MiniHuman: Simulating, Augmenting, and Regulating Virtual Human Behaviors and Responses to Environmental Stimuli

Introduction

MiniHuman aims to leverage SOTA AI models (LLM agents, RL agents, etc), sensing, and intervention techniques to simulate both human mind (mental behaviors) and physical behaviors, to support AI and HCI applications in education, health, recommendation, user experience, and human augmentation.

News

More models, behaviors, environments, and applications will be released soon!

Applications

Education

• LLM-based Student Simulator (e.g., Classroom Simulacra [CHI 2025], EduAgent, EduTwin)
• Student Behavior Regulation (e.g., PeerEdu [CHI 2025])

Mental Health

• Attention Regulation (e.g., Eyerofeedback)

Human Cognitive Augmentation

• Deep Reinforcement Learning Based Cognition Model (e.g., ReactiveAgent)
• Cognitive Training (e.g., TimeCare [CHI 2023])

Climate Change

• Human action in response to climate change

Scientific Discovery

Social Science

• Social Interaction
• Social Network

Technical Details

Human Behaviors: Mind + Physical Behaviors

We support a wide range of human behaviors to empower diverse application scenarios.

• Physiological behaviors
  • Gaze
• Motor behaviors
• Cognitive states
  • Attention
  • Workload
  • ...
• Knowledge states
• Emotion

Environmental Stimuli

We support a wide range of environmental stimuli to empower diverse application scenarios.

• Visual stimuli
  • Time Pressure
  • ...
• Auditory stimuli
  • Instructor Voice
  • ...
• Textual stimuli
  • Course Materials
  • ...

Agent Architecture

• LLM-based Agents
  • Persona Initialization
  • Task Definition
  • Behavior Definition
  • Environment Description
  • Foundation Model
• RL-based Agents
  • Action Space
  • Observation Space
  • Optimization Policy

Citation

If you use this repository, please cite it as follows:

Songlin Xu, Xinyu Zhang. (2024). MiniHuman-Toolkit (Version 0.1.2). GitHub. https://github.com/songlinxu/MiniHuman-Toolkit

Alternatively, use the following BibTeX entry:

@misc{songlin2024minihuman,
  author = {Songlin Xu, Xinyu Zhang},
  title = {minihuman},
  year = {2024},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/songlinxu/MiniHuman-Toolkit}},
  version = {0.1.2}
}