Evaluating multimodal human-robot interaction

Abstract

Multimodal natural behavior of humans presents a complex yet highly coordinated set of interacting processes. Providing robots with such interactive skills is a challenging and worthy goal and numerous such efforts are currently underway; evaluating the progress in this direction, however, continues to be a challenge. General methods for measuring the performance of artificially intelligent systems would be of great benefit to the research community. In this paper we describe an approach to evaluating human-robot multimodal natural behavior. The approach is based on a detailed scoring and spatio-temporal analysis of the structure and patterning of live behavior, at multiple temporal scales, down to the decisecond level. The approach is tested in a case study involving an early virtual robot prototype, Gandalf, which is capable of real-time verbal and non-verbal interaction with people. Our analysis includes a comparison to a comparable human-human dyadic interaction scenario. Our main objective is to develop a methodology for comparing the quality and effectiveness of human-robot interaction between a wide variety of such systems. Early results indicate that our approach holds significant promise as a future methodology for evaluating complex systems that have a natural counterpart.