AIRLab-POLIMI/mushroom_hierarchical

Code for the paper: Graph-Based Design of Hierarchical Reinforcement Learning Agents

Mushroom Hierarchical

Mushroom Hierarchical: Hierarchical Reinforcement Learning python library.

This repository implements the hierarchical framework proposed in the paper: "A Control Theoretic Approach to Hierarchical Reinforcement Learning"

What is Mushroom Hierarchical

Mushroom Hierarchical is a python Hierarchical Reinforcement Learning (HRL) library based on Mushroom, the RL library of Politecnico di Milano. It allows to perform HRL experiments exploiting the control graph formalism.

Installation

You can do a minimal installation of Mushroom with:

git clone https://github.com/AIRLab-POLIMI/mushroom.git
cd mushroom
pip3 install -e .

You can install Mushroom Hierarchical with:

git clone https://github.com/AIRLab-POLIMI/mushroom_hierarchical.git
cd mushroom_hierarchical
pip3 install -e .

How to set and run and experiment

To run experiments, you should use the run.py script in the experiment folder. If you want to visualize the results produced by our runs, you can look at the out folder, or you can re-run the plot_*.py scripts into the graphs folder.