Physics and AI

[Sonic]

Is anyone here satisfied with the current level of physics and AI seen in games? Any room for improvement in current consoles?

My own thoughts on the subject are that if devs spent a little extra time on physics and AI then games would be that much better to play. I'm really tired of seeing so many scripted events in games that it's just getting boring and too predictable to play many games. If I can pick up patterns after two times with a game than something is seriously wrong with either me or some of the games that I play. It just seems like today's games are lacking any sort of AI that's above the level seen in the 1st party games. Some games do deserve merit, but this really only acounts for about 15% of the total games out there to be bought.

Physics is a completely different story. I'm satisified with the current level of physics in games but I still think things could be done a little better.

Anybody else have thoughts on the subject?

 

[V3]

I think the AI and Physics are find as they are. But improvements are always welcome.

I want to see more of one hit, die type of game of the old. No more life bar thingy.

 

[London Geezer]

I think the AI and Physics are find as they are. But improvements are always welcome.

I want to see more of one hit, die type of game of the old. No more life bar thingy.

u are joking aren't u....

saying that physics and AI are fine as they are is just like saying Graphics and Sound are fine as they are.

with the only difference that, as we have seen in the past, games with GOOD AI and Physics are the best in their genres, even with many graphical flaws... take a look at ICO, that game is pretty much ONLY AI and Physics, with POOR graphics (apart from some original and neat effects) and OK sound... and it's considered one of the best games in this generation.... MGS2's Physics and AI systems are what gives the game its personality, not the graphics, which are unarguably less than stellar.

(DISCLAIMER)

i consider all the neat particle and lighting effects as PHYSICS, that's why i'm saying that BTW

(END OF DISCLAIMER)

in the future, games with superior AI and Physics will be the pinnacle of the next generations, not only the graphics. remember, the better the physics and AI are the more interactive (and therefore playable and enjoyable) games are.

graphics only pushes u so far.... if the game doesn't FEEL right to play, people get bored with it...

 

[Ug Lee]

My thoughts are that there's always room for improvement.

 

[Fafalada]

take a look at ICO, that game is pretty much ONLY AI and Physics, with POOR graphics (apart from some original and neat effects)

:?
I suppose if it did stuff like """Gritty! Deep shadows! Photorealism! Anisotropic normal Blinn maps with spline vectors! Whatever Tim Sweeny's doing!""" (to borrow from Tom Forsyth's depiction of publishers ) it might have avoided the adjective POOR eh :?
Either way, to each his own I guess. Personally I found the game looking great though.

Anyway, physics on current hw is still more about taking shortcuts then anything else. Still plenty of room for things to get a lot better eventually.

 

[London Geezer]

take a look at ICO, that game is pretty much ONLY AI and Physics, with POOR graphics (apart from some original and neat effects)

:?
I suppose if it did stuff like """Gritty! Deep shadows! Photorealism! Anisotropic normal Blinn maps with spline vectors! Whatever Tim Sweeny's doing!""" (to borrow from Tom Forsyth's depiction of publishers ) it might have avoided the adjective POOR eh :?
Either way, to each his own I guess. Personally I found the game looking great though.

Anyway, physics on current hw is still more about taking shortcuts then anything else. Still plenty of room for things to get a lot better eventually.

didn't mean to say something against it, it's just that all the nice things in the graphics are in the realm of Physics for me, the water, the trees and all that for me is all Physics...

besides, one room in the game doesn't even have bilinear filtered textures. and that was painfull to look at, especially since bilinear filtering on anything after N64 is pretty much free..... actually take the *pretty much* away

 

[Ozymandis]

It seems to me that improving physics is kind of dangerous. As far as improving interactivity, I'm all for it. But IMO making physics too realistic can be very detrimental to gameplay. I'm more concerned about having simple, consistent physics that don't interfere with gameplay than "improved" physics. We don't want more Trespassers or Wrecklesses 8)

Better AI would be nice. The stuff in games today is pretty basic, and it all becomes predictable after a while.


What room in ICO didn't have bilinear filtering?

 

[maskrider]

take a look at ICO, that game is pretty much ONLY AI and Physics, with POOR graphics (apart from some original and neat effects) and OK sound... and it's considered one of the best games in this generation

ICO's graphics is one of the best overall (the atmosphere is just perfect, although it is running in a low resolution), for the sound department, it doesn't need exaggarated sound effects to support the atmosphere the game is in, it is a very peaceful and yet dangerous adventure of ICO and Yorda. In another words, the sound effects of ICO are perfect for the game.

It is the most beautiful and unforgettable game to me.

 

[London Geezer]

It seems to me that improving physics is kind of dangerous. As far as improving interactivity, I'm all for it. But IMO making physics too realistic can be very detrimental to gameplay. I'm more concerned about having simple, consistent physics that don't interfere with gameplay than "improved" physics. We don't want more Trespassers or Wrecklesses 8)

Better AI would be nice. The stuff in games today is pretty basic, and it all becomes predictable after a while.


What room in ICO didn't have bilinear filtering?

Can't really remeber to be honest, havent played it in a long time... i actually thought there was something wrong with my PS2 when i first saw it.... things like that are BAD this generation....

as i said, ICO in my opinion is unforgettable thank to all those little touches that i consider Physics... the light effects, the water, the trees, the rain.... call it graphics if u want, but to me it's all animation and i would put animation under physics/AI rather than under Graphics....

 

[BoddoZerg]

"Physics and AI" is very subjective, depending on what you mean by "physics and AI".

If you're going to include stuff like animation and lighting and particles as physics then of course "physics" is incredibly important.

However, when most people talk about physics, they usually aren't thinking about mundane stuff like rain and lighting, they're usually referring to the "ragdoll" physics first seen in Hitman and greatly improved in Unreal 2. If you're going to count water and trees as physics, you might as well count every polygon of every model as "physics" because of skeletal animation.

Personally, I think physics are a natural extension of improved graphics. You can model a soldier to an incredibly high level of detail, and have extremely realistic mo-capped traditional animations, but if that soldier clips straight through the floor when he dies, it will look very bad. DOOM3's "boxes bouncing around the room" physics may just be a gimmick, but it's a really cool gimmick that adds a lot to the immersiveness.

On the other hand, I am more skeptical about whether AI has a lot of room to improve. As Bungie said in the Halo2 interview, the AI in Halo was so sophisticated that no one noticed it. Some of the most memorable AI of any game comes from Half-Life, a game that ran on 200 MHz CPUs and had very little "real" AI, as most events were pre-scripted. The problem with AI is that, unlike graphics and physics, simply throwing more CPU cycles at the problem does not guarantee an improvement. Many people would call the bot AI in Unreal Tournament "better" than that of UT2k3, despite the latter having 10 times higher system requirements. The sophisticated unit AI in the initial release of Battle Realms was a major source of complaints; they actually had to remove the AI so that players could have more direct control over their units.

The only place where I think AI could improve a lot with better CPUs is stealth games. Right now, all stealth games have the same unrealistic "as long as you're in darkness, you're invisible". Even if you cast a shadow across a light that enemies can clearly see, they will never notice you. This is one of those problems that can be solved by throwing CPU cycles at it, so I expect it to improve substantially in the next few years. Other than that, though, AI is extremely difficult to improve, and is much more of an artistic matter than one of CPU power.

 

[Tagrineth]

Heh

I suddenly think of UT2003...

Remember how they promised watery environs?

Well, there aren't any.

You know why?

The Karma physics engine is so precise, walking through water would result in single-digit FPS on a ~3GHz P4.

 

[DeanoC]

Heh

I suddenly think of UT2003...

Remember how they promised watery environs?

Well, there aren't any.

You know why?

The Karma physics engine is so precise, walking through water would result in single-digit FPS on a ~3GHz P4.

You haven't use Karma have you "Damn you Karma, your too precise" is not a term you hear escape my lips to often

You would not believe how bad both major middleware physics engines are at doing physics (unless your realm of physics involves things sinking through the floor and objects disappearing fairly often). One thing that is not realised by most people is that both physics don't like any situations where you have ratios of more than a hundred (i.e. don't have a 1KG object hit a 100KG object, bad things will happen). Classic peice of advice is don't even consider physically simulating small things (like grenades!), it won't work.

BTW Karma's water sim in Unreal Warfare (the engine that 2K3 uses) is quite fast, its the graphics effects that start to die. Its also only designed for small pools, you huge lakes or rivers are not in its remit.

Physics has a long long way to go, I reckon we will get through several generations of consoles before physics works as advertised. It takes a shitload of tweaking to get physics working at the moment.

 

[Fafalada]

(unless your realm of physics involves things sinking through the floor and objects disappearing fairly often).

Depends on the type of game. It worked great for Wreckless, helping to reduce slowdowns.

Anyway, it's not like current realtime dynamics can really afford much precision to begin with. A lot of stuff (particularly collision detection and sparse matrix solvers) is still way too expensive to afford at higher update frequencies, and god forbid you do it on large scale with a lot of objects too.

 

[Panajev2001a]

That is when you look at 1 TFLOPS and 1 Tera OPS of next-generation machines and you smile

 

[DeanoC]

Yer I know, I'm being a bit harsh

When it works its lovely... but I just hate it when it falls apart.

I have exploding things, that are really hard to control (size and speed is based on damage etc). The amount of times I get "Deano its broken again", then when I debug it, its because they made a small heavy object (usually enough mass for a small neutron star ) hit a normal object. I also got a lovely quick sand effect with my object just sinking through the floor.

Top of my list of all things game development would be enough processing time/precision to have a robust physics sim. Scary PS3 architecture won't bother me if we can get robust physics.

Lots of object over a large scale, wouldn't dream of it (only 100-200 small object (planks of wood) over a couple of KM square) Asking for trouble really...

 

[DeanoC]

That is when you look at 1 TFLOPS and 1 Tera OPS of next-generation machines and you smile

Not if you don't have enough precision....

Floats are not that good when you solve large matrices....

 

[ERP]

That is when you look at 1 TFLOPS and 1 Tera OPS of next-generation machines and you smile

Not if you don't have enough precision....

Floats are not that good when you solve large matrices....

While I largely aagree and physics is one of the few areas where precision is important. The most prevalent issue in physics engines are due to poor numeric analysis. Realistically your never goind to solve the balance a planet on a pin problem, but most implementations I've seen, make the problems much worse than they need to be.

Numeric analysis is just woefully misunderstood. It takes time and it just hasn't been necessary for a lot of what's been done in games since the old fixed point days.

 

[DeanoC]

That is when you look at 1 TFLOPS and 1 Tera OPS of next-generation machines and you smile

Not if you don't have enough precision....

Floats are not that good when you solve large matrices....

While I largely aagree and physics is one of the few areas where precision is important. The most prevalent issue in physics engines are due to poor numeric analysis. Realistically your never goind to solve the balance a planet on a pin problem, but most implementations I've seen, make the problems much worse than they need to be.

Numeric analysis is just woefully misunderstood. It takes time and it just hasn't been necessary for a lot of what's been done in games since the old fixed point days.

I've done some numeric analysis and I'm left with the conclusion that small things in a big world aint gonna fly with floats. A fast bullet vs building just gets lost in the noise of a float LCP solver with any kind of 'real' units. Of course a change of technique might make a real big change (I thought about interval arithmetic or even partial symbolic solver's). Ironically the more dynamic the enviroment (the whole point of physics ) the harder the numeric analysis gets.

Its not that I don't think were going to have 'cool' physics, its the 'plug and play' physics I doubt. I've got the feeling that people like us are going to spend a lot of our lives fighting the lack of precision.

 

[Panajev2001a]

The way I see it is this...

1) You have with Cell a lot of Integer and FP power ( you are right, 1 TFLOPS is a number achieved using floats... ).

2) You have lots of FP power and bandwidth that can be used for HOS, T&L and that REYES-like approach that was discussed in the other thread...

If I can digress I'd like to say I agree that there is historical basis on that approach... after PlayStation 2 was released, they started with the GScube project(s) and the feedback they got seemed to be positive... if you say that great Vertex Power and bandwidth would be key to this kind of approach ( micro-polygons ) I see the way Cell has evolved ( again looking at what we knew IBM was working on before [Cellular computing, BlueGene, etc...] ) as a sign that the road you mentioned is a viable one.

3) You have power to dedicate to AI routines

What does this mean to the physics programmers ? Well, frame time is a fixed quantity ( ~ ) and if you can afford pretty graphics and fast yet "smart/complex" A.I. routines for a relatively small percentage of total frame-time then you will have more time and resources to dedicate to physics...

 

[Saem]

Out of curiosity, would one course --first numerical analysis course at uni-- in numerical analysis be sufficient for modeling various things in physics engines or do you need to go beyond that and take higher level courses?

I'm curious because I'm planning out what I need if I go into game engine programming.