View Issue Details
| ID | Project | Category | Date Submitted | Last Update | |
|---|---|---|---|---|---|
| 0002587 | AI War 1 / Classic | Suggestion - AI Behavior And Tactics | Jan 21, 2011 11:44 am | Mar 23, 2011 9:38 pm | |
| Reporter | Draco18s | Assigned To | Chris_McElligottPark | ||
| Severity | feature | ||||
| Status | closed | Resolution | no change required | ||
| Product Version | 4.072 | ||||
| Summary | 0002587: Saw an article on RTS AI Design | ||||
| Description | http://arstechnica.com/gaming/news/2011/01/skynet-meets-the-swarm-how-the-berkeley-overmind-won-the-2010-starcraft-ai-competition.ars Basically a group of students designed an AI to play Starcraft at a level that exceeds a human player's ability. There's some interesting tidbits in there, particularly how they have the AI controlling mutalisks to dart in and out of range when attacking. So while I'm not requesting anything specific here, it might hold some inspiration on making the AI just that more dangerous. | ||||
| Tags | No tags attached. | ||||
| Internal Weight | |||||
|
|
A Kitemaster AI? That's scary. |
|
|
:D Although I was thinking more sophistication in the general AI thread, rather than as an AI Type. Fluctuating some of the behavior patterns results in less effective tactics, for example (in terms of pathing, threat avoidance, target priority, etc.). They did have to teach their AI how to compare two units in different situations. I.E. a smarter overall AI. One that chooses which units to use in waves or splits a CPA and holds some units in reserve, rather than "massing everything." But yes. An AI that used brutal tactics, such as you've mentioned, would be scary indeed. How many high-range, high-speed units do we have that would be good "kiters" to give it as bonus units? |
|
|
It is not this group of students. Read the 'complete' article ;) I'll give you a hint, the last bit of the link says "starcraft-ai-competition". :D |
|
|
They were participating in the competition, but that particular AI was developed by Berkeley students. It's called "The Berkeley Overmind" for a reason. :P |
|
|
Just saying it wasnt them inventing anything, they participated in an endevour set out by yet someone else. They come with the example, not the inspiration. |
|
|
I don't see the relevance of that comment to why I linked it here. :| |
|
|
I read that article too. Sparky is funny heheheh :P |
|
|
Kiting AI units would be scary. What if all this time we've been testing the experimental auto kite behavior so that the AI can use it instead? Of course then if you got a kiter vs a kiter they would just rubberband back and forth at each other. Although I've turned it off now since my units kept kiting away into other enemy ships, which didn't help too much. |
|
|
The rubber banding is an emergent behavior, if not particularly [i]useful[/i]. |
|
|
The stuff about "field control" seems interesting and possibly relevent. Teach the AI to go around turret balls instead of attacking targets in a straight line would make the planet-scale AI a lot scarier. |
|
|
You know what'd be even more scarier? Keith or Chris marking this as 'minor fix for later'... :D |
|
|
hahhahahah |
|
|
@MaxAstro: Precisely >:D Like I said, there's a lot of really good stuff in there. Maybe not for AI War per say, but really good stuff. |
|
|
Thanks for the article -- that's a good read, at any rate. A lot of it really isn't applicable on the sort of scale that AI War is on; Starcraft has tiny maps, and so you can do all sorts of clever things with field control by simple pathfinding. I've done similar things in the past with my game Alden Ridge, with 300 zombies chasing two players dynamically, actually. If Alden Ridge was a technical difficulty of 2 out of 10 for that specific sort of case, then Starcraft would be perhaps an 8 out of 10 on that specific sort of case. Maybe a 7. AI War... AI War would be an 11, no joke. It's an obvious case for quad trees, but when you get into the number of quad trees for the number of planets in play, and the number of different kinds of ships and their various properties and most importantly firing ranges... it's something I've thought a TON about over the last two years, actually, but I haven't had any mental breakthroughs on it yet. |
|
|
Like I said, maybe not good for AI War, but worth reading in any case. |
|
|
Oh, for sure. Thinking about that sort of thing is what leads to improvements, ultimately. I just have to be cleverer than usual because of the scale of this particular game. |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| Jan 21, 2011 11:44 am | Draco18s | New Issue | |
| Jan 21, 2011 1:13 pm | Prezombie | Note Added: 0009496 | |
| Jan 21, 2011 1:18 pm | Draco18s | Note Added: 0009497 | |
| Jan 21, 2011 1:19 pm | Draco18s | Note Edited: 0009497 | |
| Jan 21, 2011 1:19 pm | Red Spot | Note Added: 0009498 | |
| Jan 21, 2011 1:20 pm | Draco18s | Note Added: 0009499 | |
| Jan 21, 2011 1:21 pm | Draco18s | Note Edited: 0009499 | |
| Jan 21, 2011 1:26 pm | Red Spot | Note Added: 0009501 | |
| Jan 21, 2011 1:32 pm | Draco18s | Note Added: 0009503 | |
| Jan 21, 2011 3:13 pm | lyravega | Note Added: 0009509 | |
| Jan 21, 2011 3:46 pm | BobTheJanitor | Note Added: 0009511 | |
| Jan 21, 2011 3:50 pm | Draco18s | Note Added: 0009512 | |
| Jan 22, 2011 1:49 am | MaxAstro | Note Added: 0009532 | |
| Jan 22, 2011 2:23 am | Kordy | Note Added: 0009533 | |
| Jan 22, 2011 7:00 am | lyravega | Note Added: 0009534 | |
| Jan 22, 2011 11:05 am | Draco18s | Note Added: 0009543 | |
| Mar 23, 2011 8:32 pm | Chris_McElligottPark | Note Added: 0011390 | |
| Mar 23, 2011 8:32 pm | Chris_McElligottPark | Status | new => closed |
| Mar 23, 2011 8:32 pm | Chris_McElligottPark | Assigned To | => Chris_McElligottPark |
| Mar 23, 2011 8:32 pm | Chris_McElligottPark | Resolution | open => no change required |
| Mar 23, 2011 8:43 pm | Draco18s | Note Added: 0011391 | |
| Mar 23, 2011 9:38 pm | Chris_McElligottPark | Note Added: 0011393 |
