December 10th 2024 ai thread

Discussion in 'General Discussion' started by Andi Goodwin, Dec 10, 2024.

Thread Status:
Not open for further replies.
  1. Andi Goodwin

    Andi Goodwin Moderator Beta tester

    Joined:
    Jan 16, 2015
    Ratings:
    +848 / 0 / -0
    your videos are set to private

    thanks
    Andi
     
  2. MS_Racer03

    MS_Racer03 Well-Known Member

    Joined:
    Sep 7, 2024
    Ratings:
    +131 / 0 / -0
    Thanks for the correction, I was a bit confused there. :sweatsmile:
     
  3. nedimmb

    nedimmb Member

    Joined:
    Jan 19, 2019
    Ratings:
    +8 / 0 / -0
    This issue I did report it in a different AI thread but it hasnt been fixed yet but I still hope you can fix it this time:

    AIs are too fast and grippy through corner 2 and way too slow throught corners 8 & 10 on Slovakiaring!
     
    • Useful Useful x 1
  4. Nico Kunze

    Nico Kunze Well-Known Member

    Joined:
    Sep 1, 2018
    Ratings:
    +567 / 0 / -0
    Nordschleife 24h layout, ai strength at 106, multiclass session but driving and commenting on gt3 (and if it matters those from around 2015), only did a practice session so cant comment on race pace or post results. The ai pace seems to be imbalanced between the gp section and the nordschleife section. On the gp section i can comfortably make up a good amount of time while on the nordschleife the ai have the upper hand. Would be nice if that could be a bit more balanced. Apart from that it was one of my most enjoyable sessions in a while :) (although multiclass is of course still very far from ideal but i guess that wont change any time soon :D)
     
    • Agree Agree x 1
    • Useful Useful x 1
  5. MS_Racer03

    MS_Racer03 Well-Known Member

    Joined:
    Sep 7, 2024
    Ratings:
    +131 / 0 / -0
    Watkins Glen (Gran Prix with Inner Loop): GTR3 and DPi are ridiculously slow into turn one. They also keep crashing in the first lap in turn one. GTR3 are additionally way too careful going into the inner loop, but too quick going uphill between turn one and the inner loop (I'm full throttle, but they keep pulling away).
    That last issue is something quite common in RaceRoom, at least I've had it numerous times: with the Indycar at Road America and with GTR3 at Zolder for example. The AI is too fast accelerating out of a long corner (carousel at Road America and turn two at Zolder). I don't know how many of those issues have been fixed meanwhile, but you might think about recalculating the resistance of the front wheels while steering for AI in general, I assume that is a problem.
     
    • Agree Agree x 1
  6. Nico Kunze

    Nico Kunze Well-Known Member

    Joined:
    Sep 1, 2018
    Ratings:
    +567 / 0 / -0
    Just in case any video evidence is needed, just look at racerooms story on instagram. The repost of held3r8 quite nicely shows the difference in a 60 second span. Gotta have a look over the next 23 hours tho or itll be gone ;)
     
  7. Skidmark

    Skidmark Well-Known Member

    Joined:
    Oct 9, 2015
    Ratings:
    +240 / 0 / -0
    Track: Salzburgring T1
    Car: TCR 2020
    AI: 95%
    AI opponents:20 (started 20th)

    upload_2025-6-29_17-6-49.png

    Issue: AI frequently cut turn 1. On occasion they make unrealistic lunges up the inside of T1 which take out the player or take out each other resulting in a pile up.

    Evidence:

    I was in the Honda that got punted. Now perhaps there's an argument that as the Lynk&Co was beside me (?? was it ??) I should have left room... maybe maybe maybe. But the pics below show the AI frequently cut the corner. If you follow closely behind in their line you get a cut track warning. If you're close you won't be able to see - early in the race I wondered what I was doing wrong... I kept getting warnings. Then I realised I have to take a different line to that of the AI.

    upload_2025-6-29_17-14-17.jpeg

    upload_2025-6-29_17-14-28.jpeg

    upload_2025-6-29_17-14-43.jpeg

    upload_2025-6-29_17-15-0.jpeg

    upload_2025-6-29_17-15-15.jpeg


     
    • Agree Agree x 2
    • Useful Useful x 1
    Last edited: Jun 29, 2025
  8. MS_Racer03

    MS_Racer03 Well-Known Member

    Joined:
    Sep 7, 2024
    Ratings:
    +131 / 0 / -0
    I have a general question to the developers:
    I have tested different car classes on the Nordschleife this week. Most classes have issues in various corners, but there are corners that almost every class struggles with.
    When I make videos to post them here: Would you rather like to have "issues per corner" (e.g. every five classes struggle with the Karrussell) or "issues per class" (e.g. GT4 class struggles in T1, Karussell and Schwalbenschwanz) ?
    It's not an issue for me to do it the one way or the other way, but I think one of the two options is better for you, depending on your workflow and the way AI is realised in the code.
    So answering this question is actually helping yourself out... ;)
     
  9. MS_Racer03

    MS_Racer03 Well-Known Member

    Joined:
    Sep 7, 2024
    Ratings:
    +131 / 0 / -0
    Oh and another one, also for the community: I read in a different thread that Adaptive AI training uses the last five lap times in the race, without the first lap and the worst lap. I have two questions on that:
    1) So when I do One-Lap-Races (e.g. on the Nürburgring 24h Layout), AI will never learn, how fast I really am?
    2) If I do a Six-Lap-Race without any valid lap times, AI will also never learn, how fast I am?
    Just asking for clarification as I don't want to end up doing One-Lap-Races at the Nordschleife to train the AI when it's a total waste of time... :D
     
    Last edited: Jun 29, 2025
  10. Maskerader

    Maskerader Well-Known Member

    Joined:
    Oct 6, 2019
    Ratings:
    +547 / 0 / -0
    Adaptive AI updates and uses this file: Documents\My games\SimBin\RaceRoom Racing Experience\UserData\Player1\aiadaptation.xml
    You can do a race and then look if the file contents changed.
     
    • Useful Useful x 1
  11. ShortyBuzzGER

    ShortyBuzzGER Well-Known Member Beta tester

    Joined:
    Sep 4, 2020
    Ratings:
    +149 / 0 / -0
    The last 10 laps of yours are saved. and sometimes the first lap is saved as well (never heard that the worst lap won't get saved, that is new to me and not true as far as i can tell).
    Invalid laps sometimes also get saved.
    I haven't found out what triggers what so far, but yeah, just drive a few consistant laps and ai will learn^^
    IF you already have other data on the specific layout OR with the specific class you use, adaptive ai can make cross references and learn as well.
     
    • Informative Informative x 1
  12. majuh

    majuh Well-Known Member

    Joined:
    May 1, 2015
    Ratings:
    +298 / 0 / -0
    Here's another example for BMW domination at the Norisring. AI difficulty was 105, and this was the qualifying result:

    20250705152651_1.jpg
     
    • Useful Useful x 1
  13. Nico Kunze

    Nico Kunze Well-Known Member

    Joined:
    Sep 1, 2018
    Ratings:
    +567 / 0 / -0
    If the bmw is op then the audi might be the opposite. This was only a practice session (33ish ai at 106 strength) but after 90 minutes results should be pretty stable. The ai had best times from 8:09-8:18 with no audi quicker than 8:15 and all 7 audis in the lowest 8 positions
     

    Attached Files:

    • Informative Informative x 1
    • Useful Useful x 1
  14. Anzacastan

    Anzacastan New Member

    Joined:
    Aug 25, 2024
    Ratings:
    +0 / 0 / -0
    DTM 2016 - AI 97
    The AI are still too fast in medium to high speed corners (last sector Hockenheim, last sector Red Bull Ring etc). It’s been this way for more than a year and a half now and still isn’t sorted, as someone who races exclusively against the AI it’s completely ruining the experience for me and I’m pretty fed up.
     
    • Agree Agree x 5
    • Useful Useful x 1
    Last edited: Jul 12, 2025
  15. Maskerader

    Maskerader Well-Known Member

    Joined:
    Oct 6, 2019
    Ratings:
    +547 / 0 / -0
    I'd say it's been always like that, to more or less extent.
     
    • Like Like x 1
  16. Alex Hodgkinson

    Alex Hodgkinson KW Studios Developer

    Joined:
    Jun 5, 2017
    Ratings:
    +1,993 / 0 / -0
    It's most useful to be separated out as much as possible.

    Ie. which car, which corner. Then any other details for example they are too fast in and then run wide and are slow out.

    Generalisations are not all that useful because each individual corner and each individual car has a set of values which can be adjusted.
     
    • Informative Informative x 2
    • Like Like x 1
  17. MS_Racer03

    MS_Racer03 Well-Known Member

    Joined:
    Sep 7, 2024
    Ratings:
    +131 / 0 / -0
    Ok, thanks for the information. That will be a lot of work... :confused:;)
     
  18. Maskerader

    Maskerader Well-Known Member

    Joined:
    Oct 6, 2019
    Ratings:
    +547 / 0 / -0
    Crossle 90f at Suzuka East, standing start. AI skill 95, 98, 102

    On lap 1 there's almost always cars going wide at T1 and then rejoining aggressively towards T2 apex resulting in traffic jams or crashes/spins. Attached laptimes with AI skill 95 and 102.

    Also, AI Castelli and Leblond are always much slower than the rest, is it by design?
     

    Attached Files:

    • Useful Useful x 1
  19. ShortyBuzzGER

    ShortyBuzzGER Well-Known Member Beta tester

    Joined:
    Sep 4, 2020
    Ratings:
    +149 / 0 / -0
    Ehm..depends :D
    If they are real life drivers, they might just be bad and then they will stay like that, then it's by design. If they are made up however, they surely shouldn't have such a big gap and the drivers can be made quicker of course. :)
     
  20. GJDriessen

    GJDriessen Member

    Joined:
    May 3, 2020
    Ratings:
    +14 / 0 / -0
    Thank you for bringing this up. I also find them too fast through this corner compared to other sections/turns.
     
    • Agree Agree x 1
Thread Status:
Not open for further replies.