Springheel on 2/5/2007 at 14:28
Quote:
Is it possible to breathe some life in them by implementing time schedule (not too strict) - occationally they would go to sleep, eat etc.
As Spar said, we're implementing something like this, though not as complex as Gothic or Oblivion. The idea is that certain objects can have stims applied to them that 'invite' any passing AI to interact with them. The chance of an AI responding would be random (and can be set by the mapper, along with how far they travel). So an AI won't stop and hold their hands out to a fire every single time they pass it, like TDS, but they will stop there occasionally. Likewise, you could theoretically have a bed 'invite' AI to lie down on it, or a bottle invite an AI to pick it up and drink from it. This will give the illusion of more believable AI, but it's not without a price--it becomes much harder to synchronize patrols (as a mapper) or time precisely when a guard will return (if you're the player).
For that reason, it will probably be up to the mappers to set these stims on objects, rather than have them work by default.
242 on 2/5/2007 at 16:04
Optional stims on objects is great, but in my opinion it would be better if AI demonstrated more lifelike behaviour by default. You just throw NPC in an environment and it behave itself more randomly/lifelike than in T1/2. After all no living creature patrols with second to second accuracy in real life. If FM designer needs synchronization then he/she specifically set it, i.e. I propose inverse scheme because patrol synchrnization will be needed probably not that often. Plus, majority of designers will just be to lazy to simulate time schedules by themselves...
jay pettitt on 2/5/2007 at 16:46
Lifelike behavior would be hellish.
First off, when you add complex behavior to an AI the opportunities for them to screw up and look stupid increase exponentially. AI break the suspension of disbelief when they look stupid, even if what they are doing is fairly clever and complex in AI terms. Where as if AI are consistantly stupid, it's very much easier to live with. Witness the difference between AI in Thief 1 and 2 and the AI in Thief 3.
Secondly, it's a game. Like Springheel said, erratic behavior is hard to predict. Artificial Intelligence isn't worth dot if it's not fun to play against. Truth is, if AI do something unpredictable which cuases me to loose I'll get huffy and accuse the game of not playing by 'the rules' and switch off.
New Horizon on 2/5/2007 at 16:48
Quote Posted by 242
I propose inverse scheme because patrol synchrnization will be needed probably not that often. Plus, majority of designers will just be to lazy to simulate time schedules by themselves...
Taking the feedback we get from testers into consideration, I believe it would be the exact opposite. A Thief styled game requires the mapper to control how they want a particular AI to function first and foremost. The map and patrols needs to be built up, not down. Once a mapper is satisfied that the AI are functioning well at their default level, then they can start adding details to make the AI in certain 'non-critical' areas, less predictable. This doesn't mean the more rigid AI can't be complex, the mapper can setup numerous patrol routes...with varying chances of being chosen. There is no need to hard code these behaviors by default...scripting and stim invitations are fairly easy to setup, and will add lots of diversity when used creatively.
sparhawk on 2/5/2007 at 16:50
Quote Posted by 242
Optional stims on objects is great, but in my opinion it would be better if AI demonstrated more lifelike behaviour by default. You just throw NPC in an environment and it behave itself more randomly/lifelike than in T1/2.
That would be a nice feature, yes, but unless we get some super computing power, it will be very unlikely. :)
Quote:
After all no living creature patrols with second to second accuracy in real life. If FM designer needs synchronization then he/she specifically set it, i.e. I propose inverse scheme because patrol synchrnization will be needed probably not that often. Plus, majority of designers will just be to lazy to simulate time schedules by themselves...
The point is that we are doing a toolset to create FMs. As such there is not really a notion what constitues a realistic action. As an example. A real person might get tired and sit down on a chair. What is chair? Can you tell me the precise definition of a chair, so that we can detect it and use for AI? You can't. That's why it is the burden of the mapper, to apply the AI, supply objects and give them meaningfull interactions within the context of the map.
New Horizon on 2/5/2007 at 19:13
In addition...I think this will make for some pretty unpredictable AI, as I'm sure no two mappers will set the AI to react to the exact same things. For a game like this, mappers really need to control how the AI will behave...and as Spar said, it would be pretty costly to make such AI.
242 on 9/5/2007 at 11:33
Will it be possible to create maps of comparable size/complexity as Assassins, Ambush, LoTP, CL with Darkmod or the hope isn't realistic and they have to be devided on smaller chunks like T3 ones?
New Horizon on 9/5/2007 at 12:19
Quote Posted by 242
Will it be possible to create maps of comparable size/complexity as Assassins, Ambush, LoTP, CL with Darkmod?
Yes. They will require proper optimization for best performance of course, but these types of levels are attainable. There are few hard limits with the D3 engine, unlike the T3 engine. When the D3 engine goes open source, it can also be further optimized/ enhanced.
Yandros on 9/5/2007 at 14:17
That's some of the best news I've heard in a while. I look forward to working with TDM in the future.
codereader on 9/5/2007 at 14:24
Rest assured, Dark Mod maps can exceed the size of TDS maps by far. Angua is currently working on a map that is already comparable to (when not exceeding) the size of the original Bonehoard.
