect of a robot's predictability on the studies' participants. In viii | Abstract one study, we specifically look at people's social perception of a robot in relation to its predictability. In the other study, we investigated the effect of a robot's predic...