Pong AI

GIF
Working modular Pong

Inspiration

What it does

Teaches an AI to play Atari Pong using reinforced learning

How I built it

Read through a bunch of reinforced learning articles

Challenges I ran into

Making the pong game modular Interfacing between the two languages Setting up the server for machine learning Unable to fully get ZeroMQ message passing to work

Accomplishments that I'm proud of

Modular pong

What I learned

Interfacing between C++ and Python using ZeroMQ SFML as a graphics library Tensorflow Keras

What's next for Pong AI

Make the network learn faster

Built With

c++
cmake
keras
numpy
python
sfml
tensorflow

Submitted to

HackED Beta 2018

Created by

Worked on the communication layer on the C++ side over 0MQ to the Python Keras model runner.
Wrote most of the code on the game's non-logic parts, wrote most of the engine (inputs/outputs/game objects).

This was the most C++ I've ever written for a project, first time doing any sort of cross language IPC without a big framework backing me, and also first time working with SFML.

Michael Kwok
I worked on finding an RL model to train for the pong AI and the actual training of it on Google computer engine. I have never done anything to do with machine learning before so this was a great experience

Caleb Schoepp
Armianto Sumitro

Updates

Armianto Sumitro started this project — Nov 18, 2018 11:35 AM EST

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.