search

edbeeching / godot_rl_agents

An Open Source package that allows video game creators, AI researchers and hobbyists the opportunity to learn complex behaviors for their Non Player Characters or agents
MIT License
902 stars 63 forks source link

The framework is not easy to use. #147

Closed ChernyakKonstantin closed 12 months ago

ChernyakKonstantin commented 1 year ago

I developed my own version of Godot-Python framework for RL when found this repo. I tried to use this repo, but found it compicated in contrast to my version. So I suggest to borrow some ideas from my implementation: https://github.com/ChernyakKonstantin/godot_rl. My version can send any type of data as well through unified interface.

edbeeching commented 1 year ago

Thanks for your comments. Can you provide any more details about what you find difficult to use?

edbeeching commented 12 months ago

Hi @ChernyakKonstantin , I will close this issue unless you have any more details.

I had a look at your repo, I like the protobuf stuff, but I think it requires the user to code too much on the python side. For example, having the reward function defined in python is far less flexible than defining it in Godot. Also for each new env, the user has to create a new GodotEnvironment child class in python.

ChernyakKonstantin commented 12 months ago

Hi @edbeeching.

I'm sorry I didn't reply quickly.

I didn't like lack of capabilities to transfer large data (e.g. images), lack of modularity (it was not easy to understand how code works and whether I'm able to modify it) and that some examples are kinda broken (sorry, I can't remember exact names I tried).

From you comment I see that we follow different paradigms how the frameworks work (I raised mine from self-driving perspective with lots of data), so you can close the issue if you want.