SMILE Program · Project

EpiSimulator 🦠

Move the sliders strategically, slow down the spread, and stop the outbreak.

EpiSimulator is an interactive epidemic-spread environment built with Unity and embedded in StudyBuds. Learners manipulate transmission and mobility, quarantine, and vaccination conditions, observe their population-level effects, and explain the emerging patterns with either an inquiry-driven peer agent or a misconception-driven teachable agent.

About the SMILE Program Back to Research

Research Questions

How does interacting with a misconception-driven teachable agent, compared with an inquiry-driven peer agent, influence students’ learning outcomes?
How does the pedagogical design of the GenAI agent influence students’ behavioral engagement with the simulation-based learning environment?
How do students’ conversational and conceptual-reasoning processes differ when interacting with a misconception-driven teachable agent compared with an inquiry-driven peer agent?

Inside the Environment

The experience integrates generative activities such as predicting and explaining with simulation-based experimentation, a progress tracker to monitor learning, and interaction with two types of LLM-based agents (i.e., Inquiry-driven peer agent vs. Misconception-driven teachable agent).

Scroll sideways for more →

five selectable studybuds characters named Rafa, Nico, Shelly, Po, and Hammy, each with different appearance

🎨 Personalize the AI Agent

Learners choose from five character appearances and six signature colors before beginning the learning sequence.

EpiSimulator town view with susceptible, exposed, infected, recovered and dead counters updating as the tick count rises

🦠 Watch an Outbreak Unfold

Colour-coded agents move through the town while susceptible, exposed, infected, recovered and dead counts update tick by tick.

EpiSimulator running beside a Studybuds conversation with teachable agent Hammy and a milestone progress tracker

🔄 Experiment and Explain

The simulation, conversation, and progress tracker remain visible together so learners can test an idea and immediately explain what changed.

💬 Inquiry-driven Peer Agent

The agent asks the questions, holding back progress until the learner shares predictions and observations.

Teachable agent Rafa presenting a flawed account of epidemic growth for the learner to diagnose and correct

🎓 Misconception-driven Teachable Agent

The agent opens with a wrong statement that infections rise in a straight, predictable line, and the learner has to identify the misconception and correct it.

Tiered help panel offering a second hint that points the learner back to the EpiSimulator resource

🛟 Tiered Help on Demand

A Get Help button offers hints that redirect learners to the simulator rather than handing over the answer.

Assignment progress chart plotting 27 messages against six milestones as a rising step function

📈 Milestones as Data

Every message is mapped to a milestone, turning a single conversation into a visible learning trajectory.

Scientific Outputs

Master’s Thesis (in progress). Gupta. K. (2026). The Effects of LLM-Based Virtual Agent Design on Learning in an Interactive Epidemic Spread Simulation. Saarland University, Faculty of Mathematics and Computer Science. Supervisor: Dr. Man Su. Reviewers: Prof. Tomohiro Nagashima and Prof. Adish Singla.

Project Team

Developer

Kanika Gupta

Researchers

Dr. Man Echo Su (PI, Project Lead)

Kanika Gupta (Research Assistant)

Prof. Tomohiro Nagashima (Advisor)

Context

Tested with 50 Prolific online participants, supported by the Leibniz-Institut für Wissensmedien.