Towards the automatic observation and evaluation of ethologically inspired Human-Robot Interaction

Abstract

Human-Robot Interaction (HRI) evaluation is rapidly gaining more interest as the field of social robotics advances, but determining the effectiveness of the various HRI models can be a challenging task. Recently, ethology has been successfully incorporated in HRI models, and has been suggested to be used in HRI evaluation methods also. In ethological experiments, trained observers manually identify and record the exhibited behaviours of the participating subjects, a procedure called as coding. However, precise observation and coding of the behaviour of subjects is a resource demanding process and prone to human mistakes. Therefore, the goal of the paper is twofold. First, to introduce automatical coding as a possible method to provide accurate, unbiased, objective evaluation of HRI experiments. Second, to abolish the need for manual human work in the coding process. Accordingly, we propose a possible method which is able to automatically code simple behaviour elements in HRI experiments. The task of the proposed system is to automatically evaluate the behaviour of a mobile robot by distinguishing the cardinal behaviour elements without any information on the inner states of the model controlling the mobile robot, based solely on data gathered from the environment of the interaction. Various test scenarios were evaluated, while the mobile robot agent was exhibiting pre-programmed behaviours, where the proposed system could properly recognize and code the behaviour elements shown by the mobile robot.