Investigating the Effects of Robots Reacting to Human Backchanneling on Human-Robot Interaction

Sammanfattning: In an interaction between two participants the listener continuously gives feedback to the speaker in the form of gestures and verbal utterances; this is called backchanneling. An example of commonly used backchanneling is nodding or the verbal utterance “yeah”. Backchanneling improves the quality of the interaction as well as the participants' experience. In human-robot interaction the outcome of a robot reacting to human backchanneling has been poorly studied. In this thesis, it is investigated how a user study which examined the effects of a robot reacting to human backchanneling can be designed. Furthermore, how a robot responding to human backchanneling impacts the human’s tendency to backchannel towards the robot. In addition, how it affects specific backchanneling, encouraging/non- encouraging backchanneling. Earlier studies on interaction were reviewed and taken into consideration for the design of the study. In addition, three pilot studies were executed to get an overview of design improvements. The participants involved in the pilot studies were interviewed after completing their part of the pilot study and the result from the pilot studies was taken into consideration for structuring the design of a user study. After the design of the user study was realized, the study was conducted containing 31 participants, four of the 31 participants were later excluded from the study. The participants were randomized into a control and an intervention group. The designed user study contained a one-on-one interaction with a Pepper-robot and a participant in a controlled area; this was repeated for all participants in the study.  The Pepper-robot was controlled with a Wizard of Oz user interface during the interactions with the participants. For the intervention group the Pepper-robot responded to the participant’s backchanneling. This did not occur for the control group. The dialog with the Pepper-robot was otherwise the same between the two groups. The Pepper-robot’s utterances were pre-programmed. The interaction with the Pepper-robot was divided into five separate parts: introduction, jokes, riddles, game and ending. Each part contains a different topic of conversation. The dependent variables in the experiment were the number of times the participant performs backchanneling, as well as the number of times specific backchanneling behaviour were executed (encouraging/non- encouraging). The interaction between the participants and the Pepper-robot was video recorded. Video-annotation of the interactions was executed. The first section in the interaction (introduction) was excluded from the video-annotation due to the time limitation of the project. The user study concluded that the human’s tendency to backchannel was not affected by a robot reacting to the human’s backchanneling with the exception regarding the joke section of the interaction where the human’s tendency to execute a nod was higher for the intervention group.