The unofficial Supreme Commander map editor
You may have seen some third-party maps available for download. Unfortunately, there is no official map editor (yet?), so people are making their maps using hacked-together third-party tools like the “Unofficial Supreme Commander Map Editor“. It’s a bit basic, and doesn’t have the ability to place all of the terrain features that you see in the official maps. But it does at least allow you to sculpt the landscape of a map.
One thing that this map editor reveals that is a bit disappointing is that cues to the AI are programmed directly into the map. The AI has to be explicitly told where to build defenses, experimental units, and sea factories, and even needs cues on where battles are likely to evolve and where it should build expansion bases. Lame. The AI in Supreme Commander was supposed to be revolutionary, but it turns out that part of the AI is really just hard-coded directly into the maps! As a direct result of the way the defensive points are coded into the map, look at the way the AI’s defenses tend to develop:
Needless to say, that’s an incredibly ineffective defensive pattern, and I was able to get the AI to destroy most of its own defenses simply by running a fast scout into the middle of that mess. The AI needs to learn how to build in lines rather than clusters.
Anyway, I haven’t gotten around to using the map editor just yet, but I am going to try it out eventually and work on some maps of my own. I just wish Gas Powered Games would release whatever internal tool they use to make their maps. Many other companies release the editors that they wrote for themselves and used to make the content in the game because it helps out third party developers, which in turn makes the game more attractive to gamers.
March 6th, 2007 at 8:48 pm
Maybe you should develop an AI that learns how to play through evolutionary/heuristic searching of repeated game simulations. : )
March 6th, 2007 at 9:06 pm
That’s the problem: all of the games out there just use whatever approximation of artificial intelligence game developers think is good. What would be really nice is if a game company hired a whole team of artificial intelligence researchers and put together some truly ground-breaking AI. Then you probably would see some kill algorithms that were evolved using genetic algorithms.
March 7th, 2007 at 9:34 pm
That would be pretty kickass. You’d end up with a company specializing in AIs, which I sort of imagine as being licensed like game engines themselves. In the same way that devs talk up “Game X uses the Quake 19 engine, for superior visuals!”, you’d see “Game X uses the Electro-Death AI engine, for superior pwnage!”
March 7th, 2007 at 11:48 pm
The field is going to be heading in that direction … eventually. The game companies are still getting away with essentially “faking” the AI, unfortunately. These programmers generally don’t really have any training in AI and just approach it on a simple imitation basis.
Is it really so hard to draw a line between your own base and the enemy’s and create orthogonal lines of walls?
March 8th, 2007 at 3:05 am
The AI would need an algorithm to make to tell where a player’s bases are. This could be difficult to define on its own. Outlying MEXes with their one or two AA/PD would be frustrating, as well as trying to decide where one direction of attack begins and where it ends. Against a turtle, it’s fairly simple, but this could become very difficult against an octopus or an eagle. You would likely end up criticizing the AI’s “orthogonal lines of walls”, and laugh gleefully as your two strategic bombers decimate the AI’s base.
You’d also need an algorithm for determining chokepoints both in and outside the base. Inside to keep the TA-famouse piles of T1 munchkin-bots from piling up in the base, and outside to determine good places for other lines of extra-base defense. That is, forward outposts.
Recognizing damage as it’s being dealt is another thing that would make the AI much more formidable. You can arty a base forever and they won’t build shields to counter it. The AI will build shields, but just because it *does* that, not because you’re playing whack-a-mole with its power generation.
I think one of my biggest beefs with the AI is that it will build large piles of popcorn-bots and just leave them in its base when on island maps. It would be nice to have the AI recognize that T1 popcorn is not an effective method of attack on its own. Sure, it sends a transport full of ‘em every so often, but I want the solo campaign’s 15-transport rush of flying death.
Really, it looks like the AI needs a lot of little algorithms. Ten-ish canned strategies that would be called based on the results of various analysis algorithms seems like it would be sufficient at this point. Really, it currently has “attack with ten bits of popcorn”, “attack with two bombers”, “attack with a Monkeylord”, and “sit there and fill the base with cannon fodder”.
March 27th, 2007 at 8:39 am
There’s already something sort of out there that could really help with something like this. N.E.R.O. at http://www.nerogame.org/ which was developed by a CS department’s students. Its pretty neat, and has alot of potential.
October 29th, 2007 at 11:02 pm
All previus ideas on this blog are great but I would like to add that you could also add limited learning capability to the AI. It’s not that hard, my dad plays a 12 year old raceing game that can learn how he passes and gets better at blocking him, finds out how aggresive a driver he is and returns the favor, etc. So I think it would be simple for a subcom AI to understand threw trial and error you can’t defend against a player who favors wave after wave after wave of bombers with one flack and one SAM. Also dose anyone know how the new UEF sub orbital defence grid works I can only find pictures of the building but not when it’s doing anything. thx.
July 25th, 2008 at 12:23 am
yo, u all sound like the smart a**es that sit around all day enjoying this fine game and then sit and talk dirt about how bad the AI is, if u want better AI that doesn’t rely in being preprogramed on what to do, go to a programing school and create the AI power full enough to learn from its mistakes and yours at the same time. make a name for your self be THE hero of stratagy gaming! or play company of heros GREATEST WW2 game ever made (including the sequil opposing fronts)
December 18th, 2008 at 10:56 pm
Yeah! And once we get this new revolutionary AI, it’ll only take 5x the current CPU limitations we have!
Seriously, that kind of heavy AI would further tax an already extremely taxed CPU (as a result of this game). It’s not feasible.
March 17th, 2010 at 5:23 am
Thing is the kind of AI you people have in mind is a little out of the scope even for modern gaming, gaming typically uses the old “chess board approach” any more competitive then you would be complaining its too powerful. Fact is AI in gaming can never become anything close to artificial neural networks, not saying the possibility does not exist its just not very logical in the game sense. You would then assume AI to make human like decisions that is a subject still quiet impossible at least in gaming, it would require to train the neural networks to behave on a cognitive level of a strategist able to mimic and counter whatever you throw at it. Its simply impossible because fact is real AI is slow in the head, in fact most physicists exclaim the smartest human like AI ever created thus far has the intelligence of a retarded cockroach that is the closest we have come real life Terminator or The Matrix. That is worse than a mouse stuck in a maze trying to find its way out so imagine what AI must be like if you had to program it for games.
March 17th, 2010 at 7:58 am
I think most players actually want bad AI. Only a small percentage of the buying public actually go online with their game. They’ve no particular competitive streak so they’re much happier pummelling a bad AI opponent over and over without ever having their egos bruised.
I think “smoke and mirrors” AI is going to be with us for a long time. Trying to do “proper” AI work is nightmarishly difficult. Anyway, I think the main emphasis should be on making an AI that’s entertaining. It doesn’t necessarily have to be smart. It just needs to not be repetitive which the SupCom and TA AI’s always were. I always heard Z: Steel Soldiers praised for it’s unpredictable and merciless AI. It was one of those games I meant to get into properly but I’d no PC at the time and by the time I did nobody was playing online anymore so I wasn’t bothered.