TL;DR
This paper investigates whether human teleoperators controlling mobile robotic telepresence systems naturally adopt "F-formations" (spatial arrangements for social interaction) when joining groups of virtual agents. The study found that teleoperators do indeed adhere to these formations, but they also expressed a preference for autonomous features to assist in precise placement, suggesting a need for enhanced HRI design in telepresence systems.
What Problem Does This Paper Address?
Mobile Robotic Telepresence (MRP) systems are increasingly used for remote social interactions. A crucial aspect of human social interaction is proxemics, the study of interpersonal distances and spatial arrangements. F-formations, a specific type of spatial organization, describe how groups of people position themselves for effective communication. While F-formations are well-studied in human-human interaction, there's a gap in understanding whether teleoperators of MRP systems, navigating remotely, naturally adopt these formations when joining existing social groups. This paper aims to bridge this gap by exploring two key questions: 1) Do teleoperators of MRP systems naturally adhere to F-formations when joining social interactions? and 2) If not, is this due to technological limitations, and would they prefer autonomous assistance?
Methods
The researchers designed an experiment within a simulated conference lobby environment populated with virtual agents (VAs) arranged in various F-formations. Teleoperators controlled a simulated TurtleBot robot from an egocentric perspective (first-person view from the robot's camera) to navigate and join these VA groups. After the teleoperation task, subjects reviewed their performance from an exocentric perspective (third-person view) and provided feedback.
To analyze the adherence to F-formations, the study employed a game-theoretic model based on Vascon et al. [23]. This model uses the position and orientation of VAs and the robot to generate 2D "frustums" (representing field of view and attention). By computing pairwise affinities between these frustums, the algorithm can detect and quantify F-formations, providing precision, recall, and F-measure scores. Qualitative analysis involved interviewing subjects about their placement choices and reviewing video recordings.
Datasets
Not discussed in the paper. The study utilized a custom-built simulation environment.
Key Insights and Ideas
- Innate Social Navigation: Even when operating a robot remotely and with limited sensory input (egocentric view), humans seem to possess an inherent understanding and tendency to adopt socially appropriate spatial formations. This suggests that social cues are deeply ingrained and can translate even into mediated interactions.
- Egocentric View Limitations: The egocentric perspective, while providing a realistic experience, significantly hinders precise spatial judgment. Subjects often struggled to accurately perceive distances and select optimal spots within formations, leading to less precise placements than intended.
- Preference for Assistance: The strong desire for an "autonomous feature" to place the robot within formations highlights a critical design implication for MRP systems. It indicates that current interfaces and sensory feedback are insufficient for complex social navigation tasks, and automation could significantly improve user experience and interaction quality.
- Formation Modification: The act of a robot joining an existing F-formation can itself modify the formation, creating new emergent patterns. This suggests that the integration of robots into social groups is not merely about fitting in but can actively reshape the social dynamic.
Key Results
- F-Formation Adherence: The study concluded that teleoperators do respect F-formations when joining social groups. Quantitative analysis using the Vascon et al. algorithm showed scores of 1 for precision, recall, and F-measure, indicating the robot was successfully integrated into the formation.
- Placement Preferences: The majority of subjects (68%) stated their primary reason for placing the robot was to "see the faces" of the VAs, followed by finding an "ideal position" (54%).
- Difficulty with Egocentric View: 90% of subjects found it "easy" to select a spot, but this was in the context of the simulated environment. However, interviews revealed that precise placement was challenging due to the egocentric view and technical difficulties in controlling the robot.
- Desire for Autonomy: A significant majority of subjects (19 out of 22, or 86.4%) expressed a preference for an autonomous navigation feature to automatically place the robot within groups.
Limitations and Future Work
Limitations: * Simulation Environment: The study was conducted in a simulated environment, which may not perfectly replicate the complexities and nuances of real-world interactions. * Virtual Agents: The VAs had uniform appearances and lacked facial expressions or physical behaviors, which could have influenced subject behavior. * Robot FoV: The specific Field of View (FoV) of the robot's camera might have impacted navigation patterns. * Subject Demographics: The study had a higher proportion of male participants and a limited age range, with underrepresentation of older adults and teenagers.
Future Work: * Develop and evaluate autonomous features for mobile robots to join social interactions while adhering to F-formations. * Conduct experiments with a more balanced demographic representation (gender, ethnicity) to explore potential differences in social navigation. * Investigate how the robot's integration into an F-formation modifies the formation itself, potentially leading to new social robot designs. * Explore the impact of different robot FoVs and sensory inputs on teleoperator performance and F-formation adherence.
Summary updated