This repository is for the "Collaboration and Competition" project for Udacity Deep Reinforcement Learning Nanodegree.
In this project the environment is a simplified version of Tennis from Unity ml-agent project. This is an episodic task in which two agents control the rackets to bounce a ball. The rewards are granted as following:
- +0.1 when hitting the ball over the net, and additional -0.01 if the ball goes out of bounds
- -0.01 if missing the ball (the ball hits the ground)
The final reward of an episode is the maximum of two agents. In this project, we aim to get >0.5 average reward of 100 consecutive episodes.
The basic dependencies are listed in this document from Udacity Nanodegree. To run this solution, you will need to install some additional dependencies:
conda install tensorboard protobuf pip install torchsummary tensorboardX
To run the script, simply execute python test_agents.py
. This script will
start training two agents with multi-agent DDPG algorithm with pre-defined
model structure and hyper-parameters. The result and analysis is in
Report.ipynb.