|
05-11-2018, 03:08 PM | #1 |
Join Date: Mar 2008
Location: Oregon
|
Improve AI somehow?
Right now, the AI intelligence level is set to Katie Perry.
Unconventional, human style responses are hardly ever implemented making the AI very predictable and boring. For example, I take out as many GEVs with missiles as I can ASAP so they don't become immune harassers when my treads are reduced to 2 movement. To a computer, this would seem to be a wasteful use of missiles but 6 turns down the road but to me in 6 turns that there are not a gaggle of immune GEVs widdling me down more. A single GEV at turn 6 can do (6x2x⅓) 4 guaranteed damage. If you leave 4-6 of them running around while you have 2 movement, that's about 16-24 guaranteed damage. Why not have the game upload every game played and have the AI analyze the success of each and help build us better AI in the form of updates from SJG? It would be a very interesting experiment for sure!
__________________
HCobb's OGRE Maker Yet another OGRE designer FJCestero OGRE Maker CP Alpha's new home My email Last edited by Toltrin; 05-11-2018 at 03:15 PM. |
05-30-2018, 10:26 PM | #2 |
Join Date: May 2012
Location: Sedalia, NC
|
Re: Improve AI somehow?
Yes, Please to something to improve the experience against the AI. The AI plays much much weaker than the 1986 version even on just the simpler Ogre map scenarios even though computing power available is probably at least 1000 times what it was in 1986. The approach being used was not designed to give a strong AI. Play out a couple hundred random moves and then use the one that evaluates best, implement one of the manual AI articles where the Ogre generally move towards it objectives but has a random, small but significant chance to backup or make a surprise move. Or use the same approach as the 1986 version which evaluated distance from the current long range target, value of defenders that can be attacked, and the amount that may be suffered in a counter attack for all possible 1 ply moves but look more than one ply ahead. The AI from 1986 is extremely well documented.
The AI games are also dull. Not enough diversity in setup where the AI is defending. You have access to hundreds of games by human players with rankings give the AI a library of starting positions. For standard play pull from lower ranked human players, for more challenging play go with setups used by higher ranked players. |
|
|