Towards Adaptive Human-Robot Interaction: A Comfort-Driven Architecture for Social Robots

As social beings, humans are naturally predisposed to interact with others in an adaptive and emphatic manner. We can perceive our peers’ needs and affective states and adjust our behavior accordingly; moreover, during prolonged interaction, we learn which behaviors are the most appropriate and suitable for each of them. Although our individual differences, such as personalities and interaction styles, can influence our speed and tendency to adapt, we can efficiently establish adaptive and personalized interactions with our peers. Replicating the same kind of complex human-like interaction is a hard problem and remains an open challenge in Human-Robot Interaction (HRI). Enabling robots to reach human-like levels of social competence would require endowing them with basic cognitive functionalities and an internal drive that allows them to understand their partners’ states during the interactions, adapt their behavior accordingly, and act in ways that are understandable for users. My PhD research has the threefold objective of (i) designing a cognitive architecture to endow social robots with perceptual, motivational, and decision-making systems and their own personal traits; (ii) exploring the effects of this framework during interactions with the users on their experience and perception; (iii) including elements of adaptability in the framework and study their impact during human-robot interaction. To achieve this goal, my approach proposes a cognitive framework that supports social awareness and adaptation in robots. Drawing inspiration from the attachment theory, I assume that, as it happens in human-human interaction, a mismatch in the interaction style of a robot and a human can cause an uncomfortable interaction. Stating this, I equip robots with a motivational system based on comfort, which indicates how pleasant robots perceive the interaction. The comfort motivation is dynamic and is influenced by both the robot’s needs and goals and the stimuli it receives during the interaction with the user. The aim of this research is to allow robots to understand their partner’s interaction style and autonomously adapt their behavior accordingly by monitoring their motivation. To evaluate this approach, I developed a modular cognitive architecture consisting of sensory, perceptual, motivational, behavioral, and motor modules that operate simultaneously. I deployed it on the robotic platform iCub and validated it by replicating a paradigm from social science. Then, I tested it in HRI studies with the robot iCub and the Toyota Human Support Robot (HSR). With these studies, I aimed to understand how the cognitive framework behaves in an ecological free-form interaction and in a collaborative scenario. Moreover, I attempt to explore whether humans can be aware of and appreciate robots' behavior. Findings indicate that equipping robots with a comfort-driven architecture enhances their perception as goal-oriented and affective agents and fosters user engagement and a sense of collaboration during the interaction. Furthermore, participants appeared to respond more intuitively to robots exhibiting motivated behaviors, demonstrating a greater inclination to follow the robot’s suggestions. Overall, the comfort-driven cognitive architecture shows promising outcomes and enhances robots' social interactions. Moreover, it has a modular design that facilitates the integration of additional capabilities and makes it adaptable to different robotic platforms. The proposed framework can potentially be applied in different interaction contexts. For instance, it could be used in healthcare, where robots could offer emotional and cognitive support, and in educational settings, where robots could provide personalized assistance and tutoring. Future research could refine this framework to enrich the perception module with speech recognition and explore its impact on long-term human-robot interaction and group-robot interactions. This research advances the development of socially adaptive robots, bringing us closer to designing agents that are not simply tools but interactive partners capable of engaging in intuitive and human-like interactions.