Abstract:

A basic task for robots interacting with humans consists in guiding their focus of attention. Existing guidelines for a robot's multimodal deixis are primarily focused on the speaker (talk-gesture-coordination, handshape). Conducting a field trial with a museum guide robot, we tested these individualistic referential strategies in the dynamic conditions of real-world HRI and found that their success ranges between 27% and 95%. Qualitative video-based micro-analysis revealed that the users experienced problems when they were not facing the robot at the moment of the deictic gesture. Also the importance of the robot's head orientation became evident. Implications are drawn as design guidelines for an interactional account of modeling referential strategies for HRI.