Search Unity

  1. Unity support for visionOS is now available. Learn more in our blog post.
    Dismiss Notice

How to use ML-Agents in my GridSoccer Video game

Discussion in 'ML-Agents' started by Pouya_Jigsaw, Feb 7, 2020.

  1. Pouya_Jigsaw

    Pouya_Jigsaw

    Joined:
    Sep 9, 2018
    Posts:
    10
    I've been working on my bachelor's degree project, which is about creating a video game and implement AI algorithms on it.

    The game happens on a 6 * 10 board, and it features two players indicated by circle shapes: red and green. Both sides of the board have a goal which is a block indicated by blue color. At the start of the game, the green player has the ball, which is yellow and has to move to the other side of the board and reach the opposite goal to score. Every movement happens in a time cycle which is 0.5 seconds for my game. if the two players collide on a block, the ball passes to another opponent so the attacker becomes the defender. That's the essence of the game. it's good to note both players can only move one block per cycle.

    My question is, I want to create an AI opponent which is the red player. My issue is how can I train him in order to play with me using ML-Agents toolkit?

    The obvious thing is in order for this to happen, the opponent should learn from the player, so it means i have to play more than hours to make an agent learn from my environment? Can you suggest a better Idea?

    Is it possible that I create a Heuristic AI using the Minimax tree and let it play with the RL agent? The issue is the game is not totally turn-based, in every cycle, both players have the chance to move.

    I don't want you to finish my assignment or something I just want to learn and do it myself. my thesis project is based on ML-Agents toolkit and i really want to use its cool features.



    upload_2020-2-7_9-40-55.png
     
  2. TreyK-47

    TreyK-47

    Unity Technologies

    Joined:
    Oct 22, 2019
    Posts:
    1,810