Fighter AI Realism vs Difficulty

Collapse

Recommended Videos

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • MacGowan
    Sassy
    • Jun 2017
    • 1681

    #961
    Re: Fighter AI Realism vs Difficulty

    Originally posted by Skynet
    Hmmm, that's a pretty tricky one. CAFs are a very specialized scenario since they can be so incredibly imbalanced. The AI systems are designed mostly for AIs that have been balanced and assigned meaningful behaviors given their stats, etc.

    So yeah...that'd be a definite gap in the AI. Not sure it's one that's easy to address necessarily. It would require a very significant amount of contextual information be added to almost every AI decision system to check for such imbalances. There are systems in place which compare offensive vs defensive stats when AI an has already decided to perform an action, but less so when deciding which action to perform (at the moment in UFC 3).
    Interesting. That makes me wonder about the fictional fighters you meet in career mode. Are their AI's linked up with their stats? If so, can those AI's just be labeled and added to the CAF menu?

    Again, having more information about an AI we wanna pick would help alot.

    Comment

    • Skynet
      EA Sports UFC Developer
      • Mar 2015
      • 703

      #962
      Re: Fighter AI Realism vs Difficulty

      Originally posted by MacGowan
      Interesting. That makes me wonder about the fictional fighters you meet in career mode. Are their AI's linked up with their stats? If so, can those AI's just be labeled and added to the CAF menu?

      Again, having more information about an AI we wanna pick would help alot.
      The fictional fighters do indeed follow certain templates, though I think it's the same ones that are used for fighters already.

      There are 7 templates, which many fighters having slightly customized versions of those. These are all available to the career team, so I don't see why they wouldn't use them. As for how they then assign the movesets and stats...I honestly have no idea! I'd like to assume there's an intelligent mapping and it's not totally random.

      Those are also the ones available to you via the CAF creation, though you don't always know which ones are custom and which are templated.
      Last edited by Skynet; 10-01-2019, 07:15 PM.

      Comment

      • MacGowan
        Sassy
        • Jun 2017
        • 1681

        #963
        Re: Fighter AI Realism vs Difficulty

        Skynet
        You guys had a chance to look at this kinda stuff? I wonder how it would be to have a self-taught AI fight you :O

        Comment

        • Skynet
          EA Sports UFC Developer
          • Mar 2015
          • 703

          #964
          Re: Fighter AI Realism vs Difficulty

          I have indeed seen that paper, among others, and have tinkered around with and explored some ML applications to UFC. There's no all-powerful machine learning AI opponents just yet, but that'd be pretty sweet!

          That said, while the UFC game does in fact make use of ML, it's not always in the ways you'd think. We've been using machine learning in the game since UFC 2. I think the first case of ML was actually to help detect when the foot/leg got stuck after a head kick, and we needed to make the fighter disengage.

          There are tons of applications of a machine learned AI in games, and even if there is never a production version as an opponent, they can be super useful during development. They're great at finding bugs and exploits, for example, but also really good for testing balance during large-scale tests. I'm pretty sure the Battlefield series uses ML agents to test their huge 64 player maps without needing that many players actually playing.

          Aaaannnyway. Yes, it would be really cool to see a good ML AI in UFC one day. Maybe we'd even be able to have a user-trained version, too <3

          Comment

          • MacGowan
            Sassy
            • Jun 2017
            • 1681

            #965
            Re: Fighter AI Realism vs Difficulty

            That's a great point. Using ML to find bugs and exploits. Especially for larger multiplayer games. ML can for sure create meta gameplay the creator didn't invision.

            It's interesting to say the least. I'm coming from a media background, so this stuff is very promising since ML can for example help enlarge lower res footage into HD/4K. But that's neither here nor there.

            Comment

            Working...