Search Unity

Showcase Snowball Fight ⛄, a multi-agent competitive environment

Discussion in 'ML-Agents' started by simonini_thomas, Dec 2, 2021.

  1. simonini_thomas

    simonini_thomas

    Joined:
    Apr 30, 2015
    Posts:
    33

    Hey there , I'm Thomas Simonini from Hugging Face ,

    We just published Snowball Fight ☃️, a Deep Reinforcement Learning environment for ML-Agents. The game is inspired by Unity's Dodgeball project.

    You can play the game (and try to beat our agent) here

    Or, if you prefer to train it from scratch, you can download the training environment here.

    This is our first custom open-source Unity ML-Agents environment that is publicly available and I'm working on building an ecosystem on Hugging Face for Deep Reinforcement Learning researchers and enthusiasts that uses ML-Agents.

    For more info we published a blogpost explaining what we're going to do.

    I would love to hear your feedback about the demo and the project,
     
    simonini_thomas, Dec 2, 2021
    #1
    andrewcoh_unity, jrupert-unity and simmax21 like this.

  2. andrewcoh_unity

    andrewcoh_unity

    Unity Technologies

    Joined:
    Sep 5, 2019
    Posts:
    162
    Hi @simonini_thomas

    Thank you for sharing this! This is really awesome (and also quite fun to play).

    We love the idea of having a hub where people can share environments/models more easily. We've made steps towards this on our own with things like the Environment Registry though we never got to the phase of letting people push their own environments. There's no doubt that having many diverse and interesting game environments available will be a great thing for the ML/RL community as well as Unity developers!

    Please continue to share and feel free to reach out! We are very excited to see this project grow.
     
    andrewcoh_unity, Dec 6, 2021
    #2