To Örebro University

Social group interactions are a fundamental aspect of human communication and collaboration, characterized by dynamic spatial and orientational patterns. As robots become more prevalent in human environments, they are expected to adhere to social norms, including the ability to join these interactions seamlessly without causing disruptions.

Motivated by the need to explore how these norms extend to robots, we investigated the behavior of teleoperators during group interactions. The findings not only demonstrated that teleoperators inherently follow these socia lnorms but also highlighted their preference for robots with autonomous capabilities to seamlessly join group interactions adhering to these norms. In this regard, this thesis presents a new and comprehensive framework named as “Autonomous Group Interactions for Robots (AGIR)”, designed to enable mobile robots to join ongoing social group interactions autonomously through an egocentric camera perspective.

The AGIR framework is built upon principles from social psychology, such as Proxemics and F-formations, to ensure socially acceptable behavior. Its architecture comprises computational models for extracting spatial and orientational information, detecting groups, estimating spatial formations, and identifying optimal positions for robot in group interactions. Designed to operate in real-time using the robot’s onboard sensors, the framework is modular and adaptable to a diverse range of robotic platforms.

AGIR was rigorously evaluated through experiments conducted in both simulated and real-world environments. Real-world experiments were performed in corridor, lab, and home environments. While in simulation, three scenes were developed similar to conference lobby and coffee break scenarios. Results demonstrated high accuracy in spatial and orientational estimations, group detection, F-formation predictions, and determining optimal robot positions within groups. The framework effectively enabled operating in real-time from an egocentric view and autonomously joining group interactions without disruption. AGIR lays the groundwork for robots to seamlessly integrate into human social environments, enabling practical applications in domains such as elder care, telepresence, and collaborative workspaces.