Page 3 of 3
Posted: 11 Nov 2005, 13:33
by daraknor
I play well with others in general, but in the bigger AI field people are pretty tight lipped and territorial. So far everything looks clean and people are sharing code and concepts which I like. Alanti has contacted me, I don't know what will happen with that. If I don't partner with someone, I will make a modular system allowing others to contribute bits and pieces.
I plan to be active in some development, although I am not set on going solo or working with a group yet. I would like to work on some of my General AI theories, and this seems like a great place. Rather than a script called AI, it would actually be one.
Later the basic architecture and code may become something different/bigger like evaluating systems models. The danger for real AI comes with Goal Systems, it may simply decide people are happiest in heaven and kill everyone for their benefit. I intend to avoid the scariest part, and write AI without Goal Systems, only task lists.
I don't have anything where it counts at the moment, and that is code.
Posted: 11 Nov 2005, 16:34
by AF
I can see now why you havent released your AI just from seeing weak points in those shots.
You have a working attack system but you are happy with it. As seen by the fact that each AI instance is sending a continuous stream of units to the enemy base, giving them their attack orders as soon as they're built.
Your AI doesnt scale well, it builds small compacted bases as seen in all 3 screenshots.
You want better use of factories. The AI built a kbot and veh factory simply because they where the cheapest, but I may be incorrect on that particular one.
I dont see any scouting going on in those shots. I assume you have that covered though.
Untill the above are done I doubt you'll even consider releasing, however I am confident those are only minor obstacles.
And thankyou for the insight DeathHawk.
Posted: 11 Nov 2005, 16:47
by submarine
lol thats why you shouldn't just interpret images :)
(a) it builds the factories it needs to build the units it wants to build
(b) it sends a continous stream because i didnt implement group management yet (it puts all units in the same group)
(c) scouting does work otherwise it couldnt send its unit to the enemy base
(d) i can adjust how compact it builds its bases by changing two values of a config file
(e) it doesnt expand because i dont want the ai to expand; its much easier to debug things if bases are small and if it uses only a limited number of different units.
i tested with level 2 units as well, it works the same, but theres the drawback that they are slower to build and tehrefore i cant debug as fast as with lev1 units
Posted: 11 Nov 2005, 16:56
by AF
In my experience there are bugs that only occur when the AI scales up. I should know having playtested with NTAI and found such bugs and fixed them.
Posted: 11 Nov 2005, 17:09
by submarine
you can clearly see the scout from the other ai player and that the buildings are little bit more spread (increased values a bit)

Posted: 11 Nov 2005, 20:08
by AF
It also appears to have a grouping mechanism now, as shown by the distinct small line of densely packed units.
Also is your build algorithm a Zaphod style approach as seen in JCAI or an SJ approach as shown in callback->ClosestBuildSite()
Posted: 11 Nov 2005, 20:38
by jcnossen
Why do you have to explain submarine's AI? I think he is perfectly capable of that himself. He also has actual insight instead of the all the guessing you do.
I guess it's just your need to irritate others again...
Posted: 11 Nov 2005, 20:44
by AF
No I get to test my perdiction again.
But really if I annoy you that much then why do you insist on commenting on it? Surely you'd be better off keeping quiet after saying so for the first time.
Posted: 14 Nov 2005, 00:10
by submarine
sorry for beeing absent the last few days (university stuff took more time than expected) i've done a little work on AAI's defending behaviour and had some matches between two AAIs
the screenshot below shows that core now builds a lot more instigators than aks compared to the first matches - although i'm not sure i think thats is caused by AAI's learning system

Posted: 14 Nov 2005, 06:34
by Maelstrom
Even if its not 'learning' as such, its still very cool. Keep it up, and we might even find balance issues from this. If one unit is being used to much, then it is over powered. If a unit is never used, it sucks. When are you going to release a version for testing?