A Teacher–Agent–Robot Collaborative Model for Bilingual Hands-on STEAM Learning: Integrating an LLM-based Pedagogical Agent with Educational Robots

Abstract In recent years, Large Language Models (LLMs) have shown potential in providing dialog-based guidance and learning prompts in educational contexts. However, systematic empirical research remains limited on approaches that position educational robots as the interaction medium and LLMs as the instructional support engine to orchestrate classroom guidance using prompts and instructional workflow scripts. In response, this study adopted a bilingual STEAM (Science, Technology, Engineering, the Arts, and Mathematics) summer camp for lower elementary students as the research context, and developed and implemented an R&T (Robots & Tangible Objects) learning system that integrated educational robots with an LLM-based pedagogical agent to foster embodied interaction and bilingual learning. The system used predesigned prompts and procedural rules to organize instructional steps, question prompts, operational checks, and feedback statements, enabling the students to receive consistent and repeatable support at key task-progression checkpoints. A total of 34 elementary students participated in a five-day program (120 minutes per day). Data sources included STEAM knowledge tests, English vocabulary assessments, a curiosity scale, hands-on performance ratings, and system interaction logs. Wilcoxon signed-rank tests indicated significant improvements in the students’ STEAM knowledge, curiosity, and English vocabulary, while group-level analyses showed significant differences in hands-on performance. Process log analyses further suggested that the prompt-guided workflow scripts helped to maintain consistency in classroom guidance and supported task completion through an action–feedback–revision cycle. Nevertheless, classroom implementation was constrained by factors such as token limits, speech-recognition latency, and network/API stability, which indicated the need to incorporate exception handling and timely teacher support mechanisms to ensure instructional continuity. Overall, this study provided initial empirical support for a feasible design framework for integrating educational robots × an LLM-based pedagogical agent into bilingual hands-on STEAM activities, and its findings offer practical recommendations for curriculum adoption and system deployment.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11199056/v1
Primary Topic
Social Robot Interaction and HRI
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A Teacher–Agent–Robot Collaborative Model for Bilingual Hands-on STEAM Learning: Integrating an LLM-based Pedagogical Agent with Educational Robots

陳俊臣 Hsien-Sheng Hsiao, Nian-Shing Chen, Yi-Ying Lin, Guang-Han Chung
1 links
Research Square
Social Robot Interaction and HRI
preprint

A Teacher–Agent–Robot Collaborative Model for Bilingual Hands-on STEAM Learning: Integrating an LLM-based Pedagogical Agent with Educational Robots

陳俊臣 Hsien-Sheng Hsiao, Nian-Shing Chen, Yi-Ying Lin, Guang-Han Chung
preprint en

Abstract

Abstract In recent years, Large Language Models (LLMs) have shown potential in providing dialog-based guidance and learning prompts in educational contexts. However, systematic empirical research remains limited on approaches that position educational robots as the interaction medium and LLMs as the instructional support engine to orchestrate classroom guidance using prompts and instructional workflow scripts. In response, this study adopted a bilingual STEAM (Science, Technology, Engineering, the Arts, and Mathematics) summer camp for lower elementary students as the research context, and developed and implemented an R&T (Robots & Tangible Objects) learning system that integrated educational robots with an LLM-based pedagogical agent to foster embodied interaction and bilingual learning. The system used predesigned prompts and procedural rules to organize instructional steps, question prompts, operational checks, and feedback statements, enabling the students to receive consistent and repeatable support at key task-progression checkpoints. A total of 34 elementary students participated in a five-day program (120 minutes per day). Data sources included STEAM knowledge tests, English vocabulary assessments, a curiosity scale, hands-on performance ratings, and system interaction logs. Wilcoxon signed-rank tests indicated significant improvements in the students’ STEAM knowledge, curiosity, and English vocabulary, while group-level analyses showed significant differences in hands-on performance. Process log analyses further suggested that the prompt-guided workflow scripts helped to maintain consistency in classroom guidance and supported task completion through an action–feedback–revision cycle. Nevertheless, classroom implementation was constrained by factors such as token limits, speech-recognition latency, and network/API stability, which indicated the need to incorporate exception handling and timely teacher support mechanisms to ensure instructional continuity. Overall, this study provided initial empirical support for a feasible design framework for integrating educational robots × an LLM-based pedagogical agent into bilingual hands-on STEAM activities, and its findings offer practical recommendations for curriculum adoption and system deployment.

Research Square
National Taiwan Normal University (TW)
National Taiwan Normal University, Institute for Research Excellence in Learning Sciences, National Taiwan Normal University, National Science and Technology Council
Quality education
Social Robot Interaction and HRI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon