Enabling Robots to Adhere to Social Norms by Detecting F-Formations

Abstract

Robot navigation in environments shared with humans should take into account social structures and interactions. The identification of social groups has been a challenge for robotics as it encompasses a number of disciplines. We propose a hierarchical clustering method for grouping individuals into free standing conversational groups (FSCS), utilising their position and orientation. The proposed method is evaluated on the SALSA dataset with achieved F1 score of 0.94. The algorithm is also evaluated for scalability and implemented on a mobile robot attempting to detect social groups and engage in interaction.

The SMOOTH-Robot: A Modular, Interactive Service Robot

“…On our robot, based on the vision modules described in Section 4.1, an agglomerative hierarchical clustering method is used to detect social groups using the persons' positions and orientations, thus considering the social structures mentioned before (Kollakidou et al, 2021). The individuals are clustered with a criterion of a maximum distance allowed in case of intended interaction.…”