But the thing is, this isn’t true at all. For example in Killzone 2, a large chunk of the AI is based on precomputed values for certain locations on the map. The effectiveness of all the types of cover (both for the enemy, and where the player is) were compiled into a giant table for each map.
So instead of “looking” at where the player is and what he’s doing, and evaluating environmental factors in realtime, the CPU is going through a lookup table and making simple numerical calculations about which waypoint it should (or shouldn’t) move to next.
Why did they do it? So that they could process AI for a bunch of enemies together without hogging CPU.
But the point is… It WORKS. It’s a simple system, but in gameplay to the player it appears that the enemies are making smart decisions and are aware of the map geometry.
It’s all about what’s effictive in actual gameplay.
Another example of this is the way that top-level professional human gamers work. I’ll use Counter-Strike as an example:
The top CS players have played the competition maps so many times, that they start to understand which positions generally offer an advantage, and which are weak. They remember where enemy players will generally attack from on a map, and how long it takes them to get there.
Newbie player:
Which way should I go after spawn?
Hmm, I should check to make sure this pathway is safe
Seems like everyone else is going this way, so maybe I’ll follow them
I’ve just been shot… Where’d I get shot from?
Is there an enemy around this corner?
Professional player:
Spot A gives me cover while allowing me to scout for an enemy advance – so I’ll go there. It’ll take the other team about half a minute to get here.
The second process is much simpler. So why does it work? Because the pro has learned what gives the best results in actual gameplay. Besides their great reflexes and aim, they’ve simplified a bunch of complex decisions into something simple that just works.
And when the newbie player walks around the corner and the professional player is setup in perfect position to shoot him in the face… The truth is that the newbie player was making all kinds of evaluations on the spot that the pro player didn’t had to make.
Code warrior, you gave an example of BAD AI. BAD AI can be complex or uncomplex. Any AI that gives NPC’s knowledge outside it’s scope is BAD. They can make it as complex-ly bad as they want, but it’s still BAD.
Here two sub-questions regarding AI. Since the Black Mesa will reuse the Combine AI from HL2, will the HGrunt still be able to throw grenades from the M4? Also, will the AGrunt also use Combine AI?
How is it bad, if it works well in gameplay? Deciding if AI is good or bad based on the approach alone is like deciding whether the graphics are good or bad from looking at the source code.
If you’re infiltrating an enemy base, and the guards all run out at you like lambs to the slaughter… You’d prefer that in terms of AI?
But really, I prefer if the AI did something more than sit in corners. Taking cover is one thing, but you should be able to catch them off-guard and exposed when you show up to the scene. Just because it’s in an action game doesn’t mean the intelligence has to be prepared to fight at all times.
STALKER had pretty good AI, I was always getting the jump on those pathetic loners at their campfires. (Wait, you mean to tell me they were originally neutral?!)
Anyway, the “newbie AI” just seems more realistic. It’s more stuff, and it might not be more effective, but it’s what people do. People are often newbies.
AI shouldn’t know all the corners of a map like an uber-pro-CS-player does. AI shouldn’t hide behind invisible walls, or shoot through fake walls, even if it’d be more effective. AI should be curious, not all-knowing. If they’re always shooting first and asking questions later, when will you feel that tension of an enemy sneaking toward your low-healthed self in a hiding place? Players should not expect to walk into every room with AI camping behind every box. They should show some versatility, and care to move around as they get bored.
(Again with STALKER, sometimes those soldiers were stealthy motherfuckers. It’s not often I turn around on an open hill in a game to see a soldier crawling toward me, gun drawn but not yet firing. I don’t know where he was hiding, but he was… that game doesn’t spawn enemies anywhere near line-of-sight. It’s as if he understood the Stormtrooper Effect, and was circumventing it by getting close. The surprise alone threw my own aim off. Nearly killed me from full health despite every advantage in equipment I had.)
I disagree. Go try to get into a high-level military facility (or even a closely guarded private lab) and see how far you get without being intercepted, haha. The fact is, they have highly trained security ready to lock areas down at even the slightest suspicion of something funny going on.
It’s not like your high school where the guard sips coffee all day while playing DS, haha. A lot of the guys that provide security for installations are retired SEALs, Green Berets, or other members of special forces.
And I guarantee, they will know their environment. Even in a war you don’t just drop troops into the middle of nowhere like it’s 1945. Everything is scouted extensively beforehand, and routes of approach, areas of cover, places to set up snipers/etc are all known before anyone gets there.
And it’s not just the big country armies that do this… The taliban are known for knowing every rock, tree, hill, building and crevasse than can possibly be used to disrupt the forces they fight against. The American Rebels did it in 1776…
So the whole idea of going into some top-secret facility with advanced weaponry and having all the enemies come at you Rambo-style makes little sense unless you’re playing L4D… You can pretty much call that “Artificial Non-Intellgence”.
I think AI should be done on a map-by-map basis, and for each NPC. It would require more work, but it would make sense- Take HL for example. Under this system, Guards would know the area well, and the grunts in the first levels would seem disoriented, confused, and very cautious in their movements. In later levels, grunts would know the area better.
They’d also not be physic either way, because that just makes them seem unrealistic.
Haven’t the devs already said that they would not be overtly changing the gameplay mechanics from the original so as to remain a faithful remake?
And i thought the HECU marines were the ones who set up the trip mines in the first place. When they tripped them it was simply a case of bad observation and communication (or the player themselves setting one up in an unpredictable spot)… why would they want to remotely blow up a mine that they themselves laid down anyway?
And as for acting in a fashion where they jump to organised defensive positions… Even taking into account that they are in unfamiliar territory in the black mesa facility, wouldn’t it be part of their training to observe defensive and offensive tactical points and places of opportunity in their environment?.. so it’s not far fetched that they would be able to decide where they defend from in a short space of time. They don’t have to act like noobs =p
Which reminds me, they should also make sure they are out of the way of any explosives in proximity of the mine. Don’t want a chain reaction to get them.
I think we can all appreciate the fact that a varied AI provides an interesting AI. They don’t have to be consistently stupid or consistently professional,
consistently cautious or consistently ambitious, consistently lazy or consistently running.
Can’t you see one out of five soldiers trying to play the hero? One out of ten preferring to run when the player advances? Some preferring to hold their fire if they’re not in the player’s FOV, but instead opting to flank or ambush?
Anyway, with something like a Resonance Cascade, funky stuff is always going on. When stuff is consistently funky over the course of 18hrs, you’re going to start relaxing (or making mistakes) whether you should or not.
I think it’s ridiculous to say the grunts should know and exploit the features of terrain, even later in the game. It’s a chaotic situation. The game is supposed to give you the feeling that not even this omnipresent force is able to keep control…we see grunts eaten left and right. (They’re not even wholly obedient troops, mind you… “What the hell for? We got 'im… let’s kill him now!”)
How on earth did you know I surveilled them years ago?
I responded too hastily with my BAD AI rant to this. I looked back at it. Those massive tables of precomputed values… that’s part of the AI coding. It might not be the active AI coding, but since it is part of what makes the NPC tick, it’s part of the AI.
I think it would be really cool, if the soldiers would take use of common combat tactics as the rebels in this old E3 video: https://www.youtube.com/watch?v=y_EFqr2Bl1I (@ 04 : 34 - 05:10). Leaning against a wall, waiting for the first one to get in, then charging and spreading out for attack. Simply epic!
Even if it would only be scripted, everyone will say: What a brilliant AI!
I’d like to point out there’s a difference between smart AI and realistic AI.
Smart AI will always know where you are once they see you.
Realistic AI will lose you when they lose sight of you and you disappear in a door.
Making them seem like real people instead of supersoldiers with X-ray vision and super hearing means the player will enjoy the game more, and see the AI as better.
Actually it looks a lot better in the videos. Mainly the part where they hide behind cars, and run ahead of you, rather than crowding around you all the time.
Founded in 2004, Leakfree.org became one of the first online communities dedicated to Valve’s Source engine development. It is more famously known for the formation of Black Mesa: Source under the 'Leakfree Modification Team' handle in September 2004.
