Developing Personalized Multi‐Role Educational Robot Learning Assistants Based on Large Language Models

Human–robot interaction (HRI) learning plays a crucial role in enhancing the interactivity and enjoyment of the learning process and cultivating future talents with innovative thinking abilities. However, personalized multi‐role HRIs in educational contexts remain insufficiently explored. To fill this gap, this study developed a personalized multi‐role educational robot learning assistant (MREduRobot), based on the NAO robot and powered by the ERNIE‐4.0‐Turbo‐128K large language model (LLM). The assistant provides personalized video explanations and interactive Q&A and integrates NAO‐based voice synthesis to simulate customized teacher roles with variations in gender, personality traits, and teaching style. In addition, a textual semantics matching method was employed to generate context‐appropriate actions for the NAO robot in situated learning scenarios. Experimental results showed that the system achieved an average response time of 20.70 s during Q&A interactions (SD = 4.60 s). The questionnaire indicated that learners exhibited a preference for teacher roles centered on logic and innovation. Furthermore, the MREduRobot group significantly outperformed the video group in enhancing the learning experience (learning motivation: Sig(2‐tailed) < 0.001; teacher‐student interaction: Sig(2‐tailed) < 0.001), reducing cognitive load (Sig(2‐tailed) < 0.001), and improving learning effectiveness (retention test: Sig(2‐tailed) < 0.001; transfer test: Sig(2‐tailed) < 0.001). These findings demonstrate the feasibility of using MREduRobot to promote contextual learning through effective HRI.