bwosh / DRL_CollaborationAndCompetition

Udaticy's project solution from Deep Reinforcement Learning nanodegree program.

DRL_CollaborationAndCompetition

This project is a part of:
Deep Reinforcement Learning Nanodegree

The project uses DDPG algorithm to solve 'Tennis' environment.

Environment details

Goal is to move play tennis using two rockets. If the ball hits the grund or goes out-of-bounds the reward is -0.01. If the rocket hits the ball the reward is 0.1. The goal is to write cooperative agent that will reach score of 0.5 within 100 consecutive episodes by bouncing the ball as long as possible.

• There are 2 rockets
• Each arm rocket contains 24 local observations
• Action space contains 8 contunuous values in range from -1 to 1.

Requirements

Below you can find a list of requirements required to run train.py script:

Resources

• python 3.6
• Tennis app (this is delivered by Udacity Team) - I was using headless linux client to be able to run it seamlesly on any environment (even without display)

Python packages

• torch
• numpy
• tqdm
• matplotlib
• unityagents

Usage

Training

python3 train.py

Details

Implementation uses DDPG algorithm - the details and metrics can be found in report file.