Question How to apply 'personal algrithm' in ml-agents?

I know 'Unity ML-Agents Gym Wrapper' can support personalizations algrithm and run it through gym process,but this method has many constrains,it is a bad solution to utilize it in a complex game.How should I modify the code of mlagents(C# / Python) to make it works which I can test different rl algrithm and design the neural network structures...
Thanks for considerations and I am looking forward to any reply!

 

best way is utiliziing ML agents as a gym .
ML agents supports to wrap the envirounment to a gym

here the docs for
Unity ML-Agents Gym Wrapper

https://github.com/Unity-Technologies/ml-agents/tree/main/gym-unity

 

I think you’re better off directly using Barracuda and the ml-agents low level python API if you need that level of control, and in fact I think that’s exactly what the documentation recommends for such a situation.

 

thanks，you are right！

 

I need more low control，gym-env has many limits