ML-agents command line environment doesn't sync with Unity environment.

Unity Engine
, , ,
1

I try to start training my model with

mlagents-learn

in the my command line with an Python 3.8.6 conda virtual environment. Unity is already listening to the port (see image) but if I start training in Unity I don’t get any messages in the terminal how I expect.

7858972--997702--Screen Shot 2022-02-01 at 5.47.03 PM.png
7858972--997702--Screen Shot 2022-02-01 at 5.47.03 PM.png1005×487 140 KB

Instead I get an error after a while.

7858972--997702--Screen Shot 2022-02-01 at 5.47.03 PM.png
7858972--997702--Screen Shot 2022-02-01 at 5.47.03 PM.png1005×487 140 KB
7858972--997705--Screen Shot 2022-02-01 at 5.53.33 PM.png
7858972--997705--Screen Shot 2022-02-01 at 5.53.33 PM.png1409×499 355 KB

2

Start Python mlagents-learn first

Then hit play in Unity

Not the other way around

3

@ChillX I am already doing that.

4

In that case do you:

  • Have an agent in the scene with name matching a name in the config ?
  • If yes then Is request decision being called on agent either manually or by adding a decision requester component to it ?
  • Have you disabled auto stepping of the academy ?
  • if yes then are you manually stepping the academy ?