Collaboration and competition

This is the last project from Udacity's deep learning nano degree where we have a multi-agent environment in which two agents have to collaborate.

Details

This is a tennis environment where two players (agents) have to keep the ball in play. To meet that goal the rewards are designed this way:

If an agent hits the ball over the net, it gets +0.1 points.
If it misses (the ball bounces) or it throws the ball out of the field, it receives a reward of -0.01 points
The "game reward" is computing by taking the maximum of each agent's reward

The task is considered solved if the average score over a period of 100 games is greater than +0.5.

First create a python environment with python 3.6 and install the requirements from the requirements file. If using conda:

conda install -n myenv python=3.6
pip install -r requirements.txt

You also need jupyter notebook, install it if not already in your system. And make sure you add the environment you created to it.

conda install ipykernel

Finally, make sure you download the tennis environment (not part of this repo). Depending on yor platform, you can get it for:

See the demo in Report.ipynb

Enjoy!