Research Talk: Belief-Modeling and Inference in Human-Robot Interaction
Advances in robotics are allowing robots to take on increasingly complex tasks without the need for human micro-management, necessitating the development of high-level human-robot interaction (HRI) competencies. Constructing mental models, that is the set of beliefs, that other agents in an interaction possess is one vital capability. In this talk, I discuss the progress made at the HRI- Laboratory to develop a natural language (NL) dialogue system that supports and facilitates mental-modeling.
To facilitate mental modeling, our system utilizes adverbial cues that are routinely employed by humans. I present a novel algorithm that integrates adverbial modifiers with belief revision and expression, phrasing utterances based on Gricean conversational maxims. The algorithm is then demonstrated in a simple HRI scenario.
I also introduce my research at the HRI-Lab in the field of robot ethics, which seeks to investigate questions concerning how to engineer robots and human-robot interactions in ways that promote ethical outcomes (and avert unethical ones). The importance of belief modeling and inference capabilities in ensuring ethical behaviors from robots is discussed and avenues for future research in this domain are explored.