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

Reward Functions Magnitude

Discussion in 'ML-Agents' started by asad133, Apr 2, 2020.

  1. asad133

    asad133

    Joined:
    Apr 1, 2020
    Posts:
    22
    Is it fine to give high rewards or should they remain small?
    Is there any difference if I reward with +1000 for reaching goal vs +1

    For that matter, if I penalize on every step to make the agent finish task faster, does the PPO algo work well with small fractions like 1E-5 or is too difficult to distinguish. i.e is an episode reward difference of 0.00001 for example too small?

    Please advise with regards to this. I want an idea on how to sclae rewards and whether I should normalise them
     
  2. LexVolkov

    LexVolkov

    Joined:
    Sep 14, 2014
    Posts:
    62
    https://github.com/Unity-Technologi...Learning-Environment-Design-Agents.md#rewards

     
    andrewcoh_unity likes this.
  3. asad133

    asad133

    Joined:
    Apr 1, 2020
    Posts:
    22
    Thank you