Skirmish AI: E323AI 3.22.4 - Page 19

Skirmish AI: E323AI 3.22.4

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

Moderators: hoijui, Moderators

User avatar
Error323
AI Developer
Posts: 237
Joined: 28 Nov 2006, 16:46

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by Error323 »

Exactly, when the AI would be able to annihilate every single player we can discuss non-cheating ;). Aside from that, its goal is now to "train/challenge" human players as it resembles an agressive human strategy (or tries to anyway).
User avatar
FaerieWithBoots
Posts: 149
Joined: 17 Jun 2009, 13:21

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by FaerieWithBoots »

Error323 wrote: ...
Aside from that, its goal is now to "train/challenge" human players as it resembles an agressive human strategy
...
Make it spam flash then :lol: Just remove all tags from all units :P
elias79
Posts: 57
Joined: 11 May 2008, 23:29

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by elias79 »

I understand but cant they atleast send out scouts first and have shorter LOS hacks
User avatar
hoijui
Former Engine Dev
Posts: 4344
Joined: 22 Sep 2007, 09:51

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by hoijui »

i also do not liek this a lot, as i like to see AI as a research thing, and there the goal has to be get as good as possible without cheating, cause eg. applying a good AI in a nother field, whihc is more .. real somehow (and i am not talking about using rts AIs for real life war or something), then cheating/perfect LoS may not be anything that exists, and such a concept could no be applied.. but if the goal of the AI is more.. tied to the game, and a virtual environment, and makign the game interesting for players.. then using LOS hacks seems to benefit the AI.
related to that:
i remember reading something about clocked units beeing ignored...
does that mean, if i build a cloakcing tower, then the AI will not see the buildings/units in the towers range? i would like that, as it would make cloacking viable, like it is agasint humans.
User avatar
bartvbl
Posts: 346
Joined: 21 Mar 2009, 15:55

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by bartvbl »

http://pastebin.com/d69d81f07
crash when playing versus this AI.
It happened when I dgunned several units at the same time
User avatar
hoijui
Former Engine Dev
Posts: 4344
Joined: 22 Sep 2007, 09:51

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by hoijui »

hmm.. also when the commander dies, many of its units die at the same time usually. could it be that the units have some connection to each other, and if they get killed in the same frame, some connection to an invalid unit remains or so?
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by slogic »

Crash right before exit.
Attachments
infolog.txt
(27.44 KiB) Downloaded 30 times
User avatar
hoijui
Former Engine Dev
Posts: 4344
Joined: 22 Sep 2007, 09:51

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by hoijui »

just to make it clear again.. Error can not use infologs for anything usefull (practical problems), so if there is a crash/bug, we have to help him reprodce it, or get such a good idea of what it could be, that he knows which part of the code to have a look at. That is not optimal of course.
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by slogic »

I specially asked him about value of infolog. He told they are useful anyway. ;)

Ok, i see it's better to finally learn how to compile AI dlls locally.

Why AI does not use Advanced Solar Collectors? Aren't they are cost effective? Then it is not a true BA, you should told them with proof :)
User avatar
Error323
AI Developer
Posts: 237
Joined: 28 Nov 2006, 16:46

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by Error323 »

slogic wrote:I specially asked him about value of infolog. He told they are useful anyway. ;)

Ok, i see it's better to finally learn how to compile AI dlls locally.

Why AI does not use Advanced Solar Collectors? Aren't they are cost effective? Then it is not a true BA, you should told them with proof :)
Where exactly? Cuz its not very usefull, I was referring to the text you wrote with it :)
slogic
AI Developer
Posts: 626
Joined: 17 Mar 2008, 19:03

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by slogic »

You mean environment i used to test your AI was useful & information when it crashes, not the infolog? LOL then.
User avatar
bartvbl
Posts: 346
Joined: 21 Mar 2009, 15:55

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by bartvbl »

a small note:
the commander doesn't look where it is going to dgun. It will dgun through factories if needed.
User avatar
Error323
AI Developer
Posts: 237
Joined: 28 Nov 2006, 16:46

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by Error323 »

bartvbl wrote:a small note:
the commander doesn't look where it is going to dgun. It will dgun through factories if needed.
Epic, my code is self learning! Heh no I think u've been watching the wrong A.I.
User avatar
bartvbl
Posts: 346
Joined: 21 Mar 2009, 15:55

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by bartvbl »

cool!
Are you actually using a neural network, or more something like case-basd reasoning?
Achilla
Posts: 79
Joined: 24 Aug 2009, 15:17

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by Achilla »

I noticed the AI would often crash in latest release in any 'a bit more complicated' land maps. Sometimes after a minute, sometimes after 15 minutes. I guess infologs are pretty much useless, so I will merely generalise when it happens most often:

- AI construction units/main base is near edge of a mountain ... you don't get any information about particular unit trying to take a route but unable to reach the place, but you can clearly see some 'undecided' or 'random' behaviour following just before the crash.

- AI offensive units, which would get stuck on the slope of the hill ... technically they should be able to go from top to the bottom of it, but they will stay idle. I had one situation where AI on top would beat AI on bottom, because AI on bottom had many units stuck on hill slope which clearly were told to do something, but refused to move (sometimes they moved half a screen and stopped again). Martian Desert map is particularly known for it, when you assign one AI at the top-left and one at the bottom-right corner. 1v1 match.

There is also 'TMA 0' map which will crash upon AI initialising, no idea why? (because it gets wild when trying to create threat map? map is a bit 'original'). With other AIs it won't crash on the initialising (AAI, KAIK, RAI).
User avatar
bartvbl
Posts: 346
Joined: 21 Mar 2009, 15:55

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by bartvbl »

A crash versing E323AI. It was a quite random one, really. I just know that the commander had blown up a short while before the crash.
It blew quite a few buildings when it blew up, so I think the cause lies with the crash I described earlier: when I dgunned several units together, and spring crashed.
Translated stacktrace (hope it helps you :-)):
http://buildbot.eat-peet.net/spring/sta ... 71998.html
User avatar
hoijui
Former Engine Dev
Posts: 4344
Joined: 22 Sep 2007, 09:51

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by hoijui »

even doh you had a crash in E323AI (visible in the infolog), buildbot took the AAI debug symbols to translate the stack trace, most likely cause it does not have the E323AI ones, or because of a long known bug, which sometimes makes stack trase translation fail for AIs.
anyway, the statement that infologs are useless for Error323 remains intact.
User avatar
bibim
Lobby Developer
Posts: 959
Joined: 06 Dec 2007, 11:12

Re: Skirmish AI: E323AI v3.14.2 - High Templar

Post by bibim »

hoijui wrote:even doh you had a crash in E323AI (visible in the infolog), buildbot took the AAI debug symbols to translate the stack trace, most likely cause it does not have the E323AI ones
Actually BuildServ had the debug symbols, but it didn't know it had it ;) (still semi-hard coded list of targets which has to be updated on each AI version change because the version number is in the path :/ )

I compressed the debug symbol manually and added it in the target list, so BuildServ can now translate E323 stacktrace of current release:
http://buildbot.eat-peet.net/spring/sta ... 91026.html
User avatar
Error323
AI Developer
Posts: 237
Joined: 28 Nov 2006, 16:46

Re: Skirmish AI: E323AI v3.14.3 - High Templar

Post by Error323 »

E323AI v3.14.3 - High Templar
Changelog:
  • Removed merging functionality
E323AI SO/DLL: E323AI-HighTemplar
E323AI SRC: E323AI@github

My testing and the feedback you guys gave me (thnx) showed that the crashes were caused by the merging code. The AI just can't do it properly at the moment, partly because of bad design on my end. So this should make stuff work again.

Also, because of this design issue, it made me think quite a lot these past days and that has resulted in trying out something new from scratch (yeah that bad...). Peek in my github if ur interested as this is not the place to talk about it and its a bit to soon also.

- Error323
User avatar
Abaddon
Posts: 41
Joined: 13 Apr 2008, 18:36

Re: Skirmish AI: E323AI v3.14.3 - High Templar

Post by Abaddon »

Still unstable :(

Code: Select all

[   9582] Stacktrace:
[   9582] (0) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll [0x176DD40F]
[   9582] (1) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll [0x176DD5AB]
[   9582] (2) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll [0x176C5D36]
[   9582] (3) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll(release+0xf3f) [0x176E5B75]
[   9582] (4) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll(release+0x2a0e7) [0x1770ED1D]
[   9582] (5) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll(release+0xdc5a) [0x176F2890]
[   9582] (6) D:\Games\Spring\AI\Skirmish\E323AI\3.14.3\SkirmishAI.dll(handleEvent+0x96) [0x176E4917]
[   9582] (7) D:\Games\Spring\spring.exe [0x0084B946]
[   9582] (8) D:\Games\Spring\spring.exe(skirmishAiCallback_Unit_CurrentCommand_getOptions+0xcfc6) [0x0086A666]
[   9582] (9) D:\Games\Spring\spring.exe(aiInterfaceCallback_DataDirs_getWriteableDir+0x3869) [0x00880869]
[   9582] (10) D:\Games\Spring\spring.exe(skirmishAiCallback_Unit_CurrentCommand_getOptions+0x14434) [0x00871AD4]
[   9582] (11) D:\Games\Spring\spring.exe [0x004054C8]
[   9582] (12) D:\Games\Spring\spring.exe [0x0041E8DE]
[   9582] (13) D:\Games\Spring\spring.exe [0x00422B4C]
[   9582] (14) D:\Games\Spring\spring.exe [0x007F29BA]
[   9582] (15) D:\Games\Spring\spring.exe [0x007F98D5]
[   9582] (16) D:\Games\Spring\spring.exe [0x007EF16B]
[   9582] (17) D:\Games\Spring\spring.exe [0x007EF611]
[   9582] (18) D:\Games\Spring\spring.exe(aiInterfaceCallback_DataDirs_getWriteableDir+0xabe8) [0x00887BE8]
[   9582] (19) D:\Games\Spring\spring.exe [0x0040124B]
[   9582] (20) D:\Games\Spring\spring.exe [0x004012B8]
[   9582] (21) C:\Windows\syswow64\kernel32.dll(BaseThreadInitThunk+0x12) [0x76B43677]
[   9582] (22) C:\Windows\SysWOW64\ntdll.dll(RtlInitializeExceptionChain+0x63) [0x77299D72]
[   9582] (23) C:\Windows\SysWOW64\ntdll.dll(RtlInitializeExceptionChain+0x36) [0x77299D45]
Post Reply

Return to “AI”