Towards a Cognitive Architecture for Socially Adaptive Human-Robot Interaction

People have a natural predisposition to interact in an adaptive manner with others, by instinctively changing their actions, tones and speech according to the perceived needs of their peers. Moreover, we are not only capable of registering the affective and cognitive state of our partners, but over a prolonged period of interaction we also learn which behaviours are the most appropriate and well-suited for each one of them individually. This universal trait that we share regardless of our different personalities is referred to as social adaptation (adaptability). Humans are always capable of adapting to the others although our personalities may influence the speed and efficacy of the adaptation. This means that in our everyday lives we are accustomed to partake in complex and personalized interactions with our peers. Carrying this ability to personalize to human-robot interaction (HRI) is highly desirable since it would provide user-personalized interaction, a crucial element in many HRI scenarios - interactions with older adults, assistive or rehabilitative robotics, child-robot interaction (CRI), and many others. For a social robot to be able to recreate this same kind of rich, human-like interaction, it should be aware of our needs and affective states and be capable of continuously adapting its behaviour to them. Equipping a robot with these functionalities however is not a straightforward task. A robust approach for solving this is implementing a framework for the robot supporting social awareness and adaptation. In other words, the robot needs to be equipped with the basic cognitive functionalities, which would allow the robot to learn how to select the behaviours that would maximize the pleasantness of the interaction for its peers, while being guided by an internal motivation system that would provide autonomy to its decision-making process. The goal of this research was threefold: attempt to design a cognitive architecture supporting social HRI and implement it on a robotic platform; study how an adaptive framework of this kind would function when tested in HRI studies with users; and explore how including the element of adaptability and personalization in a cognitive framework would in reality affect the users - would it bring an additional richness to the human-robot interaction as hypothesized, or would it instead only add uncertainty and unpredictability that would not be accepted by the robot`s human peers? This thesis covers the work done on developing a cognitive framework for human-robot interaction; analyzes the various challenges of implementing the cognitive functionalities, porting the framework on several robotic platforms and testing potential validation scenarios; and finally presents the user studies performed with the robotic platforms of iCub and MiRo, focused on understanding how a cognitive framework behaves in a free-form HRI context and if humans can be aware and appreciate the adaptivity of the robot. In summary, this thesis had the task of approaching the complex field of cognitive HRI and attempt to shed some light on how cognition and adaptation develop from both the human and the robot side in an HRI scenario.