This is a code which I used to create an AI boss same like human with HRL through Malmo, (Warning! :but this code has many problems which I can't fix it )
I believe my idea is fundamentally sound in its inception, but due to the complexity of the Malmo platform, I have struggled for a long time to successfully train my model. Therefore, I am attempting to share this code, hoping to provide a new perspective on AI bosses using reinforcement learning. I also hope that someone can improve upon my code, ultimately achieving a more human-like AI boss.
Firstly, I think the main issue pertains to the API calls on the Malmo platform. Due to the changes across multiple versions, many APIs may not function as intended. This, I believe, is the major hurdle. Initially, I tried to train my model by directly obtaining data from the Malmo platform, but faced significant challenges. In my version, I attempted to use YOLO to train by capturing in-game images. However, I don't think this method is feasible.
In the model, I believe I have successfully implemented the Option-Critic architecture in my code.
I am just a university student, and what I have tried with this code is already at my limit. I hope that this code can offer a little bit of help to everyone.
_< Wish you have a nice day.
About Malmo platform: https://github.com/microsoft/malmo
I also put some articles in this it which include my through.(The article is writed with Japanese and Chinese, maybe I will translate into English)
Here is the picture when I run this py code.
