The paper explores the interactional dynamics within groups of visitors (children & adults) who are interacting with an autonomous museum guide robot. Using the example of question-answer sequences, analysis reveals that 'answering' tends to be a process of interactional negotiation at the end of which a 'result' is addressed to the robot. This process, however, is often not transparent to the robot and considered solely as "silence" which leads to the disregard of the users' attempts to provide an answer.

On the basis of a video corpus of human-robot-interaction in a museum guide scenario, we address the challenge how a robot could determine the appropriate moment of opening an interaction with a visitor. Considering such "openings" as a stepwise coordinated process, we used a wizard-of-oz setup to investigate the wizard's timely decisions about when to look at the visitor resp. to verbally greet him/her (during early stages called "pre-beginnings"). Analysis is based on both robot internal and external data and combines qualitative sequential case analysis and quantification. Analysis reveals features for classification of two visitors types and corresponding opening strategies.

An autonomous robot system was equipped with basic means to monitor the users' success/failure in following a robot's verbal-gestural deictic reference to an object and - in case of problems - to provide additional help, i.e. to suggest a `repair' action. A real-world field trial with the robot acting as museum guide constitutes the basis for analysis of the users' reactions to the first reference and the subsequent `repair' action in two structurally different conditions. Results indicate that deictic `repair' actions are mostly successful helping users to correctly orient, but they may also lead to confusion for initially correctly oriented visitors in situations of small groups of users. A recurring user practice is revealed which consists of displaying to co-visitors their understanding of the reference by pointing to the location and thus providing additional orientational help.

On the basis of a videocorpus of human-robot-interactions in a museum guide scenario and with a combined approach of qualitative case analysis and quantitative evaluation we examine two research questions: (1) What type of group dynamics result from a robot's unexpected next move - especially focussing on gaze patterns? (2) What should a robot detect when searching for trouble events and how should it identify patterns of visitor behavior that are related to the robots actions? Systematically studying opening sequences, we extract structural gaze patterns out of two cases considered from an interactional linguistics perspective as well as keeping in mind a system's capabilities of perception.

Once a museum guide robot is deployed in the real world, it is faced with the difficulty to deal with user utterances within a noisy environment. We present explorative analysis of the ways in which visitors signal trouble on a multimodal level when they have difficulties in understanding the robot. Such moments of trouble are investigated in the context of robot-to-group situations. Analysis will give insights into the visitors conduct at moments of trouble in robot-to-single-user and robot-to-group interactions. In a limited data set, we found that members of a group show internal gaze coordination between each other when trouble occurs due to the robot's inadequate behavior. Implications will be drawn for further developing the interactional capabilities of an autonomous robot system to be attentive and responsive to the users' needs.