Teaching Robots Style: Designing and Evaluating Style-by-Demonstration for Interactive Robotic Locomotion

Abstract

In this article we present a multipart formal design and evaluation of the style-by-demonstration (SBD) approach to creating interactive robot behaviors: enabling people to design the style of interactive robot behaviors by providing an exemplar. We first introduce our Puppet Master SBD algorithm that enables the creation of interactive robot behaviors with a focus on style: Users provide an example demonstration of human–robot interaction and Puppet Master uses this to generate real-time interactive robot output that matches the demonstrated style. We further designed and implemented original interfaces for demonstrating interactive robot style and for interacting with the resulting robot behaviors. Following, we detail a set of studies we performed to appraise users' reactions to and acceptance of the SBD interaction design approach, the effectiveness of the underlying Puppet Master algorithm, and the usability of the demonstration interfaces. Fundamentally, this article investigates the broad questions of how people respond to SBD interaction, how they engage SBD interfaces, how SBD can be practically realized, and how the SBD approach to social human–robot interaction can be employed in future interaction design.

Notes

Background. This article is based on the Ph.D. thesis of the first author, where sections were rewritten to emphasize the evaluations and further analysis was performed to arrive at the final implications and guidelines.

Acknowledgments. We thank members of the Interactions Lab at the University of Calgary and the Japan Science and Technology ERATO Design Interface Project for their ongoing support of various aspects of this work. In addition, we thank the excellent HCI journal reviewers and editorial board for their hard work and involvement; your input was invaluable for improving and solidifying our paper.

Support. This work was supported in part by the Canadian Natural Sciences and Engineering Research Council, Japan Science and Technology, Japan Society for the Promotion of Science, and various local grants at the University of Calgary and the University of Tokyo.

HCI Editorial Record. First manuscript received June 29, 2011. Revisions received November 28, 2011 and March 5, 2012. Accepted by Brad Myers. Final manuscript received April 22, 2012. — Editor