Reinforcement Learning with Unity and Pytorch

This repository contains several intuitive, high quality implementations of state-of-the-art reinforcement learning algorithms solving Unity-based environments.

collecting_bananas
This directory contains everything you need to train and evaluate an Agent that learns to collect bananas in a large environment.

robot_arm
This directory contains everything you need to train and evaluate an Agent that learns to control a robotic arm.

playing_tennis
This directory contains everything you need to train and evaluate two Agents that learn to play tennis in a Multi-Agent-Scenario.

Each subproject is designed as a stand-alone project and can be used without the others.
All environments were built in Unity. For more information on any of the subproject, please refer to the README.md inside the subproject's directory.

Getting started

To set up the necessary dependencies, follow the steps described here.

( Disclaimer : This steps assume that you are using Anaconda. If you don't, I highly recommend you to do so.
You also might want to use Jupyter Notebooks for a more visual experience during training / evaluation.)

Create and activate a new environment:

Linux or Mac:
conda create --name rlup python=3.6
source activate rlup
Windows:
conda create --name rlup python=3.6 
activate rlup
Clone the repository (if you haven't already!), and cd into the setup folder. Then, install several dependencies.
git clone https://github.com/HFBN/Reinforcement-Learning-with-Unity-and-PyTorch.git
cd deep-reinforcement-learning/python
pip install .
There have been some issues with Windows and pytorch as well as TensorFlow. Therefore, I recommend to install them
manually.
conda install -c conda-forge tensorflow==1.14.0
conda install -c pytorch pytorch==1.4.0
Create an IPython kernel for the rlup environment.
python -m ipykernel install --user --name rlup --display-name "rlup"
Before running code in a notebook, change the kernel to match the rlup environment by using the drop-down Kernel menu.

Author
Jonas J. Mühlbauer, Artifical Intelligence Consultant
        Intuitive, high-quality Implementations of several State-of-the-Art Reinforcement Learning Algorithms solving various Unity-based Environments.
    Topics
  reinforcement-learning
  deep-learning
  unity
  deep-reinforcement-learning
  q-learning
  pytorch
  deep-q-network
  deep-q-learning
  multi-agent-reinforcement-learning

Reinforcement Learning with Unity and Pytorch

Table of Contents

Getting started

Author

Topics