Skip to content

Udacity Deep Reinforcement Learning - Project 1: 2D Navigation using Discrete Actions

License

Notifications You must be signed in to change notification settings

mattuntergassmair/udacity-deep_rl-navigation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Project 1: Navigation

Udacity Deep Reinforcement Learning Nanodegree

Trained Agent

Project Description

This project is part of the Udacity Deep Reinforcement Learning Nanodegree program. The agent can apply four discrete actions to move through a 2D square plane: move forward, move backward, turn left, turn right. The agent can accumulate rewards (+1) by collecting yellow bananas but will be penalized (-1) for collecting blue bananas.

Each frame is observed in the form of a 37-dimensional state vector encoding the agent's velocity and ray-based perception information.

The agent learns from experience through repeated interaction with the Unity simulation environment.

Results

The environment is considered solved when the agent accumulates an average reward of +13 per episode.

Using a Deep Q-Network with hidden layer sizes [74,37,16,8] to approximate the action-value function, the episode can be solved in about 300 episodes.

Reward evolution

Installation

TODO: virtualenv installation

Download Unity simulator for Linux here or directly through the command line

wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P1/Banana/Banana_Linux.zip  # with visualization
wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P1/Banana/Banana_Linux_NoVis.zip  # no visualization

and unzip them in the root directory of this repository (simulator files for MacOS and Windows ).

Running Tests

Run the complete test suite with the command

python -m unittest qlearning tests

About

Udacity Deep Reinforcement Learning - Project 1: 2D Navigation using Discrete Actions

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published