HCC: Small: Social Agents and Robots for Open-Ended Domains

点击翻译按钮获取中文摘要

Recreational activity is a fundamental aspect of human existence and an important part of the human condition within familial and social groups, where it serves to strengthen social ties by increasing affect between individuals and as a form of education in creative thinking. Despite a sizable accumulation of knowledge about such activity from sociology, anthropology, and psychology, "play" as a first-class concept has not been studied through the lens of computation. When today's agents engage with humans, they do so in the context of structured environments and are highly dependent on well-defined goals and/or behaviors. Contrast this to the domain of pretend play, which involves non-goal directed peer-to-peer activity in a shared imaginary second reality that is continually altered. Pretend play is a common form of engagement that is relevant to an array of social domains, such as elder care, peer learning, or social skills therapy for children with autism spectrum disorders. The PI's goal in this research is to imbue robot systems with procedural and declarative representations of play so that they are capable of engaging in such activity with humans as peers. The work aims to discover how to develop humanoid robots with the ability to engage, improvise, and create with humans in unstructured environments. Such robot capability would foster perceptions of lifelikeness, social acceptance, and companionship similar to the experience of playing with other people. These agents would encourage spirited behaviors and creativity in people. They would elicit high levels of interest, intrinsic motivation, and positive affect, which in turn would lead to better concentration, learning, and personal investment by the human participant. To achieve these goals, the PI will leverage his prior work on creativity and cognition to conduct a study of adults engaging in object-based pretend play to elicit a formal understanding of it in dyads. The findings will subsequently be applied to building social robots that engage, based on the team's expertise in human-centered AI and human-robot interaction. The resulting robot architecture will be evaluated to see how it can enhance robot affect and social acceptance. Broader Impacts: This research will create a new academic research direction of Computational Play within the field of social robotics that has the potential to contribute a solid and unique advance to the field, and also to change how we interact with intelligent agents thereby increasing agents' social value and acceptance by the humans around them. The work will increase via empirical study our understanding of human engagement, and in particular of the knowledge and social dynamics involved in pretend scenarios. The playful robots to be designed, implemented and formally evaluated in this work will inform the human-robot interaction community as to how such activity can be used within HRI contexts to increase affect. This work also has the potential of improving the learning and creativity of those that interact with social agents, making computational play a valuable research direction for education. The project will provide a fertile ground for interdisciplinary training of graduate and undergraduate students, and a wealth of interaction data that will be shared with the scientific community.