Do you want a RL agent moving on Unreal Engine?

paper.

Unreal Reinforcement

This is a DQN Algorithm using Python.
and game that is a completely out of control using Unreal Engine.

Two independent processes connect to socket communication.
Train neural network and playing game asynchronously.

Please feel free to open an issue or a pull-request if you have any idea to make it better. :)

Prerequisites

This repository is tested ...

- Windows 10
- Visual Studio 2019

- Unreal Engine 4.25
- UE4 tcp socket plugin (https://unrealengine.com/marketplace/en-US/product/tcp-socket-plugin)

- Python 3.6
- Pytorch 1.7.1

We plan to shipment the game

How to use

1. Open URGame/URGame.uproject
2. Run Python Train.py
3. Start URGame

Demo youtube

Reward

stack state = 1

stack state = 4

Version

C++(Recommend)

Blueprint