Designing Empathetic Companions: Exploring Personality, Emotion, and Trust in Social Robots

TL;DR

This paper investigates how robotic personality shapes perceived empathy, trust, and enjoyment in human-robot interaction by proposing a cognitive architecture built around a three-dimensional personality space (Conscientiousness, Extraversion, Agreeableness, CEA) and an emotion-aware generator. It integrates perception, memory-driven prospection, and GPT-4o-based sentence generation to produce personality-dependent emotions and behavior in the Navel humanoid, validated through 84 dyadic conversations and a comprehensive multimodal analysis. Key findings show that Agreeableness robustly enhances Experience, Empathy, Trust, and Enjoyability, while Extraversion boosts Enjoyability and Extraversion-related expressions; Conscientiousness influences Capability trust. The results inform practical guidelines for designing empathetic, trustworthy companions by illustrating how combining multiple personality traits can improve social robot effectiveness in real-world interactions.

Abstract

How should a companion robot behave? In this research, we present a cognitive architecture based on a tailored personality model to investigate the impact of robotic personalities on the perception of companion robots. Drawing from existing literature, we identified empathy, trust, and enjoyability as key factors in building companionship with social robots. Based on these insights, we implemented a personality-dependent, emotion-aware generator, recognizing the crucial role of robot emotions in shaping these elements. We then conducted a user study involving 84 dyadic conversation sessions with the emotional robot Navel, which exhibited different personalities. Results were derived from a multimodal analysis, including questionnaires, open-ended responses, and behavioral observations. This approach allowed us to validate the developed emotion generator and explore the relationship between the personality traits of Agreeableness, Extraversion, Conscientiousness, and Empathy. Furthermore, we drew robust conclusions on how these traits influence relational trust, capability trust, enjoyability, and sociability.

Figures (6)

• Figure 1: Personality-based cognitive architecture. Three different modules can be identified: perception (green), reasoning (blue), and actions (orange).
• Figure 2: The figure illustrates personality-dependent functions (in purple and blue) and their responses to various actions, showing possible plans developed for an extrovert and unscrupulous agent, the selection of actions, their evaluation, and updates to the Episodic Memory. The contrast between the green dashed lines (indicating expected outcomes) and the solid lines (representing actual outcomes) highlights the gap between the anticipated reward and the obtained one.
• Figure 3: Evaluation of the Robot Emotion Generator, for an agreeable robot in response to different user's emotions. The three subplots distinguish the different robot's conditions (with and without emotional intelligence (EI), comfortable or uncomfortable).
• Figure 4: Experimental set-up
• Figure 5: Perception of the different traits of the CEA taxonomy during the interaction with the robot Navel. E, A, C are the three personality traits. L and H mean respectively high value and low value of the dimension of interest. (e.g. LE means introversion and HE means extraversion). The hues indicate how the specific pole of interest is combined with the poles of the remaining two traits in the taxonomy.