Unreal Engine Learns to Balance a Pole | Unreal Engine 4 Reinforcement Learning Project

preview_player
Показать описание
This is my first time using Unreal Engine. I love machine learning but there are no good tools to do machine learning stuff on Unreal. So I developed my own tools. I connected a python and a c++ process through a TCP socket and now the python process can run all the Reinforcement Learning algorithms and send control signals to Unreal Engine and get feedback.

Using this trick I trained a Deep Q Network to do the simple job of balancing a pole using only 2 possible actions. Go left or go right.

The source code for this project is available on my github. Please give it a star if you find it useful and let me know if I should make a tutorial on how to create this project from scratch.

Source Code:
Рекомендации по теме
Комментарии
Автор

Super cool man!

I took the liberty to add you in discord, because I've a couple of questions about the phisycs in UE.
But I'm not coming empty handed! I felt you with the only TCP tutorial you found, and I come with more code about it!

xmariopixels
Автор

Glad to know I'm not the only one fighting this headache, getting ml working in c++ is a nightmare of no support.

xaviermagnus
Автор

This is an amazing work. Anxiously waiting for the tutorial. Also it is a request in your tutorials do more focus on reinforcement learning and its concepts and algorithm.
I guess using them in unreal would be a separate task. Do guide me further if I am wrong

hasanrizvi
Автор

Would love to see a tutorial on this ;)
Great work 👍👍👍

raccoon_
Автор

this is like fat dino but really passive aggressive

AdrenResi
Автор

Can you explain how to implement the code to your own project?

SzymonMaslowski