Making a Case for Spatial Prompting in Human-Robot Communication

In this paper we present an analysis of a set of examples of how verbal and non-verbal behavior of a service robot influence users’ way of positioning themselves during interaction, using concepts from theories of non-verbal behavior. Based on the analysis we propose a design case where a robot utilizes a (naïve) spatial prompting strategy to influence the spatial positioning and communicative behavior of the user. INTRODUCTION A design requirement of a personal service robot is that it should be configured and provided with work tasks by the user in an interactive and intuitive way. These robots are intended to provide service tasks in the home, possibly offering wide range of services. Typically they are envisioned to be equipped with multimodal spoken dialogue systems, to reduce the complexity in the user interface. In this paper we argue that theories of spatial positioning need to be considered when developing the communicative system of the robot. Furthermore we present an empirical account of the way spatial behavior of robots influence human users. We also propose the term spatial prompting, which refers to active strategies of the robot that are intended to influence users to position themselves in a way that is advantageous for further communicative actions. Positioning, as it has been approached as a research challenge for human-robot interaction, is considered as providing adaptive physical movements of the robot. A result of this is that the communicative dimension of positioning typically has been ignored in systems that interactively position themselves in relation to their users. One requirement that is typically put forward is that the robot should position itself in a socially appropriate manner [1, 4]. The parameters that concern these approaches are typically derived from research on nonverbal behavior. In robotics the problem of maintaining the robot localized and situated within a geometric representation of the world has been framed as the Simultaneous Localization and Mapping (SLAM) problem [13]. Recent advances in Human-Robot Interaction (HRI) have raised the interest in detecting and tracking the position of users during interaction. When the position of the user is known, the robot can plan how to position itself [1]. The research on spatial reasoning applied to robotics is well advanced but primarily focused on natural language understanding of spatial relations, providing for exchanges concerning locations of objects in the environment [3]. RESEARCH ON SPATIALITY IN COMMUNICATION There are several research approaches for human-human that are relevant for spatial management between humans and robots. Hall [14] studied interpersonal distances and distinguished four different distances: intimate (0-1.5 ft), personal (1.5-4 ft), social 4-12 ft, and public (> 12 ft). These distances vary both with respect to the current activity and cultural factors. Another dimension that is relevant to spatiality is the concept of territoriality, according to Sack, i.e., “the attempt by an individual or group to affect, influence, or control people, phenomena, and relationships, by delimiting and asserting control over a geographic area” [15]. Kendon [12] also studied the spatial configuration of the participants, using the term F-formations, for instance the L-shape which describes the relation when two participants have a common visual focus. The shared space, the so called o-space, or the transactional space is then located in front of the participants, and it is within this area that the interaction is conducted. Clark [5] refers to this space as the workspace, where perceptual co-presence is established between speakers [5, 10]. In this context, research on perception and especially visual perception plays an important part for maintaining common ground between participants [10, 8]. Gill [9] has investigated the communicative effects that participants achieve by using nonverbal behavior, focusing on the functional rather than the morphological perspective of nonverbal behavior. One such function is the category focus which is a metadiscursive function that signals a shift in the center of attention in the discussion, e.g., a shift in body posture with the same meaning as the utterance “I am going to focus on this spot”. Another, less obvious, but nevertheless important concept is Schegloff’s notion of body torque [15], a state of the bodily configuration when two different body segments are oriented in different directions. According to Schegloff [15] Body torque “project change”, i.e., when some part of the body is organized in an unstable way, the participants may predict that a change in posture is pending. For instance, when turning the head, this might predict a change of the general body orientation. During interaction, speakers monitor the action of others, interpreting purposeful actions that lead towards a common joint goal as compliance [10]. Human-robot interaction is situated in a physical context, where understanding and reference to actions of the human partner during interaction explicitly needs be taken into account. This makes research on virtual collaborative [6] environments interesting also in this context, since it is concerned with models that explicitly represent spatiality and reference. CORPUS ANALYSIS OF SPATIAL MANAGEMENT We have analyzed a video corpus, collected in a European project [7], containing transcribed data of about 20 user sessions, (approximately 20 minutes each) where a user talks to a robot and teaches it the names and locations of objects using a combination of gestures and speech. By viewing the video corpus we identified and analyzed instances where the robot movements or verbal actions appear to influence the actions of the user. The examples reflect three different ways in which the robot actively influences the user to act: • Primary verbal: by using a spoken command • Primary non-verbal: by movements • Multi-modal: using movement as trigger for a verbally specified (or grounded action) 00:32:889-00:34:071 R6: Robot is following 00:40:603-00:41:744 R7: You are too fast 00:44:690-00:46:539 R8: Please stand in front of the camera 00:49:783-00:50:883 R9: Robot is following User