1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

AI skill level linearity

Discussion in 'Automobilista 2 - General Discussion' started by MEDIUMDONGER, Nov 1, 2020.

  1. MEDIUMDONGER

    MEDIUMDONGER New Member

    Joined:
    Nov 1, 2020
    Messages:
    1
    Likes Received:
    0
    Booted up the game after some time off of it to try the new update. Love the cars, hate the AI.
    Is it just me or does anyone else find the linearity of the AI difficulty setting to be completely whack? I did 2x 20 min races around the Nurburgring GP and then Imola with the 911 GT3.
    Around Nurburgring I built a 1s/lap gap to the car in P2 on 113 AI level. I increased it to 115 and got completely demolished, got gapped 1s/lap by 15 out of a total of 25 cars. Lowering it to 114 made no difference, still got gapped by most of the field. Top 10 cars were basically on the exact same pace throughout the entire race. No driver made any mistake what so ever.
    So around this track AI level 114 means I get left behind by 80% of the field, and on 113 I leave the entire field in the dust.

    Around Imola I had to lower it to 111 to even stand a chance of not ending down in the last 20. AI level at 112-113 and once again got left behind. Same ridiculous AI pace here, top 10 cars have identical pace through the entire race, and at the end of the race they cross the finishing line with just 2-3 tenths of a second from P1 to P10. Not a single car made any mistake during the race (other than bumping into me).
    I couldn't really find a level where I actually got to do some real racing on either of these tracks, never had this much trouble in any other game.
    Couldn't find any other thread with this particular issue, hope it's not the 100th time this has been brought up.
     
  2. 250swb

    250swb Active Member

    Joined:
    Apr 1, 2020
    Messages:
    125
    Likes Received:
    42
    It's also what I feel, the AI needs adjusting one percent here, one percent there to get a good race, and then do it all over again at a different track. So the work involved can be considerable, and each time the AI behavior is tweaked it's start again from scratch.

    Sometimes if I want a quick race I just make sure I start from the back and hope to either have a dice with the backmarkers or fight through the pack.
     
  3. Rodger Davies

    Rodger Davies Active Member

    Joined:
    Sep 16, 2016
    Messages:
    164
    Likes Received:
    153
    I thought this was me just being inconsistent at circuits!
     
  4. azaris

    azaris Active Member

    Joined:
    Aug 30, 2020
    Messages:
    147
    Likes Received:
    59
    One issue with AMS2 is that, apart from the bottom 2-3 cars, almost all AI cars lap at the same pace. I think the AI cars are supposed to have different personalities and driving styles, but the effect is usually that the top 15 cars are within a few seconds of each other for the entirety of the race. This means the player tends to be either in clear P1 or somewhere in the back unless the AI strength was chosen "just right".

    Other sims like R3E have a wider disparity between the fastest and slowest AI (sometimes multiple seconds/lap), which creates more strung-out fields but also helps the player to always have someone to race even if the top AIs are too fast for them.
     
    • Agree Agree x 1
  5. Pacchia

    Pacchia Member

    Joined:
    Mar 21, 2020
    Messages:
    39
    Likes Received:
    8
    Are you tweaking the setup?

    What I find in those cases is that maybe the default setup is fine for some tracks (and Im confirtably in the top 5 in qualifying), whereas the standard settings are awful for other circuits. In that case, some changes take me mid-pack.

    Something I have observed: the AI of brazilian series (stock cars, sprint race etc.) seems to be spot on for the brazilian tracks, but a bit too fast in other, european or asian circuits.

    I like the fact that the field is running similar times to be honest...makes for some exciting racing.
     
  6. Rodger Davies

    Rodger Davies Active Member

    Joined:
    Sep 16, 2016
    Messages:
    164
    Likes Received:
    153
    To me this comes down to the same ultimate point as this thread: Any way of renaming AI Drivers & Teams?

    Namely, the Madness Engine default doesn't link a driver to a car (presumably the licenced stock car series have the correct driver names for each car, right?) - I'm not sure if it has the ability to, but I'm pretty certain there isn't the same depth as there was with gMotor / AMS1 RCD files to curate drivers to have varying aggression, consistency, ability.

    Might sound daft and it might just be me, but this is my main 'feature wish' for AMS at the moment; being able to edit the AI grids/abilities in the same way as with AMS1.
     
    • Agree Agree x 2
  7. alink

    alink Well-Known Member

    Joined:
    Dec 3, 2017
    Messages:
    791
    Likes Received:
    456
    Hi,
    I also noticed that something is strange with AI.
    Originally I had AI strength at 95% and aggression at 100%.
    Since v1.0.4. I use 98%, now with v1.0.5 100%.
    But at Nurburgring with GT3 and GT4 carset, in practice I was 5s faster, in qualy 6s and in race about 5-6s faster than AI.
    Similar at other tracks.
    But it's a little bit annoying to invest hours only to find out which level is needed for AI.
    The second, I don't want to fit the AI to myself, if AI is faster than it's a challenge, if I'm faster than I have an unstressed race. But for this, I'd like to use the same strengh, independent on track or carset or series I use.
     
    • Agree Agree x 1
  8. Silvano

    Silvano Active Member Reiza Backer

    Joined:
    Mar 15, 2016
    Messages:
    263
    Likes Received:
    59
    IMO there are little chances to really have "good" AI, unless the devs find a way to give them algorythm that make them "neural", i mean to make them "learn lines, curve speed and overtaking by learning while on track. Not only that but they would learn also by learning from the user hability, witch consequently would make them on par with the user and being competitive and so giving the user a greater immersion and fun to drive.
     
  9. alink

    alink Well-Known Member

    Joined:
    Dec 3, 2017
    Messages:
    791
    Likes Received:
    456
    Papyrus with NASCAR2003 Racing Series had a nice feature:
    Dynamic AI
    AI was fitted to speed of the driver. So you had been always in mid-field.
    But there was one negative fact: you never could win a race!
    Because, if you went faster, the AI also went faster. If you got slower then AI, too.

    It shouldn't be a problem to have a good and well performed AI. In earler days of SIM racing it was possible (IndyCar Racing 1/2, Papyrus NASCAR, GPL, F1 Challenge, GTR1 / 2). Not always the behavior but speed was good and 100% performed more or less to real lap times.
    I think, the main difficult is to get all these different car classes, car performances, this wide range from Kart over many Touring and GT cars to Prototypes and Formula cars under one hood. The AI has to be specified for each car that may not be possible with one program logic. Maybe Reiza must find a way to use different AI logics depending on the car class.
     
  10. Silvano

    Silvano Active Member Reiza Backer

    Joined:
    Mar 15, 2016
    Messages:
    263
    Likes Received:
    59
    Why Reiza would lose time to give different AI logic to every single cars ?, that will never work good, but instead let the user do some laps with a given car on a given track, then take his best lap as a exemple to follow for the AI. So there will not be lot of diff with the speed in curves or straight with the user. Its important to let the user do his best lap as users drive all differently, so what i am talking about is a app integrated to the sim (or outside) that would spare a lot of work on AI, I am sure this is the way to go.
    This way i think, every driver will find the best competivity possible, of course the bad side is that the user will have to do that for every combo but he will do that just for the combos he like i guess (i would do it this way) and when the user gets really much better at a track, he just have to do his 3 laps and feed the app, good thing is we will not have to use a "strengh" slider for every track.
    At the moment the AI are totaly predicable, as a exemple," in turn 3 they are so slow i can passe 2 or 3 cars" and so on, this is really frustrating and kills the fun.
     
    Last edited: Nov 5, 2020
  11. Silvano

    Silvano Active Member Reiza Backer

    Joined:
    Mar 15, 2016
    Messages:
    263
    Likes Received:
    59
    This was a genius idea, true it worked good and at that time that was about 20 years ago !
    but today the devs (for unknowed reasons) dont even do pitstop animations, this is something i cant understand really. As for AI, Reiza and sector3 are doing a great job, very good AI so far. Of course there is still a long road ahead but they allways try to finetune and eventually they will get the AI to a even better level but nothing is better than "neural" AI, i hope that will be the next step.
     
  12. azaris

    azaris Active Member

    Joined:
    Aug 30, 2020
    Messages:
    147
    Likes Received:
    59
    It didn't actually work that well, and that is probably why few sims since then tried something similar.

    For those not in the know, NR2k3 had two "dynamic AI" options. "Auto strength" and "adaptive speed control.

    "Auto strength" would recalibrate the AI speed after every race to try to match the player's pace. It worked roughly as well as it does in RaceRoom, in that the speeds initially fluctuate all over the place but after like half a dozen races it finds some level that may or may not be what you want.

    "Adaptive speed control" was changing the AI pace during a race to match the player, i.e. rubber banding. When your tyres went off and your lap times started to fall, the AI would slow down. When you came out with freshies, they would speed up. Now I don't know about you, but I've never heard anyone say they want a rubber-banding AI in their racing sim.

    In the end, most people just stuck the AI on some fixed rating and left it at that.

    AMS2 has animated pit crews (not as detailed as NR2k3 but still).
     
  13. Silvano

    Silvano Active Member Reiza Backer

    Joined:
    Mar 15, 2016
    Messages:
    263
    Likes Received:
    59
    Really very informative infos right there, its a long time since nascar 2003 and such, my memory has leaks, so i cant remember how good was the AI, i just remember at Talladega, they pushed me to overtake, that was a nightmare to see a car in the rearview and trying to stay safe :)
    and about AMS2, i didnt know of a pit crew anim, thats good news, it was about time this would be done again.
    "auto strength" is a good option, looks like "adaptive AI" i allways use it in R3E.
    i dont think "rubberband" is a good way to go, but alone the fact that they where on par with the player is a very good point, i think this is missing in today sims.
     
    Last edited: Nov 7, 2020
  14. David Peres

    David Peres Member

    Joined:
    Nov 22, 2018
    Messages:
    55
    Likes Received:
    49
    Raceroom does the same thing with their "adaptative AI" system. It basically takes the average time of your last 10 sessions (excluding values that deviate too much from your norm) with a certain class/track combo and adjusts the AI speed around that average... more or less (it's a bit more complex than I make it sound).

    In Raceroom this does not change in real time during a session so in theory it's still possible for you to win but you need to go above your usual average to do it.

    However, in order for this to work well you need to spend quite some time "training" the AI by making a series of races, while making sure your pacing is consistent, so in the end I don't really bother and just choose the AI level by hand.
     
    Last edited: Nov 7, 2020
    • Agree Agree x 1
  15. Silvano

    Silvano Active Member Reiza Backer

    Joined:
    Mar 15, 2016
    Messages:
    263
    Likes Received:
    59
    Exactely, and its important to mention the races dont have to be more than 5 laps, i just did change the AI adaptation by making a 2 races event of 5 laps each and at the 2nd race the AI was allready adapting, not very good but if you do 2 or 3 times those mini races, the adaptation will be nearly perfect.
     
  16. Luciano Santos Filho

    Luciano Santos Filho Member

    Joined:
    Sep 7, 2017
    Messages:
    30
    Likes Received:
    7
    Another problem is the inconsistency of AI between qualy and race. Just tried to race the Classic Formulas, GEN 2, at modern Interlagos. I was driving model 1. AI at 102%. In the qualy I was 0.2s faster than P2. In the race, guys in model 3 were 1s/lap faster...
     
  17. Kolysion

    Kolysion BANNED BANNED

    Joined:
    Jan 20, 2019
    Messages:
    163
    Likes Received:
    41
    • Informative Informative x 1
  18. alink

    alink Well-Known Member

    Joined:
    Dec 3, 2017
    Messages:
    791
    Likes Received:
    456
    Adaptive speed control was a good feature always to be in mid-field, training for Oval-racing. You never could win, you never could loose. For normal racing it was not useable.
    In NR2K3 there was the best and most complex AI I ever had.
    AI was drafting, they used slipstream, if you switched the line, the AI followed(!!)
    Also the physics was nearly peferct. Not only slipstream made you faster, also when cars followed you, the rear of your car (or the AIs car if you followed one) had less drag so you reached higher speed but with less grip. I loved the races at Talladega and Daytona, it was tactical to use the other cars to cross the finish line in 1st position (3 laps before end of race, let you drop back to start the chase for the lead, otherwise you lost).
    And weather simulation: hot weather, higher top speed with less grip, cold weather less top speed with more grip. The same with wind: in Charlotte, Atlanta, Michigan with tailwind it was terrible to turn in, with upwind cornerning with full throttle
    Animated flags, when watching these you knew wind direction and speed -- all this nearly 20 years ago! I wish that REIZA can implement all / many of these features for a great simulation
     
  19. Pacchia

    Pacchia Member

    Joined:
    Mar 21, 2020
    Messages:
    39
    Likes Received:
    8
    What do you think about Copa Montana ai?

    Either I am really good with that car, or the AI there is weaker than in other series.
    At 100% I am usually mid-pack with other cars, but with Copa Montana I am first with 1 second advantage
     
  20. Wedsley Dias

    Wedsley Dias Active Member

    Joined:
    Mar 10, 2016
    Messages:
    120
    Likes Received:
    75
    AI is not the best and in some cases, it completely ruins the offline experience. As an example, yesterday at Brands Hatch with FV12 (30 cars) the difference between the first and the last placed was no more than 0.5s.

    Btw: In AMS1 you can edit talent files and make pilots faster / slower, more aggressive / permissive and more. This allows for a more exciting running experience.
     
    • Agree Agree x 1

Share This Page