Social and Cognitive Interactions in the Open World

Social robots are often designed to serve and support human’s needs in the human environment. Therefore, the fundamental capability of social robots is to be able to perceive the dynamic interaction environments with multi-modal sensory inputs and extract the underneath abstract knowledge, so that the robots can reason on top of such knowledge to plan physical interactions with the environments, via grasping, navigation, speech, postures, etc. that are both legible and socially-acceptable to humans. Importantly, perceiving and understanding the interaction environments includes not only modeling the surrounding environments but also analyzing the basic principles of human verbal and nonverbal communication. Particularly, nonverbal signals, e.g., gaze, facial expressions, gestures, vocal behavior, can carry significant information regarding the status of the human interactors including their emotions, engagement, intentions, action goals and focus of attention, as well as personal traits over the long-term. Such rich nonverbal information in company with explicit verbal expressions completes the human feedback to the social robots, and helps them to learn and adapt towards a more natural decision making and action planning. However, building up such perception and cognition capability from verbal and non-verbal inputs still faces great challenges in the open world due to the large diversity and the evolving nature of the real-world interaction environment. Encountering the unknowns, i.e. the knowledge that is outside what the robot has learnt, can easily fail the robot from proper functioning. Yet, the day-by-day richer resources in terms of data and computational resource, and advances in deep learning methods, such as self-supervised learning, active learning, continual learning, large foundation models, etc., do signal a promising future for the development of social robots that are able to address open-world challenges and evolve its cognition capability adaptively. This special issue is dedicated to developing computational tools to advance the robot cognition capability for social and natural human-robot interaction in the open-world. We are calling for innovative research ideas, cutting-edge computational methods and models related to social robot cognition, as well as new simulation tools, datasets, benchmarks and evaluation protocols. We also welcome research that focuses on the decision making and the physical interactions that foster novel methods or closely couple with the understanding of cognitive interactions between robots and their interaction environments under the open-world challenges. The topics addressed in the special issue are listed below: ● Novel deep learning/machine learning methods for social robots with open-world challenges ● Transfer learning, meta-learning and continual/lifelong learning for social robots ● Self-supervised learning and foundation models for robotic perception ● Interactive/active learning for social robots ● Recognition of individual’s and/or social group’s activity ● (Real-time) recognition of human social signals, e.g., emotion, gaze, gesture, vocal activity, etc. ● Inference of human intentions for social robots ● Social cognition (joint attention, spatial perspective taking, action prediction, theory of mind) in human-robot interaction ● Human-robot engagement prediction ● Multimodal fusion for social robots ● Applications of human-robot interactions (e.g. for heath, education, entertainment, business) ● Ambient assisted living ● Social acceptance of robots based on human nonverbal behavioral cues (e.g., eye gaze, facial expressions, group formations) ● Human-robot interaction styles ● Reinforcement learning and learning from demonstration for social robot human/environment interactions ● Novel methods for navigation in social interaction scenarios