1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.
Dismiss Notice
Like RaceDepartment on Facebook.

Plugin: To change Values in Track AIW

Discussion in 'rFactor 2' started by Melv, Jan 6, 2016.

  1. Melv

    Melv
    Premium Member

    Hi Guys,

    I was wondering if there is a way to have a Plugin or tool to change the values in the AIW files for a track without the need to repack and reinstall the tracks.
    Currently this can be done manually however then we have to repack and reinstall the tracks.

    For an offline player with multiple tracks this is a bit tedious.

    Background -
    1. On the same difficulty settings the AI are faster on some tracks and slow on some
    2. Also I am better at some tracks and poor at others
    3. By adjusting the worstadjust, midadjust and bestadjust values in the AIW
    - I am able to keep the overall AI strength the same in the UI (100%)
    - I am able to adjust the AI for each track exactly to the difficulty I like ( such that I am in the bottom half of the grid)
    - Combining this with the RCD files this ensures over multiple races I am in a tight fight with the same 2-3 drivers.

    Any thoughts or pointers would be great.

    Regards

    EDIT: The final aim (holy grail) is sort of to have adaptive AI. After each race, check players fastest lap & check AIs fastest lap (or average lap time) for a particular track.
    If the difference is below or above a certain value (configurable parameters e.g. if player lap is faster by 0.5 secs or 0.2 secs) then adjust the track AIW to speed up (or slow down) the AI. Increases could also be configurable (e.g. steps of 0.02)
     
    Last edited: Jan 6, 2016
  2. Gijs van Elderen

    Gijs van Elderen
    Premium Member

  3. Melv

    Melv
    Premium Member

    Hi @Gijs van Elderen

    Thanks for the link.

    I have used AI learning before for rFactor 1 and then too I needed to edit the AIW files (much easier because of loose files) to ensure consistent performance (relevant to me) across all tracks.

    The AI learning only ensures the AI brush up on their driving skills (which usually means they will drive faster) however this does not provide any consistency of AI ability across multiple tracks. Also if I find the AI in a track is too fast (incorrect values for AIW) AI learning will only make them faster and not slow them down.

    Only way that I know of to keep AI consistent (pace wise with respect to the players pace, which varies from person to person) across multiple tracks is to -

    1. Keep the AI strength in game constant (e.g. at 100%)
    2. Test each track
    3. Go into AIW and adjust the values for (worstadjust, midadjust, bestadjust) to get a consistent AI
    4. Edit RCD files to ensure driver consistency on top of this
    5. After few months if you notice you are faster and want to increase the AI speed. Now all you need to do is raise the AI difficulty without having to adjust the tracks individually or having to worry that on some tracks the AI will suddenly all become into schumachers, sennas or the other way round into alex yoong :p

    I could be wrong, but atleast from all my tests this is what I gather.
     
  4. I'm keeping a table of AI settings at X track with Y car for myself. It's mainly a guide as I'm not worried if I'm off by 1% either way and my skills change (hopefully always improving, but then there are the off-days). Of the ISI roadrace circuits, I've only found Sao Paulo AI to be well off the pace of the other tracks and posted a minor rant on ISI's forum about it.

    Now when you throw 3rd party tracks into the mix, I figure on begging matf1 (Mat Holloway, I think) for some of his personal AIW work... he makes new AIW files for his personal use, adding block and wet paths where they're missing and improving AI lines.
     
  5. I did quite alot of work with the ai in rF1 and in rF2 it's not as convenient to edit the worstadjust, midadjust, bestadjust. For convenience you could play around with the "AI Limiter" = used to make racing closer) I know it's not exactly the same as using the settings in the AIW but the "AI Limiter" might help to accomplish what your attempting to achieve but with a different approach.
     
    Last edited: Jan 6, 2016
  6. That's freaking awesome feature, far out some matrix stuff, AI that learns tracks, far out this sim has so many features hidden away its unreal, I love donington, but AI struggle big time to take Old hairpin at a decent speed, so this be my first test.
     
  7. Melv

    Melv
    Premium Member

    @Hash_V8

    The AI learning is an excellent tool.

    you have to remember though there is a limit to what the AI can learn and having a good AIW (AI line) is the most critical. If the AIW is not defined well for the track (hairpin) the AI learning will not resolve this issue.

    Atleast thats my undertanding and observation from my tests.
     
  8. Correct, if you already have a very poor/faulty AI line (AIW) then your AI learning isn't going to do much good. I also discovered when i made a whole new AIW for a track in rF1 called "Phoenix91" (street circuit) in which i went to great lengths to prevent the AI from scraping the walls while still attempting to keep the AI very fast while racing through the city. The outcome turned out to be very good so i thought maybe i could get the AI a tad faster using the AI learning but it actually made them slower! So, if your track's AIW is already great/flawless then you might very well risk a poorer AI line using the AI learning feature!
     
    Last edited: Jan 7, 2016
  9. Is there any guides on AIW editing\creation?
     
  10. Last edited: Jan 7, 2016
  11. Awesome, ty mate.
     
  12. Your welcome, be sure to read my additional note recently added. Enjoy, it's alot of fun to create the best AI.
     
  13. I may have to update that if anything has changed in recent Builds (highly unlikely though), but for the most part it works as intended.

    Just to correct you on this slightly, the Auto Calibrate ="1" is used to "brush up" as you say, but with "2" they actually learn different lines through the corners and use that in future races. Albeit it's still not great as such but it does a good job as is, in future there may well be more AI Learning Attributes that could be used and maybe for Ovals too, when/if that happens I don't know....if it hasn't already!
     
    • Like Like x 1
  14. Melv

    Melv
    Premium Member

    Thanks @Marc Gravett

    Learnt something new today. will try this out.
     
  15. The AI will attempt to make the "AI fast line" (AIW) faster, usually at all costs that can sometimes degrade the original AI fast line at troublesome tracks instead of making it better, though i did achieve some great results too!

    I first experimented with this feature in rF1 starting in 2005 before determining if it was necessary to make a whole new AIW for a troublesome track.

    Autocalibrate AI Mode # "0" default, edit to "1" to activate) When in a test day with 1 AI, AI will attempt to perfect his driving line, and save his knowledge for future use". Important to edit back to "0" when finished.

    AI Calibrate Sample Size #"5" default ) When AI calibration is set to 2, this is the number of trials the AI should run while testing each parameter. rF1 didn't have this option until rF2 came along which is an interesting method used by the AI.
     
    • Like Like x 1
  16. definetly gonna try this out thanks for the info !