Decline Thiaf Pre-Release Thread

It would be really fucking cool to have a system where guards don't care about footsteps in heavily travelled areas (other guard patrols, servant traffic), but will get suspicious about footsteps within locked rooms that are supposed to be empty, or un-patrolled hallways.
Oh, and what would be even more awesome, guards getting suspicious when there are no footsteps at all coming from a room that's supposed to be patrolled, and looking if the patrol is still there.

Oh how I want someone to actually make a good stealth game that tries to improve on Thief.

 

That and a stealth game should be light + geometry dependent, have true LOS. And modern games have it much more detailed than boxes of Thief 1 & 2. The more detailed objects, the further the draw distance and more of them you have in AI view that does LOS calculations - the harder the CPU load is. That's why it's either ridiculously tight corridors with 2-3 guards at once or 'Press X to crouch/stick to a box and become invisible'

 

Shouldn't be that hard - you could either flag the areas manually or do it automatically, based on pathing used and simply check area of origin when determining suspiciousness.

That's a bit trickier, but could be treated similarly to not seeing the patrol.

Still, my main peevee in wannabe stealth or hybrid games is that guards are oblivious to missing guards. This incentivizes clearing the levels by dismantling patrols guard-by-guard, bonking them upside the head (or backstabing) and shoving them into closets and dumpsters instead of avoidance.

In a good stealth implementation eliminating a guard shouldn't be a question of if, but when alarms go up and shit hits the fan.

 

You can always use simplified boxes for determining LOS and space partitioning to determine what not to check against in given spot.

Well, I'm of an opinion that since the AI is going to have suspicious/unconfirmed state in which it approaches the trigger and investigates by looking around, randomly searching or something, then it might be just as well reused for other suspicious occurences than glimpisng the player.

Similarly, missing or altered critical objects, could trigger alarm in the same manner as seeing a corpse or confirming player sighting.

As for the patrols, much depends on how they are implemented in the context of your levels. If they are designed to meet with great, pre-set regularity, than you can just make a number of missed meetings trigger alert and search.

If you want more freedom when placing the patrols, then you may need something more robust, like what I suggested.

 

I think Thief 1 had guards comment on banners you sliced apart with your sword in Bafford's manor. It was just some amusing comment, no alertness increase, but it was there.

 

Yeah and that's why you get such a shitty stealth in modern games.

 

Thing is I'm pretty sure AI in Thief is _cheating_. It's not 'searching randomly' at all. The 'event bus' of all the AI's always know where the player is - what they're looking for is for the player to fuck up in their PoV. Then they go into the 'suspicious' state, but they're not searching like a normal person, just slowly going in the general direction of the player albeit with some random changes of direction that are _always_ corrected as long as the search timer isn't over. In thief, there are no situations like 'he must have gone that way', except when they're not really looking for the player, but going to a body the player has killed etc.

 

I think we shouldn't forget about gameplay either. You can implement a lot of stuff into the game to make stealth more realistic or improve AI, but go too far and the game might be too difficult for most people to enjoy. Of course, there's always the masochist crowd, but even most old-skool gamers have their limits...

E.g. if being spotted just once for the fraction of a second by some guard will rouse the entire security force of the level into a permanent alert state, you can pretty much reload already.

 

BTW, the darkmod implements that thing with missing objects as a general feature of the editor, not a script stim.

 

I'm 99% sure that most modern games use the collision geometry for line of sight. This leads to those dumb situations where you can be hidden behind an object that clearly has holes in it, or an object much smaller than the player character. I immediately noticed this in the new Hitman but it also applies to the new Deus Ex to a degree. It's why so many games with cover systems have such easy stealth - there is very little ambiguity as to whether you are hidden or not.

 

You can throw something and they're react to where the sound is made (or shoot an arrow, whatever).

 

The arrow is just taking the place of the player on the search behaviour. Really, why should a game AI, any game AI, that has complete access to up-to-date info about the positions of all the objects in the world waste time searching for crap? It just simulates looking for things (obvious exceptions like fog-of-war strategy games are obvious).
normally, on computing: push model event notification instead of 'pull' model, where the sources of perturbation of the 'normal' state notify the interested parties instead of them scanning and noticing alterations.

That isn't to say the 'missing objects' feature couldn't be coded like this, but it's almost exactly like a script stim, not really a agent with imperfect knowledge.

The darkmod uses more than one LoS line for this IIRC. Less error prone, but more costly. Of course, they only do this when the player is there to be seen yadayada...

 

I'm pretty sure that SCO is right, although I have never had it confirmed. There are just too many instances where I have made a sound, escaped silently into a room far away, and then the guard who heard me from another hallway--without ever seeing me--has crept straight into the room that I entered. He will even walk to the proper corner of the room. They could just have Daredevil's super hearing, though, which would explain their poor eyesight.

 

But i like being able to take my clothes off and run around the mansion naked after all the guards are knocked out.

 

http://www.gameinformer.com/b/features/archive/2013/03/08/the-goals-and-inspirations-for-thief.aspx

Only the best

 

By retards for retards.

 

JarlFrank The FM Off The Record has heavy AI scripts with semi-randomized patrols and they react to missing stuff, open doors and other things. It's pretty impressive the first time you play it when you're used to the usual Thief AI.
TDM also has an improved AI that notices ropes and other little things we don't usually think about when we're used to Thief, and chances are they'll come investigate. I found this to be neat.

 

Way to misunderstand the issue.

Small objects generally have small colisions, so if you can hide behind a teacup, simplified colision envelopes are the least of your problems - the real issue is that player is treated as mathematical point for the purpose of detection.



Using arbitrarily detailed colision envelopes won't fix that.

 

You're right of course - some degree of full body simulation is required to have convincing stealth, where the space the player occupies is treated as a set of areas attached to limbs etc., rather than one single point which can be easily obfuscated by even a small object. But both problems described can result in poor stealth gameplay. On modern computers neither should be a particular performance concern, though if you are pushing last-gen stuff to the limit I can see how maybe you could save some CPU cycles on AI by taking such shortcuts.

Ultimately what matters is whether the game design necessitates the effort involved - if you treat the character in a stealth game the same way as you do in a bullet hell game, then you're going to be negatively impacting your gameplay experience. A stealth game similarly won't require the exact same accuracy in other areas as other types of games, so it's not like you can't make up for increased resource use more complex game logic requires by making cuts elsewhere.

 

I would imagine that any engine that can do dynamic lighting could, in principle, use those mechanics to calculate LoS.
If you can calculate shadows coming from arbitrary lightsources, wouldn't you be able to use the same technique for LoS as well?

 

Dynamic lighting can be very imprecise because most people don't notice if a ray of light fails to get through a notch in the side of a crate or if 3 rays get through instead of 1. That's the main problem I can think of.

 

http://www.gameinformer.com/b/featu...-thief-39-s-long-walk-out-of-the-shadows.aspx

This is sounding slightly better:

- Discussing emphasis of vertical navigation, implies grappling hook may be widely usable.

- Did some prototypes in 3rd person, but it's a very first-person game.

- Discusses most of the team left only a year ago and the game was in a completely shitty muddled state.

- They cut a lot of the previous work out in the past year.

Not sure how much of this is reactionary PR to the fact that literally every major gaming forum has been ripping them a new asshole, however.

 

The first comment made me lose my shit.

Oh yes, please do.
Other than that nothing really reassuring except the fact that TEEF is mostly a first person game.

 

The implication that this version of the game has only been in development for a year and the mention of past teams leaving the project is surprisingly candid for a PR interview. Eidos really is shitting the bed for marketing here.

 

No need for anything complex, players can't tell because normally there is no way to move the limbs in ways that would expose them but not the rest of the body - i think Garrett is essentially a cylinder in the the Dark Engine for instance.