Slashdot Mirror
A Look At the AI of Empire: Total War and F.E.A.R. 2
mr_sifter writes "The newly released Empire: Total War and F.E.A.R. 2 have both been praised for their excellent AI. In this feature, Bit-Tech talks to the developers behind these games about how they handled the challenges of creating Empire's armies of thousands of AI soldiers and F.E.A.R. 2's aggressive teams of military operatives. The discussion also talks about how game AI is 'smoke and mirrors' compared to research AI, and looks at the difficulty of improving the quality of game AI."
We talked about F.E.A.R. 2's engine and AI back in December as well.
I would revel at the day when the AI is more human like which would tremble upon sighting my avatar, run to the ends of the map at the sound of my bullets, curse me, and log out.
It's a pity that Empire: Total War lacks multi-core support, as well as having severe loading screen issues.
If FEAR 2 dropped the horror and just made it a tactical FPS, I'd have no complaints for that one.
AI is all well and good, but it is only one part of what makes a game. At least, one part of what makes a game that requires AI.
Hi, :-(.
i love the Total War Series. But the AI in ETW is a complete disappointment. I see no enhancement compared to MTW2. The opposite is true, there are so many AI bugs that battles agains the AI are close to pointless. I slaughtered enemys at the border of the battlefield, standing with theit back towards me. I micromanage my own units because grouping tends to produce strange results. Due to this, battles are a lot more point&click-work currently. I'm waiting for a patch
Sincerely yours, Martin
Not everyone is convinced that GPGPU is the best way of processing AI though. Bethesdaâ(TM)s Jean-Sylvere Simonet notes that "we might be able to take advantage of parallel architectures, but not for everything. You could probably speed up some individual parts of the decision process, such as replacing your AI search with a brute-force GPU approach, or running a pattern detection algorithm". However, Simonet also points out that "most AI processing is very sequential and usually requires a lot of data.
"For an NPC to decide on its next action, it will usually have to query the world for a tonne of information, and most of that information is conditional on a previous query result. For that reason, fewer processors that are more versatile, such as the SPEs in the PlayStation 3â(TM)s Cell chip, are ideal".
Power of Cell etc etc.
The AI in Fear 2 was terrible! Well, if it was good, I didn't notice because it was so god damned easy to play that it was more a case of "Hi, *slomo* bye." .. firefights were so short that if there was amazing AI under the hood, I completely missed it.
On the parts where I didn't use slowmo, again, nothing remarkable, and I'm not exactly an amazing FPS player either.
When has a game not claimed it has ultra-realistic AI only to end up having blatantly unrealistic AI?
I don't think you can even have ultra-realistic AI and keep the game fun in most cases anyway, if the AI was in fact ultra-realistic then they'd all just move to the same part of the map, wait until you come along and ambush you from every direction leaving you no chance whatsoever after all.
AI has got better through the years, it's started using cover and that sort of thing and the odd game has managed to have one set of AI put down supressive fire on you whilst another flanks you to an extent, but still not brilliantly. Even this time of AI is meaningless though, because there's few games that are realistic enough for you to die in one shot for example even if someone is laying down supressive fire on you you can still stick your head up and take them out even if you do take a few bullets.
So there's a distinction to make too, I don't think AI needs to be realistic else we get silly scenarios where the player has no chance but it can be a lot better whilst keeping the game fun. So far not much has managed to achieve ultra-intelligent AI whilst keeping the game fun. I guess it begs the question as to whether ultra-intelligent AI and fun AI are mutually exclusive things and you simply can't have both.
STALKER and Clearsky were both terrible with grenades as well, they would throw them so far and hit you in the head with them.
I guess in a post-apox world people have nothing better to do than sit around and throw stuff that must be why they are so good with grenades.
As a side note
Empire: Total War uses online activation (through Steam), so if you buy this and you don't have an Internet connection all the time in your gaming PC, or you upgrade components on it or you upgrade your OS or any other arbitrary condition (which can change at any time at the whim of one of the suits at Creative Assembly) then you've just threw away a nice chunk of your fun money.
Also forget about re-selling them game or giving it to a friend once you've became bored with it (or if you find out that you don't like that kind of games)
Funny part is, if this just had a key check or CD check instead of the "can be remotely disabled at any time" style of DRM, this would be one of the few games that I would buy this year.
Creative Assembly's Richard Bull notes that "there's still this disturbing mindset among programmers, particularly game programmers, that if the AI is taking any kind of considerable chunk of time, that's a really bad thing. It's only just getting to the stage now where people regard it as important enough to deserve this chunk of time in a game. If your graphics rendering is taking up 50 percent of your CPU time it's like 'well, never mind, it looks great', but if you try to tell people that you have this really intelligent decision-making system that's taking up 30 percent of the CPU time, they'll say 'you obviously don't know what you're doing, it's badly programmed' and so on".
I'd bet every strategy gamer out there would take a better AI over seeing blades of grass, but alas, AI doesn't draw marketing like grass blowing in the wind. :( FWIW, reviews are already criticizing Empire a bit for what has been a problem in past versions: all focus on the tactical AI (much needed, no doubt), but the campaign AI is still problematic. In this sense, CA has a heck of a task developing two quite different AIs. It is hard enough getting just one of them in respectable shape!
I'm about to conduct a modest study of NPCs in games; and came across this article.
:)
Having read several texts on narratives in games and played a fair share of games - many of which were well written crpgs; it seems to me that games that rely heavily on authoring player-character behavior (through non-interactive scenes - such as FMVs) - create a sort of pseudo-NPC of your player character - leaving you, as a player, passive and unimmersed in large parts of the game. This does not necessarily have to be a bad thing; but striving towards immersion and the illusion of choice and making a difference, I believe the plot is to be wrought through the NPCs - thus, I've turned to the makings of NPCs.
It's a fairly broad area; and only part of it relates to AI and programmed autonomy.
I may be looking into some of the following areas: AI/Autonomy, forms of interaction (dialogue, cooperation & other forms), Plot: Function & Relation dynamics, content (scripts, models/animation, static vs. dynamic, voice acting), extra-game presence ( npc's roles in advertising, IPs, translation between media, reception) and player reception.
I would appreciate:
- Any ideas for other NPC relevant areas
- References to articles or material regarding NPC's or AI in games
- Accounts of any particular memorable experience you may have had with NPC's in games: ( What game, What made it unique? )
Thank you
(P.s. Yes - my current scope is very broad, and I look to limit it - but I'm still in an early exploratory phase.)
- Mad, ingenous - they've both left you puzzled -
It isn't as though humans don't do stupid things, too. Even if I could make a game with a perfect AI, would people want to play it? Supposing the AI knew all the rules and never fucked up, what fun would it be? It'd always win since it was perfect.
While I am certainly a person for better AI, by better I mean "more human." I want an AI that can act like a human player of varying skills. Part of that WILL be making mistakes.
I can't decide if this is +1 informative or +1 funny. All I know is I wish I had mod points.
This is probably one of the best articles I've read in Slashdot in a while. Yes, game AI may be a couple steps below true AI, but it's certainly a step above the problem-solving your average software developer is used to, so it picked my interest.
Thing is, I'd like to learn more about this and perhaps try my hand at writing a basic AI for a game, but I have no clue on how or where to start. Anyone has any recommendations?
No problem is insoluble in all conceivable circumstances.