Fixed vs Adaptive AI

Discussion in 'General Discussion' started by stlutz, Sep 5, 2021.

  1. stlutz

    stlutz Member

    Joined:
    May 31, 2021
    Ratings:
    +8 / 0 / -0
    To me, the most helpful enhancement to adaptive AI would be if the game told you what the adaptive AI level for the selected car/track combo is. That is, instead of just having "Adaptive" in the dropdown for AI strength, say "Adaptive (101)" for example.

    That accomplishes a couple of things. One is to tell you if you're getting better or not (kind of like your rating in MP). The other is to help one adjust for a more customized experience. That is, if it says 101 but I want to race closer to the front, maybe I'd chose 100. Or if I want a tougher challenge to get better, maybe I'd pick 102.
     
    • Agree Agree x 9
  2. ravey1981

    ravey1981 Well-Known Member Beta tester

    Joined:
    Apr 15, 2018
    Ratings:
    +873 / 0 / -0
    I completely agree with this, and I would go a step further and allow the AI level to be turned up or down (in a similar fashion to the ffb multiplier) while actually in a game session rather than have to quit out and start all over again.
     
    • Agree Agree x 6
  3. ducman888

    ducman888 Well-Known Member

    Joined:
    Jan 7, 2018
    Ratings:
    +172 / 0 / -0
    I raced a P2 in a multi-class weekend of P1, P2, GT3R and Porsche 911 GT3 Cup where I selected the cars and used the same combination for 10 min practice, 15 min qualify and 60 min race each at Silverstone, Watkins Glen and Spa (single race not Championship) with 104 AI difficulty.
    Qualifying was close within each class, racing was good but faster AI struggled at times to navigate around the slower classes. I won my class by several seconds. So I decided to give the AAI another go.

    I set the AI from 104 to Adaptive and raced at Silverstone with the exact same cars I was using for the previous 3 rounds, I was fastest in my class by 3+ seconds and out qualified 2 of the 6 P1 drivers. In the race I wound up winning overall and lapped the entire field except for the P1s, although the second place car was 45 seconds behind.

    I then raced again at Silverstone (3rd time) with the same settings and was again winner in class by an entire lap and out qualifying my class by a few seconds.

    As was mentioned in an earlier post I too am consistent with invalid laps usually coming down to avoiding an unusual AI braking point or last moment adjusted line.

    When doing AAI with mutli-class, are the other classes being 'trained' simultaneously to my class and each other. Sorry, but I am confused as to why I have had very little luck with the AAI. After 1,400 hours logged with the majority of it in Single Player seems like I should have collected enough data but it seems it is either feast or famine when it comes where I finish. Should I just continue plugging away with AAI or am I doing something wrong? Thank you.
     
  4. Muzarati

    Muzarati Member

    Joined:
    Jan 27, 2018
    Ratings:
    +20 / 0 / -0
    I too gave the AAI another go, my experience is more consistent, same car/class, track and settings. R1 out qualified and won by big margin, R2 managed in dying moments to qualify in second (most fun session of the day) got a good start and won by 1.5sec. R3....couldn't touch them...qualified dead last....ended race dead last!
    My lap times were consistently the same through all 3 races....game over!
     
    • Informative Informative x 1
  5. ravey1981

    ravey1981 Well-Known Member Beta tester

    Joined:
    Apr 15, 2018
    Ratings:
    +873 / 0 / -0
    To be honest I'm not sure that AAI is working perfectly at the moment. A couple of things though. It only makes an adaption each new race session and it only looks at the last few laptimes, so in terms of training them it's better to do lots of 10 lap races than one 30 lap race (for example). It's normal that they will be too slow, then too fast and eventually narrow in on the correct level. The correct level should put you mid pack so that there is always cars to race, it's very hard to win once adapted.

    This all assumes it's actually working as intended though. I will make a backup of my file and start from scratch and do a bit of research this weekend.
     
    • Informative Informative x 1
  6. anno900

    anno900 Well-Known Member

    Joined:
    Dec 27, 2018
    Ratings:
    +47 / 0 / -0
    What means "few" laps. I posted in this thread, that AI adapted at one of my FRJ chamiopnships. I was happy that it worked and contiuned with such a setup, so many short races on the same track so that at least the last 1-2 races are fun.
    But what happened. I started a M2 cup championship with 6 5 lap races per track. I did that on 3 tracks then I stopped.
    1st track was somehow ok. In first 2 races AI was way too fast. I do not know which AI level was set by AAI but much too high, even if the M2 is an easy to drive class and with fixed AI I was on 106% to run top 5. Race 3-6 were ok. AI still sightly too fast but (as always) too slow in packed racing. Next track. AI way too fast at 1st race. I thought "ok" even if I would have thought, that based on my many DB entries it should not be that much faster. The other 5 races. A very very slight adaption to my pace but even at race 6 AI was too fast. 3rd track. Same behaviour as at track 2. Nearly no adaption even if I was significantly slower. The tracks were grand prix tracks, so laptime with M2 little over 2 mins and fastest AI times were always 2 or more seconds faster (only at 1st track, where the AI adapted to my pace).
    So I don't understand where AI takes that high level for the 1st race at a track. I have very few cars, where I needed quite high AI level to run competitive. WTCC2016 is such a case where I ran at 109% (before FFB update) whereas at many other clases I'm arounf 99-102%. After FFB upgrade my AI level increased for whatever reason. Cars are simply better to drive ;-).
    I switched back to fixed AI did a short championship with M2 at 106% and had fun at every race. I started yesterday DTM92 at 102% any nealy hit the righ strenght (maby AI is 1% too strong) but again, fun race. Quality time as someone said before. If you only have limited time then it is frustrating that you need to do "nonsense" races just to train the AI. That you have to train makes sense but after 3-4 5 lap races, the game should know which pace level you are at and then simply addapt in bigger portions.

    The disadvantage of fixed AI is, of course, that you cannot adapt AI on "bad tracks" to your pace for a propper championship. But meanwhile I know at which tracks I'm horribly bad (or AI base speed is simply wrong calculated for such a track) or way too fast and simply do not race these tracks. Luckly none of my favorites is one of those bad/good tracks.
     
  7. Maskerader

    Maskerader Well-Known Member

    Joined:
    Oct 6, 2019
    Ratings:
    +352 / 0 / -0
    What about the slowest AI and where did your end up in the race? If AAI's goal is to put you in the middle of the grid, it's expected that the fastest AI will be 1+ seconds faster than you (plus the bigger the track the more this difference will be).
     
  8. Hendrick

    Hendrick Well-Known Member

    Joined:
    Oct 7, 2020
    Ratings:
    +64 / 0 / -0
    Sorry for the late reply on that. AAI seem to have settled at being 1.5 - 2.5 sec per lap faster than me. My best lap is third worst from a grid of close to 32 cars. It needs more adjustments.
     
    • Agree Agree x 1
  9. anno900

    anno900 Well-Known Member

    Joined:
    Dec 27, 2018
    Ratings:
    +47 / 0 / -0
    Slowest AI fastest lap times were also 2 or more seconds faster. Race setup. 10 oppeonetnts, I start 6th (so in the middle), 5 laps.
    I don't know exactly but I finished last or 10th (whilst other AI had spins). The first 2-3 laps I could battle with them but then the pack in front of me drove away and I just defended against the ones behind me. Once they passed me, they drove away quite quickly. That is basically "always" those races run. First 1-3 laps I can gain more or less places and then I only defend. The slowest fatsest lap of AI driver, that finish in front of me, are much fatser than my fatest kap times. The dives behind me (if there are any, have, of course fatsest lap times like mine, cause they could not pass me).
    The fact that AI cars do get quicker after 1-3 laps (so when the full packed racing is over and they devide into smaller groups) is not an issue of AAI, that happens with fixed AI as well. But the AAI level seems to be too high and they do not adapt quickly enough if they are much faster. So for me it is easier with fixed AI once I have found my correct level. The only bad thing is, that I finish top 5 at some tracks, at other only top 10 and at few I battle with the last.
     
  10. anno900

    anno900 Well-Known Member

    Joined:
    Dec 27, 2018
    Ratings:
    +47 / 0 / -0
    An addition to my longer post above. The AAI in 2nd race was even faster than in 1st race where they were already too fast. So adaption started from 3rd race on (if there was even any adaption).
     
  11. ravey1981

    ravey1981 Well-Known Member Beta tester

    Joined:
    Apr 15, 2018
    Ratings:
    +873 / 0 / -0
    I notice you keep changing tracks though.. adaptation is best done using the same track as it's done on a class/track basis. If no data exists for a track it will use other data to make a guess but since we are all better and worse at some tracks that guess may still be way off.

    It's not a perfect system by any means and I'm not going to pretend it is either. I generally use fixed AI and just adjust per class/track.
     
    • Informative Informative x 1
  12. anno900

    anno900 Well-Known Member

    Joined:
    Dec 27, 2018
    Ratings:
    +47 / 0 / -0
    I did 6 races at each track. 6 at Shanghai, where AAI adapted. Other rounds containing 6 races a 5 laps were Silverstone and Spa. And there was barely an adaption not even in 5th nor 6th race. Anyway. I'm back at fix AI and do a championship with dtm92 in 102%. Suzuka and portimao so far with 2 races at each track and with 102% I have nearly round a sweet spot, maybe 1% too hard to win. 4 fun races.
     
  13. ravey1981

    ravey1981 Well-Known Member Beta tester

    Joined:
    Apr 15, 2018
    Ratings:
    +873 / 0 / -0
    I seem to recall from past discussion that 8 valid laps are needed for a good adaptation. That doesn't include the start lap either. So ideally you'd want to run 10 lap races.

    If you are running fixed at around the correct level that data also feeds into the AAI system and in theory should help to get a more accurate adapted level.
     
    • Informative Informative x 2
  14. anno900

    anno900 Well-Known Member

    Joined:
    Dec 27, 2018
    Ratings:
    +47 / 0 / -0
    I'm convinced that AAI Works if you drive the same class A lot and on not too many tracks. But if you switch class often and tracks then there are too many Parameters.
    That is why AAI worked so well at nascar 2003. Only one class and basically Just 4 types of tracks (short, fast, superspeedway and road).
     
  15. pixeljetstream

    pixeljetstream Well-Known Member Beta tester

    Joined:
    Jan 29, 2015
    Ratings:
    +412 / 0 / -0
    Any AI race is logged in the aiadaptation file. It contains your times per class per track as well as AIs and the level it was set to. You can use the ai adaptation tool to generate a nicer html page of its contents.

    Also would love to see dynamic adjustment of AI levels possible in garage menu or so. Fixed AI is fine. But finding the right level takes a bit of adjusting and I feel like adaptive AI doesn't give that much value given it takes time to dial in, might as well manually try a few ai values...
     
    • Agree Agree x 1