Decision Theoretic Approaches to Human-Robot Social Interaction

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Robotics and Computer Vision Program ABSTRACT Proposal #: 0329014 Title: Decision Theoretic Approaches to Human-Robot Social Interaction PI: Nourbakhsh, Illah Reza Carnegie Mellon University Social rules govern patterns of movement and allow individuals to share a space without interfering with each other's goals. These social rules of movement are rich and complicated, being highly dependent on context and governing both spatial and temporal relationships. The research question at the heart of this project is: can robots utilize cues based on human behavior in deciding when and how to apply the correct rules for social interaction? There are two main components to this question: how to acquire or design the representations of these rules and cues, and how to use this information in a robot's decision-making process. We approach this as a planning problem because a social interaction requires a sequence of actions rather than a one-time, reactive choice. We intend to extend existing decision-theoretic planning methods to semi-Markov models, which can represent more complex relationships between time and state than first order Markov models. This project has potential impact in the research fields of Artificial Intelligence and Robotics as well as impact on numerous modern applications. It can result in making robots more likely to be deployed in environments where their capacity to be embodied information providers is most useful (museum guides, receptionists, helpmates for the elderly, etc.). This project is also promoting education for underrepresented groups like women and minorities.