1. Do you need support for Assetto Corsa Competizione? Please use the proper forum below and ALWAYS zip and attach the WHOLE "Logs" folder in your c:\users\*youruser*\AppData\Local\AC2\Saved. The "AppData" folder is hidden by default, check "Hidden items" in your Windows view properties. If you report a crash, ALWAYS zip and attach the WHOLE "Crashes" folder in the same directory. Do not post "I have the same issue" in an existing thread with a game crash, always open your own thread. Do not PM developers and staff members for personal troubleshooting and support.
  2. As part of our continuous maintenance and improvements to Assetto Corsa Competizione we will be releasing small updates on a regular basis during the esports season which might not go through the usual announcement process detailing the changes until a later version update where these changes will be listed retrospectively.
  3. If ACC doesn't start with an error or the executable is missing, please add your entire Steam directory to the exceptions in your antivirus software, run a Steam integrity check or reinstall the game altogether. Make sure you add the User/Documents/Assetto Corsa Competizione folder to your antivirus/Defender exceptions and exclude it from any file sharing app (GDrive, OneDrive or Dropbox)! The Corsair iCue software is also known to conflict with Input Device initialization, if the game does not start up and you have such devices, please try disabling the iCue software and try again. [file:unknown] [line: 95] secure crt: invalid error is a sign of antivirus interference, while [Pak chunk signing mismatch on chunk] indicates a corrupted installation that requires game file verification.
  4. When reporting an issue with saved games, please always zip and attach your entire User/Documents/Assetto Corsa Competizione/Savegame folder, along with the logs and the crash folder (when reporting related to a crash).

Personal AI

Discussion in 'ACC General Discussions' started by Typer, Mar 17, 2018.

  1. Typer

    Typer Simracer

    I’ve had this somewhat out-there idea percolating in my brain for the past few days that just refuses to vacate my noggin, so here it is. It is based on a pretty massive “if” but were that ‘minor implementation detail’ overcome, then lots of interesting use cases fall out of it. That said, please tell me why my idea is stupid and would never work…


    So, what if it were possible for every player to have their own personal AI that is continuously trained to mimic their racing lines, pace and behaviour as they play the game?

    Now I don’t know how feasible this is from a technical standpoint, how much sample data would be needed or how close to a real player’s behaviour an AI could ever actually get, but if it were possible, I can think of a few interesting new possibilities such tech could open up:

    Race against a whole field of yourself: What better way to improve your skills than to take on 20-odd other yous. See the mistakes your AI makes from close quarters and learn how to capitalise on them. Careful though, as it won’t be long before your AI learns from its master.

    Share your AI with the world: Opt into having your AI synced to the cloud to allow friends to race against “you” even when you’re not around.

    Practice against the AIs of your friends, enemies or even your favourite sim racing celebs, so that next time they’re online you’ll will be ready and waiting to do battle with the real them.

    Can’t find any decently full servers that you wanna race on? Just choose to have one topped up with AIs with similar performance rankings as the real drivers.

    Joining a server filled with AI? No problem, the next AI passing the pit will be told to box. Watch as the perfectly motion-captured AI driver is helped from the car, left momentarily staring Stig-like into the middle distance, before being ushered into a cupboard at the back of the garage.

    Can’t make your scheduled league race this evening thanks to your client from hell making yet more ridiculous 6pm demands? Fear not, just send your AI in your place! (Please note that league admins reserve the right to apply a small performance penalty on AIs.)

    Smoother racing against players with poor connections: When racing online, other players’ AIs will be downloaded to your machine. The moment a nearby player’s connection starts to glitch, their AI will seamlessly step in to replace the delayed or missing packets. No more will you have to deal with opponents inexplicably spearing off into the gravel mid-corner only to suddenly appear right back in front of you moments later, scaring the sweet bejesus out of you.

    Race against the pros: Coming soon, DLC packs of AIs trained by real racing drivers! How will you stack up against the best of the best?


    Anyhoo, sorry for the marketing spiel, I might have got a bit carried away there.
     
    vilivili and silvanob2 like this.

  2. Similar Threads
    Forum Title Date
    Chit Chat Room Your personal Driving Coach Oct 6, 2023
    ACC Tools Your personal Driving Coach Oct 6, 2023
    ACC Gameplay Personal lap time displayed incorrectly Sep 18, 2022
    ACC User Interface option to have a personal saved setup for a specific car Jan 6, 2021
    ACC Rating & Leaderboards Personal stats completely wrong. Any way of reseting them? Nov 21, 2020
    ACC Rating & Leaderboards Special Event, not updating personal time Nov 21, 2020
    ACC Rating & Leaderboards Imola Personal Laps not recording Nov 20, 2020
    ACC PS4/PS5/XB1/XBX/S General Discussions Personal best laps Aug 25, 2020
    ACC PS4/PS5/XB1/XBX/S General Discussions Personal Best Laptime not saving/shown on any track Jul 22, 2020
    ACC PS4/PS5/XB1/XBX/S General Discussions personal view Jul 10, 2020
    ACC PS4/PS5/XB1/XBX/S General Discussions Has anyone noticed your best personal lap times are not shown each series Jul 10, 2020
    ACC PS4/PS5/XB1/XBX/S General Discussions Personal feedback Jun 24, 2020
    ACC Rating & Leaderboards Personal Best Lap (Circuit Screen) Not Updating Jun 9, 2020
    ACC Multiplayer Can't join own personal server due to TR, despite having 99 TR May 19, 2020
    ACC Rating & Leaderboards Fix personal best lap time May 13, 2020

  3. WallyM

    WallyM Alien

    I think it's a terrific idea.
     
  4. whereSTheFUhd

    whereSTheFUhd Rookie

    A 'somewhat out-there idea'?

    Not sure if serious. Surely, you're not trying to pass this off as your own novel idea. This has been implemented and iterated on for at least the last 12 years, from the very beginning of the Forza series. You're basically describing Forza's 'Drivatar' AI. To a T.
     
    stefanoCasillo likes this.
  5. Typer

    Typer Simracer

    Oh. LOL.

    I’ve never played Forza, so yeah that was all my own idea, just 12 years too late by the sound of it :)

    Oh well, back to the drawing board.
     
  6. Serge M

    Serge M Alien

    If a system like that actually works properly you will quickly reach a point where you can never beat the AI as they just keep learning to get faster... and then evolve and take over the world :)
     
  7. Typer

    Typer Simracer

    Ah, you’re clearly forgetting Asimov's fourth law:

    A robot must, like all good employees, let its boss win.

    Actually what I had in mind was was not for the AI to have the goal of getting faster, but instead to just more accurately mimic their trainer. Whether that would mean they would inevitably end up winning against their trainer all the time, I don’t know.
     
  8. whereSTheFUhd

    whereSTheFUhd Rookie

    No need for a drawing board. Idea's are essentially a dime a dozen - recycled, rehashed, and regurgitated over and over again (and, unless you're just recently getting into driving sims, it's probable you've received some exposure to what 'Drivatars' are about). It's the implementation and execution of it that requires the drawing board, which is why it was specifically mentioned above that the 'Bayesian' AI has been implemented and iterated on for at least the last 12 years, not just as an idea on paper, since you can be sure that they weren't the only[first] ones to have ruminations on the idea either.
     
  9. seco46

    seco46 Hardcore Simmer

    The Drivatars of Forza is only normal AI (worse than Assetto 1 imo and with diferents physics of the player) with names of Xbox users (including for example friends without the Game :p)

    So, It's pure marketing :D
     
    Poguinhas and martcerv like this.
  10. Rudski

    Rudski Alien

    A neat feature I would like to see with regards to AI would be mimicking real race drivers lines and habits.
    They all don't take the same line and some do things a little different here and there and that might add a bit more to the sim.
     
  11. whereSTheFUhd

    whereSTheFUhd Rookie

    Yes, of course, the 'Drivatars' are simply normal AI (and a large part of it is very much marketing), however, it is overlayed upon a Bayesian model.

    The effectiveness of the AI can be debated ad nauseam - I have no desire to do so. Whether they are better than Game / Simulator X or Y has no bearing on the fact that they are still an updated learning model (unless you have direct evidence that Greenawalt / Turn10 are blatantly lying, though, if history is any evidence, I certainly wouldn't put it past them).

    So long as an Xbox account holder has had any time in any of the games in the Forza series with 'Drivatars' in them (both Motorsport and Horizon - even a demo), their 'Drivatars' will carry over and appear in any and all other games that utilize 'Drivatars', regardless of whether they own that particular title or not.
     
  12. seco46

    seco46 Hardcore Simmer

    I have friends with Drivatars on Forza Horizon 3 & Motorsport 6 Apex and they never play those Forza (even the Demo) :confused:

    So I know what I'm talking about when I say it's pure marketing :rolleyes:
     
  13. whereSTheFUhd

    whereSTheFUhd Rookie

    Yeah, you're going to have to clarify that statement (though, in reality, it doesn't really matter). Are you saying that they've never played anything in the Forza series - at all - or that they've never played FH3 and FM6 Apex?

    If they've never touched a Forza series game, although it'd seem an odd choice for Turn10 to do so (so, not really out of the ordinary for them), it would still be possible that they simply source players in your Xbox Live circle and feed them into your AI list, providing you with a field of AI that are populated with your friends, if only in name, which would then be updated to more closely resemble their general driving behaviors, if and when they do ever play the game.

    To reiterate, it is essentially just a 'normal' AI overlayed on a Bayesian model - the AI gets updated with certain behaviors according to newly presented information. So, though, as already recognized, largely a marketing bullet point, it doesn't exclude the presence of an underlying Bayesian model.
     
  14. seco46

    seco46 Hardcore Simmer

    They have not played any Forza game.

    Even so I have seen typical AI failures (like the one you see below) even in Drivatars of people who have played the same title than me.



    I'm sorry but I still think this Drivatar thing it's pure marketing :D (and well, it's one thing that I'd love to see in ACC too on custom races with custom liverys for example :p)
     
  15. whereSTheFUhd

    whereSTheFUhd Rookie

    I'm not sure what I'm supposed to take away from your continued efforts to highlight the inadequacy of Forza's AI. That does not preclude the existence of an underlying Bayesian framework. Not sure how it can be stated anymore clearly.

    It's like trying to conclude that AI is not actually implemented in Forza at all, because the behavior exhibited is not what you would deem intelligent - "Forza doesn't actually have any AI, because I mean look at them, obvs." What?. Bad AI is not no AI. Likewise, bad AI doesn't mean there is no Bayesian network in it's implementation. Your argument that it is simply normal AI (without any Bayesian inferencing), isn't upheld by what you're presenting as evidence.

    In fact, if people were exhibiting such wonky behavior, then the model should, in theory, incorporate such behavior's into its repertoire (which would, of course, then have to be manually excluded).

    And, again, I'm not arguing the competence of Forza AI. In fact, I could care less about whether Forza actually even implements any sort of Bayesian model, however, it doesn't change the fact that there is an apparent gap in the logic of your argument.
     
    stefanoCasillo likes this.
  16. seco46

    seco46 Hardcore Simmer

    What? I only say that Drivatar thing Is pure marketing because the AIs only use the name from Xbox Live users.

    And that is the truth if you compares the Forza AI with other games, comparing his physics with the physics of the player or even if you seen that typical AI failures of other games with the Drivatars.

    I not doubt about the tecnology but "Skynet" is not with us for the moment buddy.
     
  17. whereSTheFUhd

    whereSTheFUhd Rookie

    No, please go back and reread your statements. In response to the notion that Forza utilizes a Bayesian model for learning AI, you stated that Forza's AI was "only normal AI," thereby essentially refuting that Forza uses any Bayesian framework at all.

    You then went on to uphold such an assertion with points that don't really have any relevance to the statement. To recap. Again. Whether Forza AI is worse than Game / Sim X or Y has no bearing on the use of a Bayesian model. Whether Forza AI utilizes the same physics model as the player has no bearing on the use of a Bayesian model. Likewise, though it'd be an odd choice, sourcing names for the AI from player accounts that have supposedly never played any Forza series title does not preclude the use of a Bayesian model (although I'll take such a claim at face value, given the gaps in your logic, which makes your position seem rooted in an emotional stance rather than one of reason, not sure on the veritability of this statement, even though, as already stated, it, ultimately, makes little difference to your initial statement). Lastly, Forza displaying 'typical AI failures' that borders on abysmal head-banging levels of idiocy does nothing to demonstrate the lack of a Bayesian model.

    So, no, the crux of your argument was not that it was pure marketing. That was a conclusion you arrived at based on your statement that the Forza AI was "only normal AI" that did not utilize a Bayesian model, which itself was based on arguments that bore no proof or relevance to your position.

    You're either backpedaling or, given some of your language, perhaps there might be some language barrier. However, I think that's being charitable, since even with a language barrier, most translational vagaries are simply awkward forms of speech, whereas what you're consistently displaying is clear logical gaps in reasoning, which can't really be consistently pinned on a language barrier. Case in point, you're still seemingly arguing the same points that have nothing to do with your central argument:

    Again, where have I made any suggestions as to the effectiveness, or lack thereof, of Forza's AI, such that anyone would even think that I would equate it, even if only in analogy, to something even remotely like Skynet? In fact, I specifically stated that I have no inclination to wade into such a discussion, particularly since it has no relevance to the matter at hand. If anything, my language would not suggest a budding enthusiasm for Turn10s choices. Buddy.

    As already stated above, it's the implementation and execution that requires the drawing board - it takes effort to implement an idea and multiple iterations to execute it successfully, if at all. Whether it was an effective implementation or ever reached optimal execution, however, does not have any bearing on whether it was ever implemented at all.

    Yet you continue to prod that stick, making it appear that your vested interest in this discussion is mainly for taking jabs at Forza and all your arguments work backwards from that point, rather than being based on a progression of logic.
     
  18. seco46

    seco46 Hardcore Simmer

  19. nobody in the thread was claiming it was tho.
    You seem to completely miss the point of the discussion and obviously not really understand the subject...so I really don't see what you are trying to prove here.
     
    chksix and Typer like this.
  20. seco46

    seco46 Hardcore Simmer

    Nobody in this thread but Microsoft yes.

    That's is my point about the pure marketing (because they use the Drivatars as if they had the best AI as marketing and for me is not the best)

    Maybe it's still too early to see an AI behaving like a human being learning from its movements (we will have to wait)
     
    SlimCharles likes this.
  21. well not really, they were pretty much describing what it is: AI that was sharing lines borrowed from players. If you understand how AI in racing games actually works you'll also understand that's like the easiest problem to solve in AI.

    Of course marketing will pimp their product as the best.. that's what marketing is supposed to do.
    I've watched some talks by the guys actually working on it and they were pretty down to earth about it.
     
    SlimCharles and seco46 like this.

Share This Page

  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice