Search Unity

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

Official Post Your ML-Agents Project

Discussion in 'ML-Agents' started by jeffrey_unity538, Nov 13, 2020.

  1. Keith-Rek


    Jun 5, 2016

    This is a game pitch of my new commercial AI horror game that uses ML-Agent toolkit for training RL agents (ghost) inside the game
    albayona likes this.
  2. albayona


    Dec 5, 2023

    Train a deep neural network using PPO and SAC to perform the following cooperative/ competitive task:
    Two teams of robots, Blue and Red, are placed inside a cage in a maze-like area with a central base. The maze is composed of blue and red squares representing the territories of each team. They must find a way out of the cage and navigate the maze.
    The agents are equipped with the ability to shoot and paralyze adversary agents. However, they can only do so if they are on a square matching their team's color, adding a strategic element where agents must choose between offensive and defensive strategies based on the location.
    Once the agents escape the cage, they can explore the maze area for food seeds. The agents do not know the exact locations of the food seeds and rely on their exploration capabilities to discover and collect them. The central base position is known to the agents, serving as a reference point for navigation.
    To keep track of the location of a food seed, agents can leave tracks or trails through the maze as they move. These trails act as markers that the agents can follow later to revisit previously discovered food seed locations.
    The agents aim to collect a predetermined amount of food seeds and return them to the central base. The team that successfully collects the required number of food seeds first wins the game.
  3. Energymover


    Mar 28, 2023
    Here is the current model I have trained for NASCAR style racing. I recorded two consecutive races with 8 cars; can be hard to tell with changing the camera. :D
  4. DrMadVibe


    Jun 29, 2022
    TensorWar is a top down ml-agents combat training environment. This has been my hobby for about 2 years and it;s my first experience programming. My hope is to release this on github and continue development, introduce new agent and game play types and get better every day. I'd like it to be a fun entry into machine learning and a framework for making simple games. Here is some agent inference.

    And some training at different timescales.

    It's still in it's infancy, but 2 years is a long time not to share this. I'm excited about the possibility of playing against someone who has trained thier own agents. I'd love feedback. Thanks!
  5. smallg2023


    Sep 2, 2018
    DrMadVibe likes this.
  6. MarkTension


    Aug 17, 2019
    msrafiyenko likes this.