> For simple games, you could probably deduce the game state by looking at what's going > on in RAM, but it's not going to be easy for anything more complicated than Pac Man. > What you would really have to do is reverse engineer (disassemble, decompile) the > entire game code and understand it.
What about reading video memory? Don't tilemap-based games use relatively simple layouts? Would it be possible to deduce from the value at each spot in video memory, for tilemap-based-games, what tile is being displayed there?
> In other words, it'll be very game-specific and a colossal waste of time.
I expected it to be very game-specific, that goes without saying.
> Attempting to build a computer vision-based AI would be even more monumentally > challenging, unless you wrote it specifically for particular games. In that case, it > might not be too difficult, as far as computer vision problems are concerned. If you > know the sprites or shapes you're looking for, it would be really easy to detect and > track their positions on the screen. > > IMHO, the effort put into such a project, and the prerequisite knowledge required, > would be better spent on more productive (and profitable) applications of computer > vision and AI. What's the best case here? An AI version of the jerk who always played > as Chun Li and never let anyone have their own shot at Street Fighter 2?
I think it would be fun. If you do not (and it sounds like you do not) then you are welcome not to participate. Thank you for the information you have provided, though; it is useful.
|