From simulation to real-world: Measuring social engagement for social robots

Webb, Nicola

Abstract

Understanding social interactions is crucial for the development of artificial agents, particularly social robots. This thesis specifically focuses on the challenge of interpreting social signals during interactions, with a primary emphasis on social engagement. Enhancing social cognition in artificial agents requires deciphering social cues and considering the dynamic nature of social interactions. Drawing upon existing literature, the study emphasises the importance of modelling individual social behaviours while prioritising the assessment of engagement levels during interactions.

The motivation for this research arises from the need to bridge the gap between human and robot understanding of social interactions. A case study involving a group interaction scenario illustrates the complexity of social dynamics and the challenges faced in accurately interpreting them. Despite humans' innate ability to discern social signals, current algorithms and methodologies struggle to replicate this capability in artificial agents. Therefore, this research aims to develop algorithms that can detect and interpret human behaviours accurately, ultimately enhancing social cognition in robots, with a particular focus on understanding and assessing social engagement.

Data collection plays a crucial role in this thesis, with a focus on capturing rich, multi-modal interactions. Two social interaction datasets are collected in this body of work. By examining various social signals, including body language, gaze, and proximity, the study aims to model individual behaviours and assess engagement levels during interactions. Furthermore, a novel visual social engagement metric is created from the datasets using proximity and orientation data. This metric is then leveraged to generate interaction profiles for each participant.

The research questions addressed in this study revolve around understanding the dynamics and underlying mechanisms of social interactions, representing interactions over time, and applying findings to real-world scenarios. By exploring these questions, the study aims to contribute to the betterment of social robotics and improve robots' ability to engage effectively and in a more socially intelligent way.

• Thesis Type: Thesis
• Deposit Date: Apr 1, 2024
• Publicly Available Date: May 7, 2025
• Public URL: https://uwe-repository.worktribe.com/output/11874309
• Award Date: May 7, 2025