Scholarlink

As robots become increasingly integrated into daily life, equipping them with emotionally intelligent behaviors is essential for improving Human-Robot Interaction (HRI). This study investigates differences in synthetic emotion generation between EmoACT–a psychology-based, platform-independent framework grounded in Affect Control Theory (ACT)–and a Large Language Model (LLM), specifically GPT-4, considering the influence of synthetic personality traits. We focused on the traits of Agreeableness, Extraversion, and Conscientiousness, comparing both systems using a dataset composed of user emotions, sentences, and robot comfortability levels. Our results show that EmoACT produces emotional responses influenced by all input variables, with personality traits having a significant impact on the generation process. In contrast, GPT-4 predominantly mimics user emotions, with only minor variations based on personality. These findings highlight the differing mechanisms and strengths of psychology-based and LLM-based emotion generation frameworks, offering insights for designing emotionally capable artificial agents.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Simulating Feelings: LLM vs. Psychology-Based Models in Human-Robot Interaction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Francesca Corrao,
  • Alice Nardelli,
  • Antonio Sgorbissa,
  • Carmine Tommaso Recchiuto

摘要

As robots become increasingly integrated into daily life, equipping them with emotionally intelligent behaviors is essential for improving Human-Robot Interaction (HRI). This study investigates differences in synthetic emotion generation between EmoACT–a psychology-based, platform-independent framework grounded in Affect Control Theory (ACT)–and a Large Language Model (LLM), specifically GPT-4, considering the influence of synthetic personality traits. We focused on the traits of Agreeableness, Extraversion, and Conscientiousness, comparing both systems using a dataset composed of user emotions, sentences, and robot comfortability levels. Our results show that EmoACT produces emotional responses influenced by all input variables, with personality traits having a significant impact on the generation process. In contrast, GPT-4 predominantly mimics user emotions, with only minor variations based on personality. These findings highlight the differing mechanisms and strengths of psychology-based and LLM-based emotion generation frameworks, offering insights for designing emotionally capable artificial agents.