Dynamic-Classroom-Control-with-RL

Problem

Students in classrooms often face difficulty focusing during lectures or other class activities, preventing them from learning efficiently. Thus, it is the primary focus of teachers to make the classroom a comfortable environment for learning to happen.

How it works

Classrooms have various environment variables that can be modified, such as temperature, air flow, and lighting. We propose a RL model that takes in some summary statistics of students in the classroom, and adjusts the environment to maximize student's productivity.

The RL agent takes in three "statistics": an attention/engagement score, a thermal comfort score, and a facial emotion score. These scores are calculated for each student, and then aggregated for each cluster of students (using Bayesian clustering).

Installation and Running

1. Clone the repository: git clone https://github.com/Saad-Mufti/Dynamic-Classroom-Control-with-RL.git
2. Install necessary dependencies: pip install -r requirements.txt. Note that tensorflow is only required for running fer_model_custom.py. Otherwise, it is fine to use the fer package by Justin Shenk. Also note that pytorch is installed to use CUDA support.
3. Install OpenFace from source as described here.
4. Run the room_agent_notebook.ipynb notebook to train the RL model.