Resolved Brain update() control?

Hey!
I am trying to control ML agents update without touching UNITY's step time and fixed time.

My main aim is to train physics character agent in a very extremely high resolution physics.
So I want to simulate 1sec real-time physics in about 30sec.
And then give observations() and get the outputs() of the brain every 30sec/60 , not every update.

Should I call the observations() and get outputs() every 30/60 second in order to control python ml_agent brain update() ?

 

First you'll want to disable automatic stepping - https://docs.unity3d.com/Packages/c...ity_MLAgents_Academy_AutomaticSteppingEnabled

Then you can just step the academy manually at whatever rate you'd like - https://docs.unity3d.com/Packages/com.unity.ml-agents@1.7/api/Unity.MLAgents.Academy.html#methods

 

If you want full control over the stepping, then disabling automatic stepping and manually calling Academy.Instance.EnvironmentStep() is the best way to do this. You should call Agent.RequestDecision() on all Agents that you want to step, then call Academy.Instance.EnvironmentStep().

You do not need to do anything else to provide the observations (the Agent's CollectObservations method will be called, along with any ISensors), and the actions will be applied in your OnActionReceived() method.

 

Thanks @everyone!!!