Policy/Guidelines/Route of ML-Agent development team for custom envs and algorithms

Unity Engine
,
1

I have been looking for with no success in the documentation and forum about the guidelines that the ML-Agents dev team will follow.

As custom env and algorithm developers we would like to take advantage of the features that you will develop as far as possible. We are arguing about to use both ML-Agents and ML-Agents-env APIs or maybe only the low one. We have found that we should make some changes to use some features like curriculum, masking… and also to receive observation in other ways. We feel that the code is a bit coupling for this. We would like to use also the current and future options for parallelization…

We would be so thankful if you could take guidelines in this way. With a decoupled code (maybe following DDD or something similar) probably we could contribute with some code.

Thank you very much.

2

Hello,

If you would like to use a custom algorithm or modified environment communication system, then the low-level API is likely the correct path. You may already be familiar with it, but for reference, here is the documentation on that API: https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Python-API.md.