Dungeon Generation
A group project that attempt to research the efficacy of machine learning techniques such as Auto encoders and Generative adversarial networks in game development.
my contribution was towards making most of the Unreal engine front end logic and helping in the development of the GAN.
my contribution was towards making most of the Unreal engine front end logic and helping in the development of the GAN.
An experimental project made to research AI usage in games
AI developed in a Python backend server and the result displayed in Unreal Engine 5
My info
Project Info
tools/engines:
- unreal engine 5.2.1
- Kingston university assignment
- 2024
- 3 months (University assignment)
- 4 people (polishing for develop)
level of education:
- studying master degree
- Game programmer
- 3 years with unreal Engine
- 3 Years of professional experience as software engine
Links to project pages
Repositories might be set to private, contact me to gain access
Original concept
Started as assignment submission for a Master degree unit, the scope of the project was to investigate whenever unsupervised machine learning could be utilised to generate dungeon levels.
The project make use GAN and Autoencoder to generate new dungeon and rooms layouts from a database provided to them in the learning stage.
The project scored: 75/100
The project make use GAN and Autoencoder to generate new dungeon and rooms layouts from a database provided to them in the learning stage.
The project scored: 75/100
Devlog
Project Scope:
Key Learnings:
- HTTP request methods in Unreal engine 5
- Level generation from 2d/3d int arrays received from the python server
- Convert image to textures in real time
This project was created with the scope of demonstrating whenever unsupervised machine learning could be use in level generation.
Team work
As the team was entirely made of programmers that lived quite far from each other, it was imperative to organise daily stand ups and set clear objective for each team member.
Credits
- Unreal engine 5 frontend, Support with DCGan: Marco Minganna
- Python Flask Backend, DCGan and Autoencoder: Navid Javdan
- Corrective Algorithms: Bhargav Rama Krishna Chitrala
- Evaluation of the models: Nitin Govindaraju