2020. 2. 18. 06:03ㆍ카테고리 없음
JediYL 9-Sep-13 14:48 9-Sep-13 14:48 The trick with neural networks is to always be recording the actions of the players and let the AI agents mutate towards the actions of others. Create a neural network sky and fly a plane around it, for no purpose. Then introduce some other planes to fly around you. Then add some guns to your plane.
Then add some guns to their planes. Then go online where players just rip the hell out of duffs like you.
Now then your computer planes will have become so skilled that no one really wants to play your suicidal game! This I term the 'Parental Discretion Method.' Writing a good AI is not about writing a clever AI. Writing a clever nightmare AI is usually easy but this AI wont be a good AI!
A good AI is stupid and its behavior contains a relatively small set of easy to identify patterns but at the same time it must.look. smart (for example soldiers play waving animations and seem to call their friends, the map script makes it clearly visible for you that they are running around to enclose you but you feel satisfied and smart when you play them (these intentionally stupid intelligent-looking bots) out.). Killing a stupid but 'smart looking' AI seems to be a fulfilling challenge for a player.
Ai Techniques For Game Programming Source Code Download
The biggest problem with neural networks is that you cant really predict some behaviors and that isn't good. Neural networks are probably rarely used in today's mainstream games. Fix map scripts and some finite state machines for AIs do the job quite well and in a much more predictable way. You don't want to find out why a neural network doesn't work when you have to ship a working version of the game 2 days later.
Why do you want to mention unity in case of A.? That would rather be an advertisement. Almost every game engine contains a path finder. However I agree that 'Path finding' would be a better name for the article. In case of AI not so many tasks can be extracted into general purpose libraries.
Path finding is one of these, and one of the most difficult in some cases depending on the parameters (type of map, 2D/3D, frequency/complexity of path finding requests, sometimes you have to do it asynchronously to prevent game from stalling.). The AI itself is often part of the game code and its a custom piece of code for every single game. For this reason I don't really like using 'AI library' to refer to a lib that contains path finding, perception system.