search

JarkkoPar / Utility_AI_GDExtension

This repository contains the binaries and example project for the Utility AI GDExtension.
MIT License
75 stars 2 forks source link

Should considerations be normalized as a mean / percentage? #10

Closed madeleineostoja closed 10 months ago

madeleineostoja commented 10 months ago

My understanding of utility AI is still nascent so this is part question part feature request — I'm in early stages of setting up and tuning my ai agent I find myself fiddling with consideration curve values a lot just to compensate for one behavior having a single consideration and another having many. This makes the system feel quite brittle, because when I add a new consideration to one behavior, I have to go and update consideration weights for other behaviors to account for the summed extra.

From my understanding, shouldn't the considerations be averaged to give the behavior a normalized score? So all behaviors are scored (for example) 0 - 1 based on the weighted curves of the considerations it has, no matter how many it has?

madeleineostoja commented 10 months ago

Ignore me I just discovered the mean evaluation option in consideration and sensor groups 🤦🏻‍♀️

Maybe a documentation issue? Maybe a note in tutorial "next steps" would help, also noticed the tutorials don't mention behavior actions