Search Unity

  1. Welcome to the Unity Forums! Please take the time to read our Code of Conduct to familiarize yourself with the forum rules and how to post constructively.

Feature Request Training multiple brains in parallel

Discussion in 'ML-Agents' started by NanushTol, Dec 29, 2022.

  1. NanushTol

    NanushTol

    Joined:
    Jan 9, 2018
    Posts:
    121
    Hi,
    I have a hierarchical multi-agent multi-brain model,
    It is a hovercraft that can jump and fly around.
    The whole model is made of 2 agents each with its own brain, both are on the same aircraft.
    The "Driver" brain is responsible for moment-to-moment collision avoidance, movement & orientation.
    The "Navigator" brain is creating the "target direction vector" & "aim look-at vector" and feeds them to the Driver.

    I have trained the Driver using curriculum learning and it is working great.
    Now I want to continue and train both brains together in a self-play mode, and I want both of the brains to train simultaneously.
    I have read that this is not possible at the moment and curios if this would be possible in the future?
     
    Last edited: Dec 29, 2022
  2. GamerLordMat

    GamerLordMat

    Joined:
    Oct 10, 2019
    Posts:
    180
    either you use POCA or you just fix one brain.
    but as far asI know you can have mutiple behavoiurs training
     
    hughperkins likes this.
  3. NanushTol

    NanushTol

    Joined:
    Jan 9, 2018
    Posts:
    121
    POCA works when all agents have the same brain,
    so I'm trying with a single brain that has 2 tasks, then have 2 agents each with a different task, it set with a different goal observation,
    when one task is selected I also pad the other task input observations.
    maybe it will work.
     
    hughperkins and GamerLordMat like this.