PDA

View Full Version : Why are UT2004 modders so incompetent at developing AI bots for realism mods?


jc!
2nd Sep 2004, 06:34 PM
Granted, UT2004’s AI is not on par with Far Cry’s AI. The mercenaries in that game were incredible; never before had I seen enemies who understood how to prepare ambushes, hide behind cover, and keep moving behind cover if I attempted to circle them.

That said, the absence of AI bots form Far Cry’s multiplayer is preposterous. However, the absence of passable AI bots from UT2004’s realism mods is more preposterous, more so because UT2004 offers a robust AI framework that modders can exploit. Modders have no excuse not to include AI bots in realism mods, not when the UT2004’s bots already do almost everything a realism mod would demand from an implementation of AI bots.

Think about it. UT2004’s bots already understand how to attack, defend, and take multiple routes to the enemy base. They understand how the retreat, stand still and crouch for increased accuracy, and switch weapons depending on the situation. They also understand how to work in teams, listen for footsteps, and capture, hold, or destroy objectives.

Bottom line is, if UT2004’s bots can play Assault, and play it well—better, in fact, than most noobs would—then they should be able play realism mods, complete objectives and fight off humans, all at the same time. So why are UT2004 modders so incompetent at developing AI bots for realism mods?

This problem isn’t restricted to UT2004’s realism mods. Nearly every realism game built off the Unreal engine has featured worse AI than UT2004’s. For example, one of the highlights of the Rainbow Six series is the planning phase, and AI—both enemy and teammate—necessary to make it work. Rainbow Six or Rainbow Six: Rogue Spear’s AI was bad, and I expected Rainbow Six 3: Raven Shield, being built off the Unreal engine, to offer better AI.

Sadly, the AI in this game of “squad-based counter terror” did not receive the same facelift that the graphics did. Rainbow Six 3: Raven Shield’s enemies still did not understand how to work in teams, lean around corners, or hear their comrades getting slaughtered in the other room. However, this was offset by your teammates’ refusal to look left or right while storming rooms—something even FBI agents learn during training, and FBI agents are hardly elite counterterrorists—or hear enemies yelling out, “I need help here!” or “Who the…?”

Once, I even heard an enemy yell out “I need help here!” and “Who the...?” right behind my teammates. Enemies normally cannot yell out more than one line of dialogue unless they glitch, so my teammates had two chances to hear an enemy behind them, yelling at the top of his lungs. Alas, I remembered that my teammates were deaf, right before he killed down. He then wiped out my team, but not by spraying them. First, he killed my second teammate, then turned and killed my third teammate, then turned again before methodically killing my last teammate. Not once did my teammates turn to see what all the gunfire behind them was about. Even UT2004’s AI was better than this.

I won’t even go into how my teammates look straight ahead while storming rooms and getting killed by enemies standing to their left or right. With the exception of Far Cry, UT2004’s AI is still the best on the market—hey, bots can even hear gunfire behind them!

So where the hell are AI bots for UT2004’s realism mods. I understand why the AI in commercial realism games like Rainbow Six 3: Raven Shield is so awful—their programmers can’t program shiit—but modders have no excuse not to take advantage of UT2004’s robust AI framework. Why do developers refuse to exploit it for realism mods?

Taleweaver
2nd Sep 2004, 07:43 PM
<sigh>Let me tell you a little bit about bot AI. Bots know nothing! They aren't real. They don't know how to act properly. They don't know how to improvise. They don't know how to properly take on a given command. What bots do know, however, is what their AI (in this game: whatever Epic tells them how to act like) tells them to, and what the specific map details tell the bots (defending spots, attack courses, and so on).
So...Since bots start out with nothing, how can they be clever in UT2004? Simple: Epic knows how they should react in a UT2004 game, so that's what they coded. This is completely different from the AI used in Farcry. Yes, it could be ported over, but this is stealing! The programmers of Farcry spent weeks (if not months) to come up with a decent code, so you can't just 'copy-paste' that in a free mod for UT.
You also state that bots can play assault better than most n00bs...True, but if you had played it online a couple times, you'd agree that any single player can come up with better strategies than the bots could (they just follow their objectives). Assault isn't a 'realism' mod, since everyone keeps respawning all the time (telling bots to be carefull isn't gonna work). You can't just 'tweak' that setting, you'll have to rewrite everything about the bot AI.

Let me summarize my rantings here: UT2004 bots aren't made for realism mods (hence the name UNreal tournament), so they cannot 'simply' be altered to include it. If you disagree with me, fine. If it's really that easy, post all the solutions to the shortcomings in the realism mods in patterns that computers might understand (eg. IF enemySpotted THEN Hideout ...of course, you'll have to define "enemySpotted" and "Hideout" too), and I'm sure someone from those realism mods would love to code it for you...

igNiTion
2nd Sep 2004, 08:11 PM
And to add to that, coding takes time. Time that mods don't have, especially considering there is usually only one or two coders on the team (from most mods I've seen). And they would much rather code new guns, a better menu interface, or improve gameplay with tweaks than write code to tell a bot when to reload his mp5.

Not to mention, since most of the realism mods have 0 replay value offline (how often have you played cz with bots?) why would they bother? I too, love having working bots, but I can at least understand why the teams would much rather focus on the multiplayer aspects of gameplay, which all of the long-time players will stick to, than worry about some petty bots just so people can play offline once or twice and get bored of the mod.

cole_bie
2nd Sep 2004, 08:32 PM
I say just be happy that UT has such a great mod community.

jc!
2nd Sep 2004, 09:54 PM
What good is a mod community if you can’t play the mods? UT2004 is the only game I play because I can play it offline; playing online is totally unfeasible on my connection.

If you load up a weapons mutator that replaces default weapons with realistic weapons, load up a player mutator that moderates UT2004’s over-the-top player movement, and load up a realistic Assault map, the AI bots should be able to handle it. The mapmaker places waypoints so bots understand where to attack, defend, and snipe, no programming required. And bots automatically understand how to use their weapons by analyzing weapon functions via UnrealScript.

From what shortcomings do UT2004 bots suffer? They already understand most tactics essential to realism mods; they cover each other, assume defensive positions, retreat when injured, reform on you on command, obey radio calls, listen for footsteps, etc. And most importantly, they know how to capture, hold, or destroy objectives, which is pretty much the basis for every realism mod out there.

T2A`
2nd Sep 2004, 10:23 PM
What good is a mod community if you can’t play the mods?You mean I, don't you? ;)

I do think it would be possible and it probably wouldn't be too hard compared to writing a whole AI from scratch. However, these are mod teams full of people who do this in their spare time, not paid professionals. Mods are most fun online and will get the most play online and the mod makers know this. Why waste their time with something that isn't required, when all this would do is push back the mod from being released? If you're implying you can't play UT2004 online with 56k, you're wrong, because it's completely possible. Just ask Ren. Is every mod team supposed to take all that time to implement a different AI for each mod just because a few people can't play online? Maybe in a perfect world, but as it stands now, most mod teams are strapped for time and personnel already. If you're so adamant about getting a fully-functioning realistic AI, why don't you try writing it? You make it sound like it would be easy.

jc!
2nd Sep 2004, 10:45 PM
My point is, you don't need to write new code. Just add realistic weapons, realistic maps, and bots should automatically know what to do, as I explained in my previous reply. The biggest responsibility falls on the mapmaker, who must place waypoints so bots understand where to go, attack, defend, and snipe.

There shouldn't be a big difference between how bots play UT2004 and how they play a realism mod. They still need to run toward the enemy base, shoot, and maybe capture something if everyone isn’t already dead. :lol:

T2A`
2nd Sep 2004, 11:04 PM
No matter what kind of waypoints and weapons you put in the maps, the bots will never crouch beneath a window and peek out to see if the way is clear, then toss a grenade out to create a diversion while they charge off to hide somewhere else unless you tell them to through new code. Here's a basic breakdown of UT2004's bots:

1) Get weapons and items.
2) Kill things with those weapons if they cross my path.

Yes, the basic functionality is there, but putting it all together will require coding. Even if you don't have to recode the bots, you'd still have to code the waypoints to tell them what to do. UT2004 bots do not know nor care about the other bots or people by default. Their main objective is to get stuff, then if they encounter something they can kill, to kill it. They're not overly aware of what's going on outside their bubble.

I do agree that UT2004 bots are probably a good jumping-off point for getting some good AI, but making a map a certain way isn't going to make the bots smart.

HuggyBear
3rd Sep 2004, 12:11 AM
bots are bots and are always lame. They take my manta and run off with it!

Taleweaver
3rd Sep 2004, 03:51 AM
My point is, you don't need to write new code. Just add realistic weapons, realistic maps, and bots should automatically know what to do, as I explained in my previous reply. The biggest responsibility falls on the mapmaker, who must place waypoints so bots understand where to go, attack, defend, and snipe.

There shouldn't be a big difference between how bots play UT2004 and how they play a realism mod. They still need to run toward the enemy base, shoot, and maybe capture something if everyone isn’t already dead. :lol:
Is it so hard to see that there should be written new code for that? Realistic weapons are almost completely trace-hit, hard to trace the shootist and do much more damage/second than the average standard weapon. Abd don't forget that with realism mods, you got only one life.
This combination makes it that your bots can't afford to run around like they do now (they would die before they even saw that person with the machine gun up there). They have to rely on stealth and teamwork to be succesfull, things that aren't sufficiently coded in regards of the weaponry.

I'm just curious: you have played those realism mods (red orchestra) online before, right? There's no point in telling everyone 'they should do it', when you don't even see the enormous differences in gameplay

JaFO
3rd Sep 2004, 04:54 AM
....

Yes, the basic functionality is there, ...
So why have all these mods such a hard time using it ?

Any mod that features 'objectives' is just using another fancy name for Assault.
The bots already know how to complete objectives (destroy items, push buttons, control area) ...
Yet for some strange reason the bots in mods suddenly forgets this stuff.

Take Red Orchestra ...
All the average map requires is for bots to stand near an objective ... like they do in DDOM, like they need to do in Convoy for the bridge-objective ...

Frag.Ops has a map that's basically CTF (capture a suitcase).
Yet in the version I've had the bots didn't even know how to do it anymore.

Perhaps they can't get them to do advanced squad-tactics, because that's something that's missing from the core-routines used by the "AI"
Never mind that on public servers most players act dumber than the bots ...

As for teaching bots the weapons ... they use the same interface the players use. If a player presses 'fire' the the exact same routine is used !
Most 'realism'-mods have an auto-reload feature. Guess how the bots know how to reload ?
Perhaps you'd need to teach them 'tactical' reloads, but the *basics* are in there already.

I think the real reason why most teams are incapable of adding basic bot-support is that they aren't working within the framework of the classes as they exist.
Why else would anyone even need to spend one minute on coding an objective-based gametype for UT2k4 if the game itself provides all the tools already ?
Just add a weapons- and realisme-mutator to UT2k4 Assault and you've got your wannabe CS-clone with objectives done.
Perhaps they simply suffer from the "not developed here"-syndrome so they are convinced that they need to code it themselves.

As for teams not having the time needed. That's a BS excuse. As an amateur you've got as much time as you want.
There's no punishment for failing to meet any deadline.
They're completely free to set their own deadlines ...

//
The fact that "realistic" weapons are trace-hit based makes them easy to use, because as soon as the bot sees you he's doing a trace already ...
In effect the majority of the code is used to make sure the bot does not hit you dead-center every time it takes a shot.

KenjiE20
3rd Sep 2004, 08:04 AM
On the subject of ANY AI; GIGO (Garbage In, Garbage OUT)
i.e. AI is only as good as you tell it to be.
Therefore if the triggers don't exist for them do to something they wont.

JaFO
3rd Sep 2004, 01:35 PM
but the point is the triggers *do* exist.
They're part of the default objects available in Assault gametype.

In fact the only thing a mod-team would need to do for AI to work adequately would be to implement recoil (= make sure they don't fire at exactly the same spot) and tactical reloading (telling the AI to reload their weapons when not in combat). Anything else (the bot-pathing, placement of triggers, etc.) can be done by competent mappers ...

Mad Max RW
3rd Sep 2004, 01:54 PM
I agree, but I realize how much work it takes for good bots. It's a lot easier to just leave the game up to human players. Unfortunately there's too many mods and not enough players. So everybody loses. The teams who spent countless hours are left with a tiny group of devoted fans.

Right now I'm primarily offline and will once in a while join a server. Any mod that has bots (ChaosUT, Clone Bandits, Alien Swarm, Jailbreak) occupy most of my time. They can't be sucky placeholders like certain mods.

Hell, I find on average bots put up more of a fight than the idiot humans I've been running into for years. I'm less frustrated, and at least they follow orders. Overall, I have a better and more fun gaming experience. Of course nothing can beat a human that knows what he's doing, but each time you join a server it's like you're rolling the dice.
I'm not even gonna start on my hatred of lag :P

Mychaeel
3rd Sep 2004, 02:10 PM
Mods [...] will get the most play online
Obviously, mods without any offline play value will. That's circular reasoning.

And obviously, there's demand, or we wouldn't be having this discussion. Of course, the demand doesn't come from within those mods' communities, but taking that as an argument against the value of decent offline playing support would be circular reasoning once more since it's hard to build a community of people who can't play the game to start with.

Parser
3rd Sep 2004, 02:31 PM
I don't know why the big mods don't have coders competent enough to even put a smidgen of bot support in. I'm hardly the best coder ever but I can implement bot support into a gametype that changes gameplay and uses a custom pawn class with new functions, so why can't these "uber coders" doing the big mods such as Red Orchestra and Frag Ops do the same?

It's not hard to make the bots even run around and shoot weapons, or in fact use objectives like in Assault. Everything the coders need is already there, so use it, dammit. If I can do it, you certainly can.

Angel_Mapper
3rd Sep 2004, 02:40 PM
Well I'd like to know your personal experience with mapping and coding jc, it doesn't sound like you know the amount of work that has to go into AI to get it decent. For my mod, I'm actually dumbing them down since they're only animal intelligence, but that's still going to take a lot of work.