Description

Gym Environment for Deep Reinforcement Learning is a plugin which allows to easily create a “Reinforcement Learning” environment with Unreal.

It provides features to set up the observation space, the action space, the reward, etc. in an Unreal level and exposes them on the network through an API.

This API can then be called from any RL framework (TensorFlow, PyTorch, Stable Baselines, etc.) to easily train a neural network to play your game.

Note: Some Python files are provided as an example for the Reinforcement Learning part. To make the example work, you should install OpenAI Gym (gym) and Stable Baselines (stable-baselines3[extra]) for your Python environment.

Technical Details

Code Modules:

Number of Blueprints: 8

Number of C++ Classes: 12

Network Replicated: No

Supported Development Platforms: Win64

Supported Target Build Platforms: Win64

Documentation: https://docs.isaratech.com/ue4-plugins/drl-unreal

Example Project: (incoming later)

Important/Additional Notes: The plugin is composed of two parts: the Unreal part and the Python part, both are provided but a Python runtime environment is required to run the Python side (it’s also possible to replace Python by something else but the appropriate code for it is not provided)

Supported Engine Versions

4.27, 5.0 – 5.3