Design and Evaluation of a Novel Framework Inspired by Affect Control Theory to Create Social Robot Affective Identities

Abstract

Developing agent personalities and behaviours that can be adjusted based on individual factors such as personal preferences, context, and lived experiences is important to improve user acceptance and engagement with social robots in varying scenarios, such as for assistive robots. Current approaches used in the design of social robot personalities (e.g., extroversion manipulation-based) often do not encompass the entire personality (e.g., Big 5) as only a subset of dimensions can be manipulated. We propose a novel framework using Affect Control Theory (ACT) that enables the manipulation of all aspects of an agents' affective identities, as a proxy for personality. We evaluate this framework in two studies, one online and one in-person, involving an "at-home" medication sorting task to determine (1) whether participants are able to identify the manipulations in the social robot behaviours, and (2) whether this framework can be used for personalizing social robot identities, in other words, if participants will prefer to interact with specific robot identities considering their own affective identities. The results confirmed the design and manipulation of two out of three ACT dimensions and showed that participants preferred interacting with robots that are perceived to have an affective identity closer to them. Proposed framework and results can have significant contributions to the personalization of social robot behaviours in assistive robots and beyond.