Earthquaker-AI brings RAG-guided assistance to earthquake education for primary schools

Introduction

Earthquake preparedness is difficult to teach as a purely verbal lesson. Children need to recognize safe actions, understand their sequence, and remain calm enough to apply them under stress. A new arXiv paper presents Earthquaker-AI, a hybrid educational framework designed for primary-school earthquake education. Rather than simply adding a chatbot to the classroom, it combines Lego WeDo2 robotics, retrieval-augmented generation, and rubric-based feedback into one learning flow.

Key points

• From physical simulation to reflective learning: Earthquaker-AI builds on the earlier award-winning Earthquaker STEM project. The original robotics activity used Lego WeDo2 automation to simulate seismic response, giving students hands-on access to sensors and actuators as tangible representations of protective actions. The AI-enhanced version adds a conversational layer so students can explain, question, and reflect on what they are doing.

• RAG keeps responses tied to safety guidance: The dialogic module uses retrieval-augmented generation to semantically match student queries with official earthquake safety guidelines before producing answers. This matters because emergency education leaves little room for speculative or inconsistent advice. The paper reports that experimental evaluation showed strong groundedness and accuracy, along with a low hallucination rate.

• Learning tasks change by grade level: The framework follows a progressive trajectory aligned with cognitive development. In early grades, students focus on recognizing basic safety actions through multiple-choice questions assessed with a two-dimensional rubric. In middle grades, they identify correct action sequences and are evaluated with a three-axis rubric. In upper grades, the task shifts toward short written responses, assessed with a four-dimensional rubric that includes clarity of expression.

• Feedback goes beyond right or wrong: The assistant acts as a guided learning mechanism. It provides rubric-based verbal feedback that helps students align their answers with safety guidelines while supporting self-regulated learning. In crisis education, the ability to explain an action and stay composed can be as important as selecting the correct option.

Why it matters

Earthquaker-AI is notable because it connects three educational approaches that are often used separately. Robotics creates an embodied scenario, RAG supplies a source-grounded information layer, and rubrics make feedback age-appropriate and structured. For early crisis-management education, that combination can help students move from memorizing instructions to understanding and practicing them.

At the same time, the work should be read as a framework and evaluation study rather than proof of universal classroom effectiveness. Broader deployment would require longer-term classroom trials, teacher workflow analysis, and careful governance around children’s data and the boundaries of AI-generated safety advice.

The broader lesson is pragmatic: AI in education may be most useful when it is constrained by trusted sources, tied to concrete learning activities, and evaluated through transparent rubrics rather than left as an open-ended tutor.

Source: arXiv