General AI Performance Discrepancy Reports (No Discussion)

Discussion in 'Automobilista 2 - General Discussion' started by Renato Simioni, Sep 14, 2023.

Thread Status:
Not open for further replies.
  1. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    I think they were updated in the last beta. I was quicker than the 120 AI in that class by around a second a lap, which can't be the goal.
     
    • Informative Informative x 1
  2. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    20231112214109_1.jpg F-VintageG2 AI at Kyalami is another one of the too slow ones I think. The picture below is on a heavy rubbered in track on 113 difficulty. Could it also be that the Lotus 49C is by far the fastest of the bunch on straights?

    Also, after re-testing V10G2, I can confirm it's still slower than me on 120 by around a second on Spa-Francorchamps and a little more than half a second on Interlagos, so that report still stands.
     
    • Agree Agree x 1
    • Informative Informative x 1
  3. sampopel

    sampopel Active Member AMS2 Club Member

    Joined:
    Mar 31, 2020
    Messages:
    330
    Likes Received:
    147
    Opala Old Stock at Taruma (no chicane), default AI @100% / default setup / dry - going through T3 heading towards the hairpin, AI are slow, I easily power past them on full throttle - not sure if they're lifting off to preserve tyres? They're a bit easy to pass through the hairpin, too.
     
    • Informative Informative x 1
  4. Mazdaspeed

    Mazdaspeed Well-Known Member AMS2 Club Member

    Joined:
    Aug 25, 2022
    Messages:
    1,402
    Likes Received:
    727
    Strange, this is the AI pace in my case at 108, faster than yours o_O. Automobilista 2 Beta 11_12_2023 6_12_19 PM-min.png



    Also the fastest model at Kyalami at least for me at default setup is the V12 model. The Lotus is the most balanced and overall quickest car though, but on tracks with long straights the V12 is hard to beat.
     
  5. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    That is extremely weird ... guess I have to verify my files...
     
  6. sampopel

    sampopel Active Member AMS2 Club Member

    Joined:
    Mar 31, 2020
    Messages:
    330
    Likes Received:
    147
    I haven't raced in the wet much but tried the Formula Hi-tech Gen1 (I was in Model 1 racing against all models) @ Guapore, AI @100%, default AI (ie. not custom) and default setups, wet tyres, TC off, and was significantly faster than them through T1, T2 and T3, also coming off the final turn onto the start/finish straight - easy overtakes every lap.
     
  7. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    I have to amend this report: On 120 difficulty in a 10 lap race, the AI now get into the 1,18's (well, one AI did). However, they are losing at least half a second if not more in the T3 (the tiny kink) leading up to the long right hander. The player can go through there full throttle easily, the AI can't. This leads to situations like this one:

     
    • Informative Informative x 1
  8. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    Group A AI are decently quick in the corners, but they brake much too early on 112 difficulty, making them easy pickings for the player in every longer braking zone.

     
    Last edited: Nov 15, 2023
    • Informative Informative x 1
  9. kkdrummer

    kkdrummer Well-Known Member AMS2 Club Member

    Joined:
    Jun 25, 2016
    Messages:
    1,560
    Likes Received:
    751
    My baseline with the stock 2023 is 114% at Barcelona. This is me matching the fastest ai in quali, rubbered track 31degrees track temp, sunny weather.
    Porsche gt4, same conditions as above, I am 1.5 sec a lap slower than the fastest Porsche.
     
    • Agree Agree x 1
  10. whokilledduncan

    whokilledduncan Active Member AMS2 Club Member

    Joined:
    Mar 11, 2016
    Messages:
    229
    Likes Received:
    125
    SC2023 Barcelona no chicane baseline 108% with everything set to default.

    Omega Stock Car, identical settings has a ~1 second per lap swing in my favour. Their lines seem pretty fine to me except that they don't use (abuse?) the kerb to straighten T13 like I can / do. They are very slow out of turns 7, 9, 13 and 14, especially 9 and 13.

    I often felt like had had much better top speed down the home straight, but hard to be sure how much of that was better exit and slipstream.

    Screenshot 2023-11-18 130846.png Screenshot 2023-11-18 133515.png
     
    • Informative Informative x 1
  11. whokilledduncan

    whokilledduncan Active Member AMS2 Club Member

    Joined:
    Mar 11, 2016
    Messages:
    229
    Likes Received:
    125
    SC2023 Barcelona no chicane baseline 108% (High aggression) with everything set to default.

    Lotus 23, identical settings is near perfect for overall lap time, but the sectors are somewhat uneven. They are a bit faster out of turns 3 and 4, a little slower through T8 and T10-11 and slow out of T9 and T13. They also brake much better than I can, especially into T1, T4 and T10. I'm not the best braker anyway and have recently adjusted my pedal settings though, so..... Screenshot 2023-11-19 131010.png
     
    • Agree Agree x 1
    • Informative Informative x 1
  12. whokilledduncan

    whokilledduncan Active Member AMS2 Club Member

    Joined:
    Mar 11, 2016
    Messages:
    229
    Likes Received:
    125
    SC2023 Barcelona no chicane baseline 108% (High aggression) with everything set to default. (I can qualify P1 or P2 at this level).

    MCRS2000 with identical settings I'm almost two seconds a lap slower than the fastest AI. I have no idea exactly where they are gaining (beyond the sector splits, which indicate mostly in S1 and S3). In the race I climbed from P22 to P17 and was reasonably well balanced with the cars around me. My fastest lap was over 1.5s down on the fastest AI.

    Screenshot 2023-11-19 162118.png
    Screenshot 2023-11-19 162152.png
     
    • Informative Informative x 2
  13. sampopel

    sampopel Active Member AMS2 Club Member

    Joined:
    Mar 31, 2020
    Messages:
    330
    Likes Received:
    147
    Stock Car 23 Baseline - 108%

    Ginetta G40 Cup
    Snetterton 300
    AI @108%, Max Aggression
    Default set-ups.
    Default track progression.

    AI are very slow. The slowest were over 6 secs back.
    upload_2023-11-20_20-24-52.png

    I started the race but was already 3 seconds ahead after the first lap, so quit. ie. they're were equally slow in quali and race.

    As appears to be the case with numerous classes, their first flying lap is usually the fastest they do. Should they not get faster as the session advances - ie. track rubbers in and they get into their groove/rhythm on that track?

    Incidentally, the field was seemingly identical cars but when I was monitoring the AI in cockpit view, there were 2 different engines tones. Some cars had a slightly deeper/throatier tone, the others had a slightly higher, more metallic tone. What's the reason for this?
     
    • Informative Informative x 2
  14. Mazdaspeed

    Mazdaspeed Well-Known Member AMS2 Club Member

    Joined:
    Aug 25, 2022
    Messages:
    1,402
    Likes Received:
    727
    Did some more testing of the AI Fuel weight penalty on some of the Formula classes.
    108% difficulty, max aggression, normal fuel and tire wear.

    F-ClassicG4: Hockenheim 1988.
    The AI is a good 1+ second faster than me from qualy to race, didn't take any pictures, but I was able to match Nyles on qualy, but on the race he was 1.2 seconds faster

    F-HitechG2: On my previous test at Imola I could match the AI with the rich fuel mixture on this class.
    Barcelona 1991 however has a huge discrepancy, basically the same as before the update. I lose a good 6.5 seconds from qualy to race, while the AI only loses 3.5 seconds, photos below.
    Automobilista 2 Beta 11_20_2023 11_51_24 PM-min.png Automobilista 2 Beta 11_21_2023 12_13_24 AM-min.png
     
    • Informative Informative x 1
  15. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    V10G2 in qualifying on Nürburgring (please notice how they are faster in S1 and S3 and lose their time in S2; the same thing also happens with most other cars) vs V10G2 in the race on wet tires with weather set to "rain" (both on 120 difficulty)

    Me: 25.855 - 36.666 - 26.354 -> 1,28,875
    AI in the Quali: 25.361 - 37.115 - 25.879 -> 1,28,355
    AI when simulating rest of Quali: 24.702 - 36.968 - 25.857 -> 1,27,527

    Me (Wet Race): 32,601 - 46,301 - 30,299
    AI (Wet Race): 31,600 - 44,781 - 30,140 (2nd place, 1st place was even faster)

    Now I have nothing against being slower than 120 difficulty, but it needs to be consistent. In general, Nürburgring GP AI seems to have too much pace in 1st sector (especially T1) and 3rd sector (especially chicane exit) and too little pace in S2, quali looks roughly the same for F-UltimateG2.

    However, the AI are ridiculously quick through the S-section and braking for the corner after it; I was unable to catch any of them through there. In general they seemed to be able to accelerate a bit too well in the wet, but these aforementioned corners were really bad.

    If I can keep up with non-simulated top AI times in quali, I should also be able to in the race.

    Also, Nürburgring GP in general might deserve another look in terms of AI lines because of the pace discrepancies throughout the lap (5 tenths quicker, 5 tenths slower, 5 tenths quicker amount to more than one second, even though I admit my lap wasn't perfect).

    In case it matters, this was my quali lap:

     
  16. BRIAN ROGERS

    BRIAN ROGERS Member AMS2 Club Member

    Joined:
    Mar 2, 2021
    Messages:
    109
    Likes Received:
    22
    Build 1.5.3.1
    Track: Hockenheim Short A w/Heavy Rubber setting
    Cars: Copa Fusca and Fusca Classic FL VW
    Setup: Single Player, 110% AI at Meduim Aggression

    Issues: AI speeds extreme vary from car to car.
    Copa Fusca: +2.500 slower than pole, getting run over.
    Fusca Classic FL (VW): Pole by -1.300.

    This I feel is a normal representation of AI performance for many cars. When racing Single Player races, it can take me :30+ minutes to find an AI strength that makes an enjoyable, competitive, race. I'd expect for one's skill level to be able to set AI at or near the same from car to car, track to track. I can certainly understand a 1-5% variances to account for knowing the car/track or just being better/worse skills from one to another (for me I suck at fast open wheel cars), but I've seen a range anywhere between 95% and 118%, and that is not realistic and again very time consuming when playing offline to find a "sweet spot" to race.
     
  17. Dicra

    Dicra Local Gamepad Ambassador AMS2 Club Member

    Joined:
    May 1, 2020
    Messages:
    1,460
    Likes Received:
    1,004
    With the M1 Procar @Jerez Moto I'm 0.5 seconds quicker than the AI on 115 difficulty in qualifying (exclusively because they don't drive the almost flat T4 as quickly as the player can), however in the rain (normal rain setting) I'm 5 seconds and more quicker per lap on the same difficulty. Deserves another look imo.

    With the Procar you can slide excessively in the rain and gain time; in not doing that the AI are losing out a lot.
     
  18. sampopel

    sampopel Active Member AMS2 Club Member

    Joined:
    Mar 31, 2020
    Messages:
    330
    Likes Received:
    147
    Stock Car 23 Baseline - 108%

    Opala 86
    Cordoba TC
    AI @108%, Max Aggression
    Default set-ups.
    Default track progression.

    upload_2023-11-22_21-54-27.png

    Fastest AI is just over a second slower (although I'm sure I can go faster and the rest were quite a bit slower). I tried a quick race @116% and the fastest AI were doing similar lap times to me.

    They weren't particularly slow anywhere but I'd say T1, T2, T3 and T4 were where I tended to make gains - they're just a touch slow getting back on full throttle.

    And again, their fastest laps were pretty much always their first flying lap.
     
  19. Brainbug

    Brainbug Inside Virtual Reality AMS2 Club Member

    Joined:
    Mar 2, 2020
    Messages:
    412
    Likes Received:
    660
    i just can second that. Beside the fact it takes time…. It also is no fun.

    on a side note and lemme dream a bit:
    How awesome it would be to have a ai strength difficulty system, driven by ai to identify player strength and adapt automatically from race to race, car to car, and “learn” constantly.
    :po_O
     
    • Dislike Dislike x 1
  20. morpwr

    morpwr Active Member AMS2 Club Member

    Joined:
    Dec 10, 2016
    Messages:
    491
    Likes Received:
    219
    I honestly dont see that with ai times. 10 minute practice 10 minute qualifier and race at 100 works every time. If it doesnt its me not figuring out the car or track yet. Normally not the case after 20 minutes of practice and qualifying though. Now open wheels are a different story because its 100% me.lol
     
    • Dislike Dislike x 1
Thread Status:
Not open for further replies.

Share This Page