Skip to content

My attempt to run DQN on Jetson TX1, which learns how to play Nintendo Famicom Mini games through reinforcement learning directly.

Notifications You must be signed in to change notification settings

jkjung-avt/dqn-tx1-for-nintendo

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

48 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

dqn-tx1-for-nintendo

Aka, a Deep Q Learner Network (DQN) on NVIDIA Jetson TX1 which learns to play Nintendo Famicom Mini games.

I'm developing an AI program, based on DeepMind's DQN, on Jetson TX1 to play Nintendo games. I use this repository to keep track of all source code of the project. Please refer to the following blog posts for more information.

DQN (RL Agent) on TX1 for Nintendo Famicom Mini

Nintendo AI Agent Training in Action, Finally...

Training an AI agent to play 'Galaga' on Nintendo Famicom Mini

Installation Instructions

The following are required for this dqn-tx1-for-nintendo to work.

Training DQN on Nintendo Games

To build the code on Jetson TX1, just run make at the project root diretory.

 $ make

Then execute train-deepmind.lua to train the DQN (use -h to see help messages).

 $ th ./train-deepmind.lua

Modules within This Project

The following modules resides in the corresponding subdirectories of the repository. There are also test scripts for most modules as described in the next section.

Testing Individual Modules

All test scripts are organized in the 'test' subdirectory. They are meant to be run from the project root directory. Most of the test scripts accept cmdline options (use -h to see help messages).

 $ qlua test/test_vidcap.lua
 $ qlua test/test_galaga.lua
 $ th   test/test_gpio.lua
 $ th   test/test_imshow.lua
 $ th   test/test_gameenv.lua

About

My attempt to run DQN on Jetson TX1, which learns how to play Nintendo Famicom Mini games through reinforcement learning directly.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published