Spring RTS Engine

Rather than keeping download links and updates for KAI 0.12 (renamed to KAIK 0.13 as of 9/9/2007) tucked away deep inside Krogothe's old thread about the unofficial release of 0.2 and various other places, I'm centralizing things a bit by starting my own "official" thread about it. The latest version at this time of writing (23/10/2009) is:

(no link, comes with the installer)

0.13 is not as well-rounded an AI as NTAI or RAI, but gives decent (in AI terms) games on average if you don't rush it too much. Features that are supported presently are:

• * Most convential mods (it won't take well to eg. Kernel Panic)
  * EE-style hubs
  * Nukes
  * D-Gunning
  * Extractor upgrading
  * Serialization (saving and loading games)

What it doesn't support (yet, subject to change) includes mainly:

• * Basically anything that floats (0.13 can't swim )
  * Shields and anti-nukes
  * Nano-towers

Why are you not developing with Victors 0.23?

You AI guys should work together on one AI!

clericvash wrote:You AI guys should work together on one AI!

Yeah, they should focus on one or two or maybe 3. But not two versions of the same ai.
Okay, the lobby client developers are not better.

@Agon

As I mentioned in the old thread, Viktor's megapatch and later commits already took care of a lot of things that I had planned to do for 0.23, so improving 0.12 instead seemed more worthwhile. Moreover, 0.12 and 0.23 were (and are) hardly the same where playing style or internal workings are concerned so room exists for both IMO. There is also something to be said for variety, more AI's never hurt (especially on the Linux front) and if anything I'd like to see more AI projects spring up to spur some friendly competition again, not less.

Besides that the idea of AI coders joining forces to create some sort of uber-AI is just not realistic or practical, there are too many different approaches and points of view going around to unify into a single coherent whole and in general people prefer doing their own thing to hard-to-organize online cooperation, so it probably won't happen unless you offer a large sum of money in return.

Kloot wrote:...

Besides that the idea of AI coders joining forces to create some sort of uber-AI is just not realistic or practical, there are too many different approaches and points of view going around to unify into a single coherent whole and in general people prefer doing their own thing to hard-to-organize online cooperation, so it probably won't happen unless you offer a large sum of money in return.

Bull, you should not try to make one 'uber-AI' I can agree with that, but you should try to abstract the general AI functionality into separate libs that can be used by all AI projects. Like that code that can read the metal map to help the AI place the extractors.

Finding all the metal spots is basically an engineering problem though, not an AI reasoning one. There are very few AI components with such clearly defined tasks that you could move them into a library and use each independently. And anyway, that's sort of beside the real point.

this is playing very nicely in XTA now. as EE GDI, all the commander does is spam tonnes of gas turbines. necro/fark only spams adv radar towers in XTA.

Yeah, I noticed the EE issue too. It's fixed in SVN along with the teching-up problem.

The fark/necro thing is weird though, since I spent several hours hunting that bug down and eliminating it for this release. I'm pretty sure it can no longer happen, I've watched a lot of XTA games and KAI always used them to assist stuff and to construct moho mines, Vipers, etc, never to pump out radar farms like it did in older builds. Do you have a demo by any chance?

no, i was reporting from previous version sorry. nice to know its fixed :)
have you fixed the thing of EE never making a tech 2 nanotower?

also, should consider changing the behavior of units with flamethrower type weapons so they engage from minimum range rather than maximum

Sadly weapons don't have a flamethrower tag, I don't really know of an easy way to detect them so I can't modify attack behavior like that.

EE con-units indeed now build L2 and L3 hubs too.

New release (31/7/2007) is up on UF which takes care of the EE problems.

nice one

Kloot this is working really well now, prob the best AI at the moment. Thanks for all the work you've put in to get it working again.

Yess. So far this has been the most challenging (and intelligent... some build useless crap too much to useless places) AI I've tried. Thanks :)

and +1 for the linux version too!

Thank you. All credit to Krogothe/Firenu/Tournesol for writing an AI that still holds up well against the competition, my upgrades to KAI haven't advanced it nearly as much (I would rate RAI the top config-free AI out there right now).

Please, please rename the AI and give it a sensible version numbering system. I've been reading forum threads for the past three hours and I still can't figure out what version is which.

The first post in this thread should tell you that, but
here's an overview for the sake of historical accuracy:

• KAI 0.11: the last official non-development version of KAI
  that was released by Krogothe et all (over a year ago) for
  Spring 0.71 or thereabouts. The 0.11 source he uploaded
  to SVN contained a few logic bugs that prevented it from
  building anything though, so KAI was unfortunately thought
  to be dead for a while.
  
  KAI 0.12: "my" partially patched, unofficial version of KAI
  0.11, initially built for Spring 0.74b2. Binaries of this one
  have been bundled with Spring since 0.74b3, and I've put
  out several interim builds as well (AI development doesn't
  have to be synced with Spring development). None of my
  updates have really moved KAI far enough away from 0.11
  to warrant increasing its version further however, so it has
  remained static at 0.12 (update: 0.13 now). Also, starting
  with the next release of Spring, all 0.12+ binaries will print
  a version string on initialization that'll make identifying their
  exact revision easier.
  
  
  
  KAI 0.21: a development version of KAI featuring a lot of
  stuff that 0.11 didn't (and 0.13 still doesn't, like full water
  support), but playing less well than the release version
  overall. Krogothe never finished it, but made an unofficial
  preview build available once and committed its source to
  SVN along with 0.11's.
  
  KAI 0.22: in essence "my" version of 0.21 plus rudimentary
  d-gun logic and minus a crash-bug, called 0.22 to indicate
  its unofficial status. Due to the fact that 0.21 didn't really
  offer a challenge, it wasn't packaged with Spring like 0.12.
  
  KAI 0.23: a recent total overhaul of 0.22 by Viktor that turns
  it into a real competitive AI (save for some economy issues).

I think that about covers it.

Wow, someone actually managed to make sense of my messy, uncommented code!
very nice
Whats the d-gunning routine like? Id upload my code for predictive dgunning (remember the video i made of it hitting planes/jeffys with it?) but a hard disk failure wiped all my old code...
If you have questions about it, let me know