This is obviously a personal thing for you. That’s ok, we’ll let it rest.
Sebmojo
1582
I believe the leaked version is like the first …third? of the game.
Something like that. It took a …“friend” of mine… about 8 hours to get through it, and I know for a fact he didn’t explore all he could have, not by a long shot.
:)
Careful what you wish for. In the PCG review, the author mentions creating a conspicuous cardboard box hallway to get to a hack terminal. I took that to mean a guard was watching this box hallway being built, but did nothing because the player was behind a physical object. The author thought it was lolarious and part of me sympathizes with goofy emergence like that (I’m not too much of a nerd rager about Deus Ex) but don’t let the developers fool you about their highbrow new design. It’s a system that breaks realism just like shadows and nearly-blind enemies in previous stealth games.
Speaking of that review, the author also mentioned how much he liked the killer takedowns. He said they never got old. I can enjoy big blasty fun at the expense of depth (Dawn of War II, for example) but games like that never quite become an all time favorite. So the review did nothing to change my expectations: it should be a good romp even if it doesn’t gel into reverence. We’ll see.
Well, if i can “wish for”, i wish for an enough advanced AI which will investigate a strange cardboad construction in the middle of the corridor. :P
Out of curiosity, how would that work? A stationary guard that behaves the way we’re afraid the game will behave is simple. You just have the guard throw a ray every cycle in the range he can see and if it runs into the character model, you flip him to aggressive mode. Would we rather he had some sort of motion sensing capability? That, plus audio sensing, seem reasonable (though you’ll still run into the incredibly fake feeling areas where you’ve gone one step further than he can see and you can clearly see him), but I wonder how much of a pain that would be to implement.
It would be obviously hard to implement, that’s why it haven’t seen that behavior in lots of other games. But i don’t think it would be impossible. It could be divided up in two areas:
-
Detection of suspect items.
-
Detection of “motion sensing”, as you say.
-
I have seen it in other games like the Dark Mod, where the AI will do a check to every object it does have in visual range (well, not every object, only the objects the game know are interactive by the player) to know if they are moved from their original positions or are in a different room than they should be. Obviously all this is data that has to be set up before hand in the level editor, in the properties of the objects. It should raise the suspect level of the AI and examine the surroundings of the object.
-
It’s basically to evade a specific cheap trick. The player could use the strength aug. to move a very big box toward some AI and because the box is always between the AI and the player, the AI could not see the player. A cheap way to solve the problem is just to make the object transparent for the AI if in the moment is being moved by player. Not very elegant solution, but hey, it’s just to evade this loophole in the AI system, we could say.
Prioritize the security level of various locations. Track the displacement and duration of each object.
Assume that an object displacing at a significant rate in a high-security location will eventually draw some attention.
I don’t think anyone here really, ACTUALLY, wants believable behavior from armed, trained guards. :)
DeepT
1590
I suppose there could be some simple AI put in. If an object has been moved, the AI will move it back. So the player starts to stack boxes while the patrolling guard isn’t around to create his cover. However when the patrolling guard comes around, he notices the stack of boxes and begins to un-stack them.
A further enchantment is that a guard who sees a moved object will call in a report, especially if it is a larger object (moving a coffee cup will not cause a reaction, but a large box will). Additional guards will show up and investigate while the original guard puts things back where they are supposed to be.
The more highly-trained they are, they easier to manipulate …
Blips
1592
And then they call the maids in to tidy the room up and dust :D
I look forward to seeing some realistic, highly-trained maid AI - and whether it can pass the white cursor test.
Timex
1594
Investigating something like the cardboard box fort requires the bot to have some mental model of the world around him, to the extent that he has expectations of what should be there, and what shouldn’t.
On a really simple level, to answer your question, you could perform the same kind of ray casting that you describe, but additionally keep track of what those rays are hitting… essentially, remember what’s in the environment around him. When that changes, trigger an investigation response.
Of course, this quickly becomes impractical when the bot is moving around instead of being stationary… or when other things which are ok are moving in and out of his field of vision. You really need to start tracking a more robust mental model of the world, more in lines with how humans do… When you’re in your office, you expect to see certain people wandering around… maybe even total strangers that fit into certain categories (like a guy in a UPS uniform, for instance). But certain things break your expectations, and that act of breaking your expectation (if we accept explanations of our neocortical functions put forth by people like Jeff Hawkins) trigger interest (or more strong responses like terror) in our minds. It’s one of the major mechanisms by which we learn new things, and don’t waste all of our processing on mundane crap.
Track the object, not the guard.
If he sees something which has had a high rate of movement, increase the likelihood he might investigate.
JZigish
1596
If you’re doing a proper weirdness-detection AI, you’d actually NOT want a more advanced mental model. The human brain is very good at detecting change and will immediately pick up a moving box. However, were you to time your movement of the box to when the guard was not looking at it, he would probably never notice that a (reasonably placed) box had moved several feet.
Anyway, to make that model work in a game you need sight cones or something to make it really obvious when a guard is looking at you. If I were architecting it I would make any movement of an object generate a certain amount of “visual noise”, that would increase with amount of movement, then do some sort of ray trace to the area of that visual noise to see how much of it is visible to the eye. Kind of treat it like footstep detection, but with a huge emphasis on directionality.
I think that would be fun. What I absolutely DON’T want out of my video game AI is some sort of “absurdity detector” that would know that a box shouldn’t be in that weird place to start with. That fights directly against being able to understand the rules of the game and veers into what the designers think is the "right’ solution.
Timex
1597
I think that would be fun. What I absolutely DON’T want out of my video game AI is some sort of “absurdity detector” that would know that a box shouldn’t be in that weird place to start with. That fights directly against being able to understand the rules of the game and veers into what the designers think is the "right’ solution.
But that could lead to some hilarious things…
Like arrange the boxes in the shape of a smiley face, and the guard recognizes the weirdness. :)
The problem with those kind of grand schemes isn’t that it inherently prevents the player from being able to understand the “rules of the game”… After all, we exist in the real world, with this kind of crazy complex stuff going on in all of our brains, and we somehow manage to work within the rules.
The problem is that with the AI solution, you’d almost certainly end up having some tremendously inferior implementation of that “grand solution”, and that failure would be what leads people to not being able to understand what the rules are.
But as the entities become more and more human in their behavior, then the player is able to fall back onto the same rules that they use to deal with other humans on a regular basis. You can know how the bot will react, because he’ll react like you would in that situation.
The problem arises from that grey area where the AI is complex enough to make it difficult to understand, but not complex enough to function like a human such that you can leverage your natural empathy to predict its activities.
The other example from the PCG review was using a box to creep up on a security camera. He carried it and then went into super cover mode every time the camera swung his way. I’m not sure if I like that more or less: cameras aren’t as smart as humans so maybe you can fool them, but on the other hand there are image processing algorithms to detect a new large object in view since the last image from that position.
It’s not like any of this nerd analysis matters. I’m ready to play the game. But when the developers say “oh gosh those blind enemies are so unrealistic” they’ve called attention to whatever they replace it with.
Zylon
1599
Nonetheless, it’s disappointing that they thought it was a smart move to gut light-based AI awareness from Deus Ex. Making a point of turning off the lights when sneaking through offices in DX just felt right.
Sure, but it’s less relevant in a world where you can get thermal imaging contacts so cheap online.
