Is a more consistent AI possible?

Discussion in 'General Discussion' started by Racefreak1976, May 18, 2015.

  1. Marc Collins

    Marc Collins Registered

    Joined:
    Jan 11, 2012
    Messages:
    3,159
    Likes Received:
    162
    This is all really interesting, but since the first days of rF1, one of the biggest complaints and problems is that AI are not consistent from one track to the next (your original point).

    You can delve for hours into one particular car at one particular track to try to figure out of it's realistic or not, but it will be to no avail. As MarcG already said, the AI don't use the same physics as we do, so even if they are magically consistent with a real-life comparison, it is not 100% correlated to what time you will be able to do in the same car at the same track. There are way too many variables, with no consistency in how they are applied, in the world of AI.

    Hopefully ISI does have something planned to solve the following two major problems:

    1) Few mod tracks have good quality AI because it takes a lot of painstaking work to do it properly. A simpler, one-time AI training routine that could then be translated into a permanent track component would be a godsend.

    2) Some way to retain consistency of a car's speed and skill level from one track to the next. For example, if you can hop onto Mores with the Clio's and match the AI speed at 100% during practice, three things need to happen:

    i) The AI should be the same speed during a race, so that once you have adjusted them to be competitive, the % strength doesn't need to be adjusted
    ii) When you take the same car to another track, like Lime Rock, the same % strength should result in very close to the same level of competitiveness
    iii) This works for all ISI content (it sure does not at the present), so there is some hope that third-party content will also be able to be configured similarly

    Right now, it's a dog's breakfast and many third-party modders don't even bother to try to figure it all out or spend weeks calibrating it...and many players just give-up and go online. We've been waiting for 10 years for a fix to this, so I wouldn't hold my breath that it will all be sorted in the next build or two.
     
  2. P.S.R.

    P.S.R. Registered

    Joined:
    Jun 15, 2014
    Messages:
    1,794
    Likes Received:
    4
    The waters are pretty muddy in here with people mixing different "issues".

    On the "issue" of "a consistent LEVEL OF CHALLENGE for ME across tracks", that is an invalid goal for so many reasons I don't have time to type them all.

    On the "issue" of AI need to "be fast" or "represent a fast driver" or whatever words you choose to use regarding any meaningful standard of performance relative to real life or whatever you can forget about that and there is already general agreement that this is not possible because differences in IRL tracks and grip levels and absence of seat of the pants etc etc so you should give up on that too. So it follows that without a standard then there is no way to do what you are asking for.

    Furthermore, this is almost entirely needless given the immense control you have over the AI and also the fact that you are talking about changing an AI percent. How hard can that be?

    This all smells very arcade and lame and stupid. Furthermore, the software doesn't even support cross-weekend racing out of the box anyway such as consistent drivers so you have to fiddle anyway or you should be asking about consistent drivers across tracks instead but then that will lead to arbitrary selections of those drivers from the pool of drivers so then series by series will be different etc etc.

    The fact is that rF2 kicks ass and there is no way to make certain things easy without simply dumbing them down or worse distorting things are losing flexibility etc etc.

    My graduate school dean always said "organizations are suboptimal" and she was right. Flexibility is king and technical people will always rule.

    Stop trying to make everything easy. There is a reason people play arcade fake sim games and that is because they don't want to think for themselves or learn anything or even decide what they want to drive ffs.

    There is ONE and only ONE issue with AI and that is ****ty AIW from modders who don't finish the job and that is why I don't drive those tracks.
     
  3. peterchen

    peterchen Registered

    Joined:
    Oct 25, 2011
    Messages:
    2,099
    Likes Received:
    287
    Here I agree. Rest is rubbish...
     
  4. Racefreak1976

    Racefreak1976 Registered

    Joined:
    Apr 16, 2012
    Messages:
    1,085
    Likes Received:
    63
    Maybe it isn't such a bad thing after all. If the AI was perfectly optimized there would be no point for multiplayer. So there is a good thing in a bad thing. Or Thesis and Anthesis as Hegel would call it. ;)
     
  5. MarcG

    MarcG Registered

    Joined:
    Oct 4, 2010
    Messages:
    6,854
    Likes Received:
    2,234
    Just on your 1) point: It is is relatively easy to create "good enough" AIW from scratch once you've learnt how, if a track modder has spent 100s of hours creating a track in the first place then the 6-10 hours to do the AIW (from complete learning to finish) is nothing in comparison, it's the fact (as you say and one which I think we all agree!) that they are simply too lazy to do it properly. Once you know how it becomes quicker and the "painstaking work" is still just a fraction of the time that it takes to make the track in the first place.

    I agree there needs to be some sort of consistent strength value across all tracks, how to get this to work though is another story and I just can't see ISI doing anything about it any time soon, although I disagree we've been "waiting 10 years for a fix" - it's ISIs way of doing it therefore a "fix" isn't needed, a Change to the system is what's needed (see last paragraph).

    As P.S.R. somewhat goes into, if you start to dumb down the AI to make it level across all tracks then you'll end up with worse AI than what we've got....and who wants that? I know I don't. Maybe there could be a way that the AI learn the track from within the Corridors somehow, but then I can see Track Makers becoming even more lazy if all they have to do is "press a button and the AI will learn", imagine the mess when half-arsed tracks are released (as they already are and never finished) with AI that can barely make it past Turn 1 even though the AIW creation is "easier" - we're back to square one, who wins then? no one, it's us having to "re-do the AIW" just like we already are.

    The rFactor franchise has always been about tweaking here and fiddling there to get a desired result due to the complexity of the Sim, this isn't a simple GP2 from Crammond style game after all, to get the best you have to go beyond simple otherwise you're releasing a game every 6-8 years that never progresses.
    Point being this is the way ISI have learnt to do AI, it maybe rough around the edges, it may have it's flaws in dealing with certain aspects which may seem "simple" to you or I, but it's their way and still to date in my mind the best damn AI in any game and I wouldn't have it any other way right now. You don't create Great AI like this from the start, it takes time to perfect as is perfectly evidenced by ISI.

    The best, that I believe, we can hope for is that ISI change the code or style of the AI for RF3 (if that ever comes), they sort of Rehash what they've done using what they have learnt to date to create an AI Learning procedure that is more consistent on any given track, no matter what the Track Maker has done with Corridors!
     

Share This Page