Pathfinding Using Reinforcement Learning

• Q-table
• Q-network

• PPO
• DQN (with the Double DQN extension) - work in progress, not currently working as intended
• DRQN - work in progress, not currently working as intended

python + pip, version 3.6 or greater

Note: It's recommended that you install the Python dependencies in a virtual environment. virtualenv and virtualenvwrapper:

pip install virtualenv
pip install virtualenvwrapper

Note: To get matplotlib to work on a Mac with a virtual environment you have to use venv instead

mkvirtualenv purl

or with venv:

python -m venv purl-venv

First, switch to the virtual environment:

workon purl

or with venv:

source purl-venv/bin/activate

Note: You can set up an alias in your shell to make the virtual environment more accessible, e.g alias actpurl='source /path/to/purl/purl-venv/bin/activate'

Then, install the dependencies by running:

pip install -r requirements.txt

or if you have pip-sync installed:

pip-sync

There are two main subcommands to PURL

To train a model, run:

./purl train

For example, to train a model using the PPO algorithm on the MiniGrid-LavaCrossingS9N1-v0 environment, use the following arguments:

./purl train --algorithm ppo --environment MiniGrid-LavaCrossingS9N1-v0

To visualize a model, run:

./purl vizualize

Python dependencies are managed by pip-compile

To add a new package, simply add it to the list in requirements.in.

You then update the requirements.txt-file by running

pip-compile --output-file requirements.txt requirements.in

• gym-minigrid - Minimalistic gridworld environment for OpenAI Gym
• PyTorch - Tensors and Dynamic neural networks in Python

• Anne Engström
• Joel Lidin
• Gustav Molander
• Olle Månsson
• Noa Onoszko
• Hugo Ölund

This project is licensed under the MIT License - see the LICENSE file for details