Skirmish AI: E323AI 3.22.4 - Page 33

Skirmish AI: E323AI 3.22.4

Here is where ideas can be collected for the skirmish AI in development

Moderators: hoijui, Moderators

slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

Thank you. Nice demonstration. I marked the following new problem also: AI issues ATTACK command even when target is not seen, so it attacks in disorganized manner.
Image

Was green AI (in the left top corner) also E323AI? Its behaviour is weird.
Image

PS. This is because it built factory upon metal spot. Hope Error323's build map development will fix this.
Attachments
screen00062.jpg
(147.57 KiB) Downloaded 2 times
screen00061.jpg
(137.81 KiB) Downloaded 2 times
User avatar
Zydox
Lobby Developer
Posts: 453
Joined: 23 May 2006, 13:54

Re: Skirmish AI: E323AI 3.19.1

Post by Zydox »

slogic wrote:Was green AI (in the left top corner) also E323AI? Its behaviour is weird.

PS. This is because it built factory upon metal spot. Hope Error323's build map development will fix this.
Yeah, all four were E323 AI's... first time I've seen it done something like that though...
User avatar
Error323
AI Developer
Posts: 237
Joined: 28 Nov 2006, 16:46

Re: Skirmish AI: E323AI 3.19.1

Post by Error323 »

It build something else on top of a mex spot. Next it wants to build a mex, the mex is build as close as possible. But its still outside the extraction radius of the map and the spot (since a building is there). So the mex is build and the spot is still not marked as being "taken", this keeps on repeating. This requires a buildmap... working on it...
User avatar
romulous
Posts: 39
Joined: 31 Jan 2010, 12:36

E323AI 3.21.1

Post by romulous »

Anyone played a few bots on Comet Catcher Redux ? They are aggresive at the start. I suspect the stall mid game could be a unit limit issue, since they build a lot of buildings.
SirMaverick
Posts: 834
Joined: 19 May 2009, 21:10

Re: Skirmish AI: E323AI 3.19.1

Post by SirMaverick »

slogic wrote:>> In CA...
At current stage I personally don't care about CA, because it uses plain tech tree which needs another approach in AI planning. Also no config files for CA are deployed with AI.
Config files for CA: http://trac.caspring.org/browser/trunk/AI/E323AI

When I tried to make all factories t1 it always just build one type of vehicles and kbots. Is that a limitation that E323AI doesn't expect e.g. 2 lvl1 vehicles factories?

Splitting them up in lvl1 and lvl2 works fine.
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

SirMaverick wrote:...When I tried to make all factories t1 it always just build one type of vehicles and kbots. Is that a limitation that E323AI doesn't expect e.g. 2 lvl1 vehicles factories?
Yes. Each factory type should have separate tag. There are: KBOT, VEHICLE, AIR, SEA, HOVER (moreover, SEA & AIR tags are overlapped with unit environment tags). So, the only way to force AI to build more than one KBOT factory is to assign them different tech levels (as you did). Also there is hardcoded limit to 3 tech levels.
User avatar
Zydox
Lobby Developer
Posts: 453
Joined: 23 May 2006, 13:54

Re: Skirmish AI: E323AI 3.19.1

Post by Zydox »

Soo... Anything new happening in the E323 land?
User avatar
Idleking
Posts: 28
Joined: 23 Dec 2008, 18:40

Re: Skirmish AI: E323AI 3.19.1

Post by Idleking »

Hmm...
Does anyone else get some utterly high CPU-usage later in-game?

Usually just right before I set my army to move towards the enemyline, the AI starts doing... whatever it does when engaged :P - resulting the CPU-usage going up to the high 90's, which results in a horrible loss of fps making the game, well, unplayable.

Since it's getting better when I nuke the bot's base (which is half of the mag usually), I guess it messes up using a lot of units at the same time.
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

Zydox wrote:Soo... Anything new happening in the E323 land?
Refactoring is going which is very boring, so i can't promise any ETA. AI needs task queues support as human needs oxygen.
User avatar
AF
AI Developer
Posts: 20687
Joined: 14 Sep 2004, 11:32

Re: Skirmish AI: E323AI 3.19.1

Post by AF »

While I would normally be in support since I appear to be the most excessive proponent of task queues in AIs in this community, Im curious what alternatives you've discussed and your justification? It's unusual that one should do this since it has only happened once before in this AI community with OTAI and veylon
jseah
Posts: 42
Joined: 10 Jun 2008, 03:28

Re: Skirmish AI: E323AI 3.19.1

Post by jseah »

My brother looked at the source for slogic's E323AI 3.21.1 and we made a hack that forced the AI to only build air.
Combined with normal 3.21.1, I find it provides a more challenging gameplay due to air raiding.

Idea to make one AI do air start was taken from BAI's land/eco/air division between three AI instances.


Replay attached.
Can see the AI's weakness against long range bombardment, penetrators were not handled well by the AI.
Dragonflies were used by me as a risky strategy and the AI lost it's powerful units to T1 stumpies. (goliath and tremors)
Top right AI was killed near the start by a light gunship rush (and minimal land support). Of course, that means that I was useless later in the game so it's more or less even.
Attachments
20100626_232712_Comet Catcher Redux_0.81.2.sdf
3v3, top left is air start hack
(7.78 MiB) Downloaded 51 times
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

AF wrote:While I would normally be in support since I appear to be the most excessive proponent of task queues in AIs in this community, Im curious what alternatives you've discussed and your justification? It's unusual that one should do this since it has only happened once before in this AI community with OTAI and veylon
I would like not to fulfill this thread with discussion of bonuses or negative factors of task queues. We just need a tool to make AI behaviour more versatile and easy understandable even by AI developers. Original design is: one goal and some quick hacks (enemy scan, resource scan, repair scan) while target is moving to its goal depending on unit group type. I can't issue with ease behaviour like attack then retire to the base, or, _insert_ new order for builder like reclaiming enemy unit (which attacked a builder) while it was constructing some building.

jseah, well, long ago i pushed some changes into master (not tested though) which allow AI players of the same team (team of E323AI instances only) to build different types of the very first factory. If real player (or AI of another type) is playing with AI in ally then AI can't track its partner(s) and wait for him/them to decide which factory to build.
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

Please, try new 3.21.2 version with Spring 0.82.x. I'm interested in crash reports only.

Changelog:

* implemented http://github.com/Error323/E323AI/issues#issue/26
* tasks refactoring (no changes for end-user)
* no more unlimited number of tasks being executed per update
* possible number of tech levels increased from 3 to 5
* fixed improper area threat calculation near map edges
* fixed rare selecting the edge metal spot instead of the closest one (due to the fix above)
* fixed invalid handling with techlevel flags when selecting next factory to build

NOTE: DLL is NOT compatible with any previous version of Spring.

PS. This AI has been included in Spring 0.82.2.
Last edited by slogic on 08 Sep 2010, 02:13, edited 2 times in total.
merijn
Posts: 59
Joined: 17 Feb 2007, 18:43

Re: Skirmish AI: E323AI 3.19.1

Post by merijn »

Played several games, and no crash, so thats good ;)

However, to test the new spring release, i set up an 1v1v1 between KAIK, RAI and E323. (BA7.14)
Usually the outcome is that E323 wins and KAIK goes down first.

This time KAIK went down first again, after being pestered by E323 and RAI. However, after about 1:10h of game time, RAI won the game.
What attracted my attention is that E323 didn't upgrade a single mine to moho....and RAI could keep ahead in economy and eventually took down E323 by attrition.

I was using the E323 included with the new spring release.
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

You tested exactly 3.21.2.

For FFA scenario this AI (actually any AI which can't play porc style game) could loose easily of course (it is a matter of luck). Try team fight 3xE323AI vs 6-8x(KAIK or RAI in any combination) next time. This is more interesting.

Generally AI lacks long term gameplay. Upgrading mexes is the least one of AI can't do. I know this well. There is much to do but i can't divide tasks cause Error is busy with its own stuff and i'm working on this AI from time to time when i have spare time.

I already told i'm bored with refactoring. So, i'm periodically switching to more interesting tasks. I've already included revised CA config files so it should work from the box in the next version (don't expect perfect gameplay, though it kills CA noobs with ease). Made revision of ugly XTA config file. Trying to make it compatible with NOTA (at least with static builders). Made first steps towards naval units (and again so much problems arose during tests).
merijn
Posts: 59
Joined: 17 Feb 2007, 18:43

Re: Skirmish AI: E323AI 3.19.1

Post by merijn »

Yeah, we played with the dll you put up in that post, on the spring RC's several full games.
But the behavior i described in my previous post was with the version included in the new spring release (and the replay screen says its 3.21.2 too).

The only reason i remarked about this E323 loosing the game is because i was suprised, he never lost this set-up before. So i mentioned it in case something related might have been changed.

In no way i mean to criticize you, I am very happy with any work you do on the bots at all, and i know it takes alot of time. It is a great bot and a enjoy playing against it with friends.

Making the bot work with NOTA would be great, because right now i don't know of any bot which is able to work with NOTA and it would be very nice to have one. But i can imagine this being more difficult, because there are alot of different elements in NOTA.

Cheers,
Merijn
User avatar
Zydox
Lobby Developer
Posts: 453
Joined: 23 May 2006, 13:54

Re: Skirmish AI: E323AI 3.19.1

Post by Zydox »

I've tried the last version provided in the installer, and I must say that the AI has become much harder to beat :)
Though, if you look at the attached replay... the commander handling needs some tweaks :)
Attachments
20100821_144134_Tabula-v2_0.82.3.rar
(2.28 MiB) Downloaded 21 times
User avatar
romulous
Posts: 39
Joined: 31 Jan 2010, 12:36

Re: Skirmish AI: E323AI 3.21.2

Post by romulous »

Zydox wrote:I've tried the last version provided in the installer, and I must say that the AI has become much harder to beat :)
Though, if you look at the attached replay... the commander handling needs some tweaks :)
I did not find any difference in the AI on spring 0.82.x. Normally play 2 AIs on maps like Comet Catcher. Still not that hard to beat.
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI 3.19.1

Post by slogic »

Zydox, you were cheating with dragon teeth :P AI too stupid to reclaim them or move around. Anyway, there was a bug in BA config file for Core aircraft plant.

romulous, he played vs 3 AI players, and one of them always starts with aircraft plant. I think this could be a bit harder. With 2AI players they usually start with kbot & vehicle factories.
User avatar
Zydox
Lobby Developer
Posts: 453
Joined: 23 May 2006, 13:54

Re: Skirmish AI: E323AI 3.19.1

Post by Zydox »

slogic wrote:Zydox, you were cheating with dragon teeth :P AI too stupid to reclaim them or move around. Anyway, there was a bug in BA config file for Core aircraft plant.
That's ok... but the fact that all three commanders ended the game by walking into my defence line is what's bothering me :)
Post Reply

Return to “AI”