Rebecca Stower

• As children develop, they learn to copy from others, interpret their intentions, and evaluate their reliability as sources of information (social learning). Today, these ‘others’ include not only children's peers, parents, or teachers, but also technological devices with which they can interact, including in educational settings. Social robots occupy a particular niche among such technologies, with the ability to embody specific social behaviours such as gaze, gesture, and verbal communication. Nonetheless, the technology underlying the design of social robots remains far from perfect, and opportunities for failures are rife. Understanding how the social capabilities inherent with social robots interact with these inevitable failures is therefore necessary for the design of robust educational interactions with robots. Consequently, the goal of this thesis is to develop an understanding of how errors impact children's social attitudes and behaviour towards social robots. First, a meta-analysis on children's trust in social robots was conducted, creating a theoretical baseline from which to study robot errors. Second, a learning task and measurements for use in child-robot-interaction (cHRI) were developed, through which robot errors could be manipulated and their effect on constructs such as trust, liking, and perceived agency captured. Third, the role of robot errors in a real-world implementation of the learning task was examined. Finally, how current social cognition paradigms can be used to explain children's perceptions of robot errors was explored. Throughout these studies, a theme emerged towards errors not being detrimental to children's perceptions of social robots, at least for more short-term interactions. The significance of robot errors and, more generally, social behaviour when evaluating social robots as tutors is discussed. The thesis concludes with recommendations for how cHRI research can draw from psychology in the design of future cHRI studies.