Abstract:

Large language models (LLMs) have changed how novice programmers, including students and junior software engineers, seek help with programming. However, effective learning depends on structured guidance and collaboration, such as the software engineering practice of pair programming. While LLM support is typically accessed via individual chat interfaces, often disrupting shared workflows, this paper explores an interactive approach in which a humanoid robot mediates interaction with an LLM to support collaborative learning and encourage good programming practices.

We present a tool concept where a robot facilitates LLM-driven feedback during pair programming. The robot, in the role of a navigator, gives feedback to the driver responsible for writing the code based on code snapshots. The feedback is delivered via LLM-generated responses optimized for efficient human-robot interaction. The paper details prompt design for educational scaffolding, interaction patterns for maintaining shared attention, and a technical architecture linking the coding environment, robot client, and LLM service. It also discusses practical deployment challenges, including response formatting for text-to-speech and rate limiting.

The paper contributes design guidance and implementation lessons for integrating humanoid robots and LLMs in programming education and highlights the potential of physical embodiment to support collaborative workflows throughout the learning process.

Keywords:

Large language models (LLMs), pair programming, humanoid robots, human-robot interaction, programming education.