Deep Reinforcement Learning-based Active Queue Management

In this repository, I invesitgate the implementation of a DRL-based agent for dropping the packets in a queuing model. I use an stable version of OpenAI baselines for this purpose. The queuing simulation is implemented using MATLAB in another repository.

Setup

1. Install stable-baselines3

Development Guide

I started by creating a custom environment that follows OpenAI gym interface based on this notebook and this tutorial from OpenAI baselines. The result is a shared memory communication module queue_env.py that connects to MATLAB and exchanges the information.

In run.py, a PPO RL agent is instantiated using the custom environment and MLP policy.