Search Unity

  1. Megacity Metro Demo now available. Download now.
    Dismiss Notice
  2. Unity support for visionOS is now available. Learn more in our blog post.
    Dismiss Notice

Discussion Adaptability of Ml-Agents' ppo algorithm to different environments.

Discussion in 'ML-Agents' started by SkeVinci, Jun 17, 2022.

  1. SkeVinci

    SkeVinci

    Joined:
    May 8, 2022
    Posts:
    6
    It seems that when we use ml-agents to train an agent, we just create our own environments and set up parameters in the .yaml file. Thus, I'm wondering the adaptability of its ppo algorithm to different environments. How can it train different agents in different environment well?
     
  2. JulesVerny

    JulesVerny

    Joined:
    Dec 21, 2015
    Posts:
    47
    Not sure I understand the Query. The Agent will try to learn to perform in the environment you set up and train it for.
    Yu can create lots of different environments, different sensor and actuator(action) for your agents to operate in. But you will need to Train that Agent to perform in each of those environments. And you will probably need to configure, or hypertune the Agent PPO,GAIL, SAC algorithms with different training hyper parameters for each. You may be extremely lucky, or very patient with the use of the same algorithm training parameters for use on different agents in different environments. But your Trained Brain, will be optimised against the specific environment it was exposed and trained against.
     
    SkeVinci likes this.