I have parameters like -detect enemy range -firing range -coverage zone width -etc.

for a dodgeball game and ideally would like to run the game over and over until I get the highest win rate against an AI opponent (my AI that I froze at an earlier iteration).

This sounds like a neural network problem now, sort of RL-ish except my characters move based on an FSM.

  • \$\begingroup\$ This sounds like something you could solve in your own script. Write a script that configures these parameters at the start of a match, plays to completion, records the result, modifies the parameters, and repeats. What engine-level support are you looking for to aid with this? There is an ML Agents package, but that may be more involved than the simple parameter optimization you're looking for. \$\endgroup\$
    – DMGregory
    Mar 18, 2023 at 14:45
  • \$\begingroup\$ Thanks @DMGregory, I'll do that and look into running it from Python? I've used Bayesian Optimization with success in the past but it was for optimizing a Python program. \$\endgroup\$ Mar 19, 2023 at 8:45


You must log in to answer this question.

Browse other questions tagged .