Sorry this looks noob since AI is not my specialty, and I ask just by curiosity to expand my general knowledge base. But else I am an experienced (3D) coder. Is anyone here aware of the lastest AI systems used in commercial games ? I knew roughly the state of art 10-15 years ago (KO2,SS) but now it looks far more complex. Specially how the physics blend with tactics and tactics with strategy. K thanks not to point me on the robot cup things. I doubt it''s really state of art compared to the best soccer games. For instance potential sub topics : - are learning processes used in preprod or RT ? - how is space and time modelled ? (I mean discrete steps or some 4D objects) ...

So you''re saying that RoboCup is not good enough?

Don''t be too hasty to judge before you even understand what
has already been done in this area of research.

The sub-topics you listed are pretty well researched there.

If anything, Soccer GAME AI would have to be "worse" than
those found in RoboCup because you don''t want an unbeatable
computer-controlled team against a human player.



Kami no Itte ga ore ni zettai naru!

Hmm don''t take me for a complete noob too. 25 years of game coding and I have also checked RoboCup projects but quiuckly I admit. I don''t jusge RoboCup inabsolute, my issue is about real soccer sims based on humans, not just virtual CPUs+cylinders. Sure there is a lot of AI developped in RoboCup ... and I admit I havent worked on it so I don''t pretend to know anything in depth.

The difficulty in AI is not making unbeatable abstract machines it''s making human models able to rival real humans. A good AI must remain efficient while human abilities and weaknesses are modelled more and more precisely. For instance zero second reactions or exact moves and movement prediction is not what I value in AI. Giving such inputs is cheating the higher level AI algos.

All I am sure is a real soccer sim has to model human biology, kinetics, psychology. A robot is not tired and not emotive. Subtlety in sport certainly comes from these weaknesses in the models. So in fact it''s how this stuff is currently managed I''d like to know a bit more.