|
Hi everybody,
As part of a research team at the IT University of Copenhagen I am currently investigating automatic generation of maps for strategy games. In a recent paper we describe how we use evolutionary algorithms to create StarCraft maps:
http://julian.togelius.com/Togelius2010Multiobjective.pdf
Now, we would like some experienced StarCraft players to help us validate our algorithms by looking at the maps we have created and giving us your opinions. This is why I am posting in this forum. We would be very happy if as many as possible of you went to the following address, looked at ten algorithmically created StarCraft maps, and told us what you think about them:
http://itu.dk/people/cogr/starcraft_experiment/
With your help, we might come closer to our goal of automatic generation of quality game content.
One more thing: if you intend to read the paper, please do the survey first, so that your reading the paper will not bias your answers.
Thanks a lot in advance for this!
Julian Togelius
julian@togelius.com
|
United States5162 Posts
Wow, really cool man. I'll definitely help out and look at the generated maps.
|
Creating maps through the use of evolutionary algorithms sound really interesting. Will this be how the map with ultimate balance will be constructed? ^_^
|
Sounds awesome, would be great to help you guys out testing.
|
|
|
If you guys actually want to do this help yourselfs out by making 2 player maps only at first, there is absolutely no one way you are going to make these anywhere near balanced with 3 player maps. Also you have to make things more symmetrical with terrain and location of naturals and 3rds.
Good idea, just needs a lot of tweaking :D
|
You seem to have this problem of making the player in the top left get trapped on an island.
|
I have played Battleforge (great RTS mixed with trading card game) a lot.
They introduced randomly generated maps at some point and I really loved those.
Not having to play the same maps over and over all the time is pretty cool, I would love to have that for starcraft laddering.
It just has to be done right, so that the maps don't turn out too weird/imbalanced.
edit: I just took the test, and to be honest it was a little disappointing.
The most important thing for a map to be fair is the symmetry. Every map must have perfect symmetry to even have a chance of being fair...
I really recommend you to look into how the random maps in Battleforge are made, it could help you out a lot.
|
Layout prevented me from completing survey, about 5% of the picture took up my entire screen, had to scroll like mad to see anything, sorry.
But tehre wasn't anything on any of the ten maps... it was a blanket map with scattered resources, no choke points, high ground, ramps, etc. Just literally an open plain with no design whatsoever!
|
Just from glancing at the maps that were generated, you have alot of work ahead of you.
|
Yeah, the maps are pretty imbalanced. Like some bases can just defend a really small choke.
|
Very cool, I did the survey. How complex is the algorithm planned on being? I.E. will you eventually incorporate things like high ground and destructables? Or perhaps SC2 things like towers and vision blockers? Looking forward to the results!
|
Maps won't be balanced unless they're symmetrical. Also, what's with the random gas and minerals scattered around?
|
Is the jungle texture supposed to denote high ground, and the dirt texture low ground? As far as I can tell there are 0 choke points on any of the maps.
|
Creating balanced, fun maps for broodwar is very difficult. Here's the list of maps used by professional Korean gamers and leagues.
http://www.teamliquid.net/tlpd/korean/maps
Key points they have in common:
Main is usually on high ground or at the very least has a choke that can be blocked off by a barracks and 2 supply depots.
Natural expansion should be easy to take, there should only be 1 entrance point to it and it can't be too wide to cover with cannons, or other static defense.
Third base becomes vastly more difficult to defend, and anything beyond 3 bases you're going to need to make yourself vulnerable from the open middle area.
Most maps have a generally open area dead in the middle, usually there are some sort of obstacles there to allow for tactical play.
Symmetry! Where you spawn should not give you any advantage or disadvantage. Especially for 2-player maps.
|
Just finished the survey. Only map out of those that had any potential was number 7, and even then it was still way off. Keep it up and see what happens.
|
Your website crashed my computer. Too bad BW couldn't have real random generated maps instead of some half-assed school project.
|
Things i saw a lot of:
The top left player was frequently in a well designed base with 1 additionally easily defended expansion via choke points.
The bottom middle player was almost always surrounded by no choke points.
The top right player was half of the time in an alcove that is easily defendable and half of the time similar to the bottom middle player.
More symmetry is required for good starcraft maps.
|
Starcraft is played on mirrored maps and your algorithm does not seem to take that into account. Every starting position needs to look exactly the same and have the exact same distance to the natural etc. Otherwise the game will be extremely imbalanced.
As someone said, attempt to create two player maps first, that will be much easier.
Still, a good job so far, just a lot of work ahead of you.
On July 19 2011 07:50 FabledIntegral wrote:
Layout prevented me from completing survey, about 5% of the picture took up my entire screen, had to scroll like mad to see anything, sorry.
But tehre wasn't anything on any of the ten maps... it was a blanket map with scattered resources, no choke points, high ground, ramps, etc. Just literally an open plain with no design whatsoever!
They are likely in the early stages so the maps are supposed to be flat. When they ask you for chokes they mean the stuff they have created with grass. There are two different ground textures and you just have to use a bit of imagination 
|
for science !
i dont think any of the random maps i looked at would be any good to play on, but if this helps people make better maps im all for it.
|
United States5162 Posts
As others have said, none of the maps are symmetrical and only the last few even start to get the initial bases close to fair. I would definitely agree to do 2 player maps rather than 3 player, not even humans can get those things perfect all the time.
|
Cool concept, but nothing resembling balanced. Wouldn't it be easier to do 2 or 4 player maps? It seems like it would be easier to mirror everything so even if you don't have an optimal layout you get reasonable starting areas.
It might be useful to look at played maps and try and mimic the distances between the key points of the map, like distance to base to base, natural to natural, secondary Tertiary bases.
|
The idea sounds really nice
It might take a lot of time but if you make it with ultimate balance and pathing it would be a very good achievement for BW players and a big step for RTS gaming. Cool stuff.
|
Wow, that's a lot of responses in a short time! Thanks a lot - I hope all of you take the survey as well, and fill out the comment field so we can have your input nicely collected in our database, and include in our next paper on this.
While I know all of you know StarCraft better than you do (and most of you probably better than my co-authors do) there are a few things that puzzle me.
In particular, I wonder about symmetry. The argument I hear is that maps are never fair unless they are symmetric. I personally think symmetric maps are boring - symmetry means that you know exactly how the enemy's territory looks even before visiting it. (Of course, if you play the same map over and over there you would know what the enemy's home territory looks like anyway - but our idea is that with automatic map generation, you would never need to play the same map more than once, like in Civilization.) What we are trying to do is to measure and optimize various types of balance so that maps can be balanced without being symmetric. I realize this is quite hard to achieve, but we need to start somewhere.
As for the comments that we should have started with two-player maps rather than three-player maps: you are probably right. But we wanted a challenge. And the nice thing about three-player maps is that it is impossible to make them perfectly symmetric.
A note about the maps: all the maps have a single level only (no ramps etc) and consist only of ground and mountains/cliffs. I'm sorry that we apparently didn't get the textures right...
Thanks a lot for your comments, we will read all of them - keep them coming!
|
United States17042 Posts
moved to broodwar general.
Seems really exciting though ^^ Great to see that research and innovation is still happening in this space.
|
United States5162 Posts
togelius,
I agree you probably don't need perfect symmetry for your maps as you have a different target audience than what everyone here is used to, but I do think you need general symmetry. I don't see how you'll overcome the huge difference in initial advantage that a choke provides compared to an open base.
edit: Although something I saw in your maps and it struck me as kinda cool, a huge impassible terrain that provides a choke, but it's so far away from the base it 'protects' that it's not really possible to wall it efficiently. This makes from interesting non-symmetry that doesn't give one player a huge advantage. I'll still agree with my initial point though, having a direct choke into someone's base while another player doesn't seems impossible to make fair.
|
You have to take into account that different races values different terrain features differently when you're working with your algorithm.
Many differents...
|
Guys, map being symmetric is only imprtant in the case of perfect information. Which is the case for korean proscene with thousands of hours of practice on all maps. In the case of extremely limited information, like no preknowledge of map, symmetry is quite secondary to impulsive adaption to uncertain conditions.
But the problem with maps THIS open is that zerg can just dominate every game with speedlings. Terran and Protoss t1 units have no chance against them without tight choked bases.
|
The maps were too assymetrical, take a look at the current korean map pool, they're good maps. GL on your investigation
|
![[image loading]](http://i.imgur.com/D4ehz.png)
This was one of the most interesting maps that it chewed out, quite unlike any of the other ones I got.
The layout of the chokes, with a little tweaking could make something legitimate, my biggest problem is the proximity of minerals to the starting bases, in everycase, the mining times would be drastically different, and one player would be very far ahead of another in the first 2 minutes.
Really interesting project however, good luck in your further investigation.
|
Not really sure how to answer these questions.
"Fair resource distribution" - is that relative to each base, including expansions? Or is it just talking about the mins/vespene next to the starting bases? Also, concerning "distribution" - all the bases appear to have the same number of resources, but the mins/vespene are of varying distance from the nexus, so some bases would get resources faster. Is that "fair"?
"This map has choke points" - Existence of choke points or fair choke points?
Good luck in your project!
edit: sp
|
Good idea, the idea of randomly generated maps is a really cool one. Unfortunately you're probably going to have to make it random with some strict guidelines for it to be fair.
1. Symmetry, sorry but it has to exist for it to be fair. Randomly generate a quarter/third/half of the map and put one starting point on it, then copy it 4/3/2 times to have your map. This can make simple designs seem more intricate and interesting as well. Think cutting out a snowflake shape in paper.
2. SC generally requires a natural that is close distance for it to be fair for the less mobile races. There should be a starting base with somewhat of a choke into a natural expansion rule.
If you incorporate this kind of thing into your maps it'll make them far more playable. I really like that somebody is researching this because it'd be so cool to have randomly generated maps making each game unique, but those maps kindof look like my 9 year old cousins attempt at making a SC map. Keep trying, but unfortunately you can't have it completely random and have it be playable or fair.
|
In general - Left can wall off and take way too much crap without punishment, even take additional crap right outside his choke. Right is too far away from anything, especially his second gas, and a lot of times has to choose between a wide open mineral only or 1/2 gas expo. Bottom enjoys larger rushing distances, but has no space whatsoever in his main, and is fragmented and wide open for any kind of harassment.
I would have used "How imbalanced is this map on a scale of 100 to over 9000" and "Is Left imba? Fuck yes / Hell yeah!" and "Is this map frustrating? Yup / Well duh" options to rate them.
I liked only the 5th map, but just because Right and Bottom can battle over that 3 gas expo.
|
On July 19 2011 08:01 Stenstyren wrote:
Starcraft is played on mirrored maps and your algorithm does not seem to take that into account. Every starting position needs to look exactly the same and have the exact same distance to the natural etc. Otherwise the game will be extremely imbalanced.
Well, not so much in Immersive maps, but the algorithm lost alot of basic points.
Im not sure if we all get the same map to judge, but take a look at number 4, i can see major turtling from the 9 o-clock start, but in my head, adjusting the map alittle (two players) moving resources about, i started imaging a PVT on that map, Terran making choices such as wraiths, mass goliath (counter carrier mass) strong bio-tank and going for a deathpush, i started picturing observers with templars storming over the ridge, reavers and shield batteries holding chokes, drops and air from protoss.
I started getting nerd chills.
These maps COULD very work well in a casual, lan environment, imagine that you sit with a group of buddies, you roll a map, air vs turtle, then come up on the spot a gameplan and play.
Though, i think this needs to stay FAR FAR away from 3 starting locations, or else its asking to be mirrored.
EDIT:
I didnt know you could take the picture from the thing, i ment this map
http://itu.dk/people/cogr/starcraft_experiment/img/EvoMap6.png
Just start twisting a few mineral locations around, walls/cliffs, make 2 players, could be entertaining.
|
17020 Posts
This is really cool
My independent study project next year is to use Markov Chains and genetic algorithms to compose Bach chorales :D
|
Most people seem to be forgetting maps like Fantasy II. Making a somewhat balanced map while eschewing symmetry is difficult but not impossible. In the context of the maps generated so far, however, there's a lot of progress to be made. There are many common features for almost all modern maps, symmetry or not.
![[image loading]](http://www.teamliquid.net/tlpd/images/maps/167_Fantasy%20II.jpg)
|
OP you might try to ask a few map authors, I am pretty sure they would be eager to give you some advice in your project regarding map layout and balance.
Edit: A more complex task would be the creation of a simulator that plays CPUs on the generated maps and rates them according to positional and racial balance.
|
I stopped after 3. Those maps were mega imbalanced due to spawn positions.
|
On July 19 2011 10:00 Bwenjarin Raffrack wrote:Most people seem to be forgetting maps like Fantasy II. Making a somewhat balanced map while eschewing symmetry is difficult but not impossible. In the context of the maps generated so far, however, there's a lot of progress to be made. There are many common features for almost all modern maps, symmetry or not.
Oh god was that map beautiful. Just like that picture, the starting locations need to support game plans to take out other locations.
Im thinking the algorythm, seperated into 3 parts (for each spawning location, if you are deadset on 3player maps) could take into account of different playstyles such as air, turtleing, fast expansions and that sort of jazz, the you plug 3 gameplans/bases into a map, see how that looks.
|
Is there a tool in BW that compares maps by things like rush distances, main base space and so on? What are the spans of fun playable maps for the variables?
As for this. Top left would win on most of the maps, no problem. Having a choke and large main is huge in most RTS games.
Some game designs ignore the chokes or the main size. I can't think of any that ignores both except for 4x type of games that still has chokes/defenders advantage to a certain degree.
The only way I see these maps working is if the miners move to the minerals and then sit on them. Per the wc3 tree mining of night elves or undead gold. After that you have to have a game design where the majority of units ignore chokes. Thus most units have to be flyers/diggers, removing most importance from the chokes.
Perhaps the top tier units are effected by chokes, that would still give them meaning.
I can't think of a single game like this though.
I liked the random maps in RA (I think it was this c&c game) and AOE2.
|
I suggest you take the most basic of looks at other games that employ randomly generated maps, most notably Age of Empires (e.g. make it churn out a whole bunch of maps and analise general trends, if you want to), because ultimately the issue with the maps you've provided is that, I'll repeat it, they arent fair, because mineral distances/terrain features that apply to one spawn don't apply to another. That's just inherent map imballance, there's no way you can say that a spawn with a single tight choke is any way equal or equivalent to a spawn with just an empty expanse around it. Also in my experience 3 player maps are the HARDEST to ballance... would it be possible if you could do a quick run of 128x96 2 player maps that have diagonal symmetry + Show Spoiler [like this] +
But in any case props to you for doing this, I've always loved fantasizing about maps and new maps; if you want ANY sort of advice on Starcraft Maps from english-speakers, you really need to contact some old BWMaps people (off the top of my head, the only ones that I've seen around recently are neobowman and Chef (PsycHOTemplar) ), the know their shit son!
|
It's a great idea to work on this. Please consider adjusting the algorithms to produce symmetric maps though. It would improve the effect dramatically.
|
I suggest you stay away from 3 player maps, as those are very difficult to make and from all the maps i saw there were none that were balanced for all 3 players at all. Still like the idea though, but 2/4 player is much easier and lets you define the algorithm better, the symmetry may suck because it makes your challenge easier, but it makes the enjoyment of games played on those maps that much better.
|
Canada2480 Posts
On July 19 2011 07:50 101TFP wrote:
I have played Battleforge (great RTS mixed with trading card game) a lot.
They introduced randomly generated maps at some point and I really loved those.
Not having to play the same maps over and over all the time is pretty cool, I would love to have that for starcraft laddering.
It just has to be done right, so that the maps don't turn out too weird/imbalanced.
edit: I just took the test, and to be honest it was a little disappointing.
The most important thing for a map to be fair is the symmetry. Every map must have perfect symmetry to even have a chance of being fair...
I really recommend you to look into how the random maps in Battleforge are made, it could help you out a lot.
Fantasy 2 would like to have a word with you
|
I was excited to check out this experiment and participate... until I saw the maps.
I'm afraid my answer was the same for all of them, nothing was fair or even and every map was flawed.
|
I'm not sure how far along this project is supposed to be, so I don't want to be too harsh, but at the same time all of these maps seemed nearly unplayably bad. Resources weren't even on any. Most had a pattern: top left - choke, "standard" main; right side semi-choke (or chokes); bottom wide open. I'd have to agree that it would seem to make more sense to try 2 or 4 player maps first but I don't know what constraints you're working with in the actual programming.
|
You should attempt to make 4 player maps. 2 Player and 3 Player maps are very difficult to balance. Almost all the 2 and 3 player maps at the professional level have balance problems. Most 4 player maps are alright.
The maps that you generated though are unplayable.
|
Hi OP,
I'm sorry to be cruel here, but if your maps are not symetric noone will play those maps, NOONE. So you and your team can enjoy your wonderful project by yourselves and noone will care. They are BAD maps.
In the other hand, just make them symetric and a bunch of people will play them and give you feedback despite them still being imbalanced.
|
These maps are terrible, not to mention imba. Those algorithms need some serious reworking to accommodate balance.
|
|
|
Link doesn't work for me, redirects to some random site...
|
Okay, I have sent them a line, as after the survey they gave an email to do so, if I wished.
This is what I wrote:
+ Show Spoiler +Hello,
I don't know why, but you seem somewhat randomly put resources at a map, and by this I don't mean that you make new bases randomly, but that you don't keep them tight.
You should recognize a basic structure of a base in SC, and try to distribute such a structures around a map (I assume you don't do it on purpose with a reason I can't imagine). There are three requirements for a base in SC:
1) It needs a place to put a main building.
2) It require minerals to be put in the shortest possible distance form the main building; usually half of them is placed one grid further.
3) it needs gas in the shortest possible distance form the main building. In starting position it should be exactly above main building, so 3 workers can mine it effectively, which is extremely important in the early game. Some bases can have minerals only.
There is also a single requirement to keep resources distribution fair:
1) From each starting location you should have VERY similar distances to new bases. It is allowed for bases not to be in a number that will be suitable for equal division among players, BUT they have to have equal chances to get it, so it must be in equal distances from all players.
Once you keep bases organized under those conditions, you will immediately get fair resources distribution.
Just place whole bases around the map!
good luck!
Double T from TeamLiquid forums
We need to find more such an "objects" like bases, so they can make more of a jigsaw puzzle than... well, whatever it was
|
None of the maps generation have fair starting locations... I wonder if they are still working on it.
|
|
|
|
|
|
EPT![[image loading]](http://www.teamliquid.net/tlpd/images/maps/188_Destination.jpg)
