The GT5 expectation thread (including preview titles)*

[Mintmaster]

No, nothing YOU describe involves any of these. You simply state that x is sufficient to obtain convincing realism, and I state that y is required to obtian convincing realism. We disagree in that regard. There have been several years that 95% of my considerable amount of gametime was devoted to Gran Turismo and I come from a family of car-nuts, so it may well be that I am more demanding that way though.

It has nothing to do with being a car nut or not. It has to do with understanding how physics work in a car sim, and understanding what can tax a CPU.

You said y is required, and I'm telling you that y doesn't need any graph traversal, searching, iteration, etc. I have no problem with accepting that those factors you mentioned are needed for realism. Nonetheless, you haven't given me one single example of a computationally intensive task.

Yes, but the whole discussion is moot when you think you can separate that bit from the 'rest'.

If you can write a car simulator that knows when a wheel is off the track, then it is locating the surface parameters. Clearly the strength of PD is not knowing where a wheel is on the track (every driving sim can do that), but what to do with the information you have about the position.

When I say driving physics, I'm referring to everything that makes Gran Turismo handle differently from, say, Ridge Racer. Both games do collision detection (although collision reaction is handled differently), do rendering, do AI, know when a wheel is off the track, how far the ground is, etc.

The driving physics are what determine the character of the car and how it reacts to the road and user input. It's what separate the men from the boys. It's very hard to model, but once you do, it's easy to calculate.

So forward motion of the different weighted and 3d positioned objects do not matter here, you say, and I disagree.

I have no idea what you're trying to say here. A car is a rigid body. All you need is the mass, centre of gravity, and moment of inertia tensor as parameters. It doesn't matter how the mass is distributed around the car. This is indisputable physics.

I know you on this forum and on the one hand I appreciate having a discussion with a fellow programmer who is working in a field I would love to be working in, but on the other hand it is a shame that you are so convinced of your own viewpoints that the discussion will probably remain one-way for a long time yet.

I am not working in the field, actually. I did the car simulation as a hobby, and the physics engine was for a startup that has now disbanded. However, I have more than enough experience from that as well as a very good understanding of physics (if you or phat want to know how good, PM me) to be very confident in these claims.

This is not just opinion that I'm spewing. It is fact based on first-hand experience.

 

[Arwin]

It has nothing to do with being a car nut or not. It has to do with understanding how physics work in a car sim, and understanding what can tax a CPU.

Understanding how physics work in a car sim and understanding how cars work are very much related. I learnt most of the things I know about cars not primarily because I love cars, but because I love car sims.

You said y is required, and I'm telling you that y doesn't need any graph traversal, searching, iteration, etc.

Because those are the only things that tax 'a' CPU, right?

My problem is how you gloss over the details of knowing that a car is 'on the track.' What matters here for instance is the precision of the collision detection. in real life, the track is never a completely flat surface, and you can add as much resolution in your collision detection between the surface of the wheel and the surface of the road as your computer can handle at those speeds. There are bumps in the road, there are curbs on the track, there are off-track surfaces, two wheels can be on grass while the other two are on the track, etc. For the collision detection to work properly, you also have to calculate how the softness of the tire influences the actual surface contact, not to mention that pressure division over that surface is going to be a complex bit of calculation in itself.

Sure, some of these things you should be able to handle fairly easily. But a lot of them not so much. I will agree with you on the basics, that many variables are in principle easily modelled. I know fairly well actually which variables were used in GT3 and GT4 Prologue, as a few years ago I wrote a fairly widely used replay manager that parsed the binary save files for among others settings information, and here you also found a lot of the basic parameters that PD uses internally.

I have no idea what you're trying to say here. A car is a rigid body. All you need is the mass, centre of gravity, and moment of inertia tensor as parameters. It doesn't matter how the mass is distributed around the car. This is indisputable physics.

Even the rigidity of the body of a car is variable in GT. It's one of the parts in GT that actually wears and it's rather expensive to bring it back to its original state (in the game anyway ... ).

As far as physics go, my knowledge of them is limited to what my general curiosity taught me (and a few years of very basic stuff in school). But I understand what you mean with the centre of gravity. You are saying that if I put a 75kg driver on the right or the left side of the car, its centre of gravity just shifts to a different point in space, but it remains 1 point, no matter how many people I put in the car. So whereever the weight of objects is a factor in other calculations, you only need to work with the single variable 'centre of gravity'.

I am not working in the field, actually. I did the car simulation as a hobby, and the physics engine was for a startup that has now disbanded. However, I have more than enough experience from that as well as a very good understanding of physics (if you or phat want to know how good, PM me) to be very confident in these claims.

This is not just opinion that I'm spewing. It is fact based on first-hand experience.

I understand you. But I still have a sneaking suspicion that in your enthusiasm about how easy you discovered it really was in your project, you are too eager to overlook at which point things become complex.

Your division in work that taxes the CPU and work that doesn't is alien to me. I've been trying to get into the whole 'building your application around the data' thing, and so far it is been working out for me quite well. Whenever I try to build something that is fast, you simply make sure you don't have to search or recurse. (Iteration isn't so evil imho)

Hopefully I get a chance to play around with some realtime stuff, but for now I'll have to stick to designing flexible and easily maintainable user interfaces.

 

[Shifty Geezer]

I have nothing to contribute to this thread, other than I stumbled across a vid about modelling physics, specifically cars, by a chappy who worked on Forza.

http://www.gametrailers.com/player/usermovies/74862.html : 35 mins long

Here's part 2
http://www.gametrailers.com/player/usermovies/74906.html

 

[Mintmaster]

Because those are the only things that tax 'a' CPU, right?

Yup, because we aren't talking about thousands of cars on the track, and PD is not going to implement any calculation in a stupid way. You will not occupy millions of CPU cycles (and remember that Cell has 4 orders of magnitude more than that) without long (i.e. n>100) loops, recursion, or algorithms tougher than O(n^2). You certainly haven't mentioned anything yet that is taxing.

My problem is how you gloss over the details of knowing that a car is 'on the track.' What matters here for instance is the precision of the collision detection. in real life, the track is never a completely flat surface, and you can add as much resolution in your collision detection between the surface of the wheel and the surface of the road as your computer can handle at those speeds. There are bumps in the road, there are curbs on the track, there are off-track surfaces, two wheels can be on grass while the other two are on the track, etc. For the collision detection to work properly, you also have to calculate how the softness of the tire influences the actual surface contact, not to mention that pressure division over that surface is going to be a complex bit of calculation in itself.

All the collision stuff you mentioned is done by nearly all current games. How do you think they make a different noise when going over the curbing, or knock signs when the car goes off track, or kick up dirt/leaves only on the wheels that are off, or avoid drawing tires through or above the ground, etc?

If you want a curved surface, you simply use one of the many ways available to deal with that. Nobody is going to use implicit surfaces (e.g. metaballs) to model a track surface, so the explicit definition is again just a simple formula. Yet again you are confusing modelling with simulation. Tire softness and it's impact on traction is again just a formula arising from a model. It's tons of work for PD, but very easy for any CPU.

Sure, some of these things you should be able to handle fairly easily. But a lot of them not so much.

You still haven't given me a single example of the latter.

I will agree with you on the basics, that many variables are in principle easily modelled.

No, that isn't what I'm saying. Modelling is tough. Once you have the model, however, it is easy for the CPU.

Even the rigidity of the body of a car is variable in GT. It's one of the parts in GT that actually wears and it's rather expensive to bring it back to its original state (in the game anyway ... ).

Okay, but rigidity is not something you can feel on a TV screen. Fully accurate modelling of it is a waste of time, and even if you did go down that path there are tons of ways to simplify the realtime task as well like modal analysis.

As far as physics go, my knowledge of them is limited to what my general curiosity taught me (and a few years of very basic stuff in school). But I understand what you mean with the centre of gravity. You are saying that if I put a 75kg driver on the right or the left side of the car, its centre of gravity just shifts to a different point in space, but it remains 1 point, no matter how many people I put in the car. So whereever the weight of objects is a factor in other calculations, you only need to work with the single variable 'centre of gravity'.

Yeah, that vector along with the moment of inertia tensor (a 3x3 matrix, of which usually only 3 elements are non-zero for a car due to symmetry). Looking at this and the first sentence of your post, though, I have a feeling that you're overestimating your knowledge of the physics in a car sim and solution of computational physics problems.

I understand you. But I still have a sneaking suspicion that in your enthusiasm about how easy you discovered it really was in your project, you are too eager to overlook at which point things become complex.

All you have to do is give me a counterexample to change my viewpoint. I knew it was easy for the CPU before the project. It was more of an exercise in execution and applied math. I'm not understimating the effort needed to create a Gran Turismo game in any way. Lots and lots of work there. However, the computational load for the specific part of the game that handles driving characteristics is very low. Just like a scientist can spend years or decades developing a formula, but putting it to use on a calculator simply requires a few operations.

The complexity of the physics in any Gran Turismo game is a statement of how superb the talent is at PD, not the platform it runs on.

 

[RobertR1]

No, that isn't what I'm saying. Modelling is tough. Once you have the model, however, it is easy for the CPU.

I think this is the part Arwin is having an issues with. Trying to keep both seperate. He's assuming that the difficulty in creating a good simulation model translates to needing a lot of raw cpu power to execute it. You're stating that for CPU's executing the model, regardless of how difficult it was to design, is an easy task. Close?

 

[bRoNx]

For the love of God, can you please change the title of this thread! I'd bet that at least 25% of the views this thread has had is people going "New GTHD video? Yeah!" - *click* - and then !

Look at the "New Forza 2 Screenshots" thread, probably suffering from the same thing!

Please remove "New" from the title.

That's all. Please continue.

Edit: Thank you

 

[Cyan]

I have nothing to contribute to this thread, other than I stumbled across a vid about modelling physics, specifically cars, by a chappy who worked on Forza.

http://www.gametrailers.com/player/usermovies/74862.html : 35 mins long

Here's part 2
http://www.gametrailers.com/player/usermovies/74906.html

Neat find! Very interesting. The interviewed person is very articulate and goes on the subject to a great deal of depth.

I remember reading Dan Greenawalt in the FM2's LCE exclusive booklet days ago, where he stated that, for the purpose of being able to recreate accurate physics (and, therefore, accurate scenes of racing) they did conceive a formula which is made up of 9000+ variables.

All I can say is that the game is well done. A bit of a classic; infinite gameplay, simple but dynamic playability, etc.

I also remember he said something about some magic formula called "Pacejka Formula", which is a physics formula that works great in racing games. He stated that he can inmediately recognize whether a sim (or arcadish) racer game uses this formula or not to recreate physics.

Pacejka formula works great but it's also somewhat limited in that it can push accurately.

So they made a new formula, based on some books (a great addition to their knowledge), specially "Milliken and Milliken" and some others about race car vehicle dynamics.

Back to the original game of this thread.... I wonder, what's their model physics like? I am pretty sure it is quite different from any other racing title out there but they don't seem to be that open-minded with their fans or attentive to feedback, perhaps because the main language of PD staff is not english -excuse me if I'm wrong-.

Btw, GT4, technology-wise, is quite impressive for a machine like the PS2 and it doesn't have any fundamental flaws (although I don't like skid's sound, almost make my teeth shiver as when I try to unzip a sleeping bag). I am going to give them that.

Cheerio

 

[Shifty Geezer]

I also remember he said something about some magic formula called "Pacejka Formula", which is a physics formula that works great in racing games. He stated that he can inmediately recognize whether a sim (or arcadish) racer game uses this formula or not to recreate physics.

Is that Beckman you're referring to? He said that when playing a game, he can tell when the Pacejka Formula is swaped out for an arcade-y formula at a point where the Pacejka becomes unstable. His example was at low speeds pulling into the pits, where there's a threshold at which, above is the sophisticated physics model, and under is the simple model, and between it's just hokey!

I think the main point was real car modelling isn't easy. You can knock up a racer with an off-the-shelf freebie physics engine, and it'll handle with a sort of 'real-world' physics, but accuracy needs lots of fidgetty stuff that can be unstable. And in some cases, impossible to resolve. I was a bit unsure of processing requirements though. The general topic was 'this is all too much for computers' but it was starting way back in history. There wasn't a given example of a, say, last gen racer and how much computation it was doing. He didn't give an example of Forza's calculations, accuracy, limitations to keep it doable, and such. So it doesn't really help with the Mintmaster<>Arwin debate about how demanding physics calculations for a racer are.

I thought the most interesting thing was the Rods example though! I saw that vid without it being explained, and wasn't at all impressed. But now I understand it, it's majorly cool! It's a simple system that's extremely scalable and ought to map fantastically well onto Cell and other fast vector parallel vector processors. I don't know how well it can handle the complexities of racing exactly though. The example looked good, with physical deformation, but is it accurate? 'Coz that's what a racing sim needs to be (look at the length of this thread to prove how demanding sim enthusaists are!) and they need to have those downforces, drags, temperatures, etc. all working. A simple rod system doesn't seem able to handle that level of data.

This discussion would maybe be better in the Tech forum. Heck, much of this thread should be there!

 

[Cheezdoodles]

Btw, GT4, technology-wise, is quite impressive for a machine like the PS2 and it doesn't have any fundamental flaws (although I don't like skid's sound, almost make my teeth shiver as when I try to unzip a sleeping bag). I am going to give them that.

Cheerio

Correction:

Gt4, technology wise is not quite impressive, its a goddamn masterpiece.

And i can point out a fundumental flaw right away, the game has no online, and the AI is worthless. So unless you like time trails, its not really a challenge.

 

[Mintmaster]

Shifty, those videos were pretty neat. The funny thing is that simplification plot in part 2 is how I've always modelled traction. I guess it matters where you start from. If you work for an automobile industry, you do things in terms of the standard parameters. If you're starting from scratch, using vectors is clearly the best way to proceed.

Take notice at his point about lookup tables. Yeah, in 1990 thousands of bytes is meaningful, but now it's irrelevant. The fanciest function ever is simply a bicubic interpolation of a table lookup. You consider the rippling of the tire and friction and everything, but in the end it's just a function of a couple variables. Lots of work to model, but easy to compute.

He seems to be a bit behind on integration methods, especially implicit integration. It's amazing how much stuff has been invented years ago and even presented at game developer or graphics conferences but wind up unnoticed. For that rigsofrods demo, there are some amazing ways of precomputing and simplifying a lot of that stuff (the modal analysis stuff I mentioned) so that it would run on slow computers too. Real-time soft body dynamics for characters, for example, has been largely solved 5 years ago, but nobody uses it.

 

[Mintmaster]

Correction:

Gt4, technology wise is not quite impressive, its a goddamn masterpiece.

To me it's not a masterpiece of technology, but one of content and execution. Brilliant artists, great physics modelling (to me that's content, because it's the parameters and formulas that are responsible), and very well put together by the visionaries at the top.

Gran Turismo has built an aura about it that will forever make it the undisputed champ of car simulation. Someone could make the perfect simulation in another game and it wouldn't get near the praise of GT5 because the latter will be close enough and you simply can't tell the difference (in a "blind" test, if that was possible) from your couch with a controller watching a TV screen.

Don't get me wrong, though. PD totally deserves it.

 

[Fu3lFr3nzy]

Aaaand after countless pages of arguments and zero gameplay videos....

New GTHD Video

Its still the LeMans Track with the Audi R8(full course), minus the shaky cam and annoying people chatting in the background

 

[RobertR1]

That's some nicely exaggerated body roll, dive and squat. Wasn't aware the Audi R8 comes on pogo sticks! http://youtube.com/watch?v=_ro30yBUUjs notice how flat the car is in cornering with Tiff driving wildly on a track.

 

[Nesh]

That's some nicely exaggerated body roll, dive and squat. Wasn't aware the Audi R8 comes on pogo sticks! http://youtube.com/watch?v=_ro30yBUUjs notice how flat the car is in cornering with Tiff driving wildly on a track.

Its very hard to compare with the video you posted.

 

[phat]

That's some nicely exaggerated body roll, dive and squat. Wasn't aware the Audi R8 comes on pogo sticks! http://youtube.com/watch?v=_ro30yBUUjs notice how flat the car is in cornering with Tiff driving wildly on a track.

You're making the mistake of comparing an internal cam view of GTHD with an external cam view of Tiff driving an R8. The sense of lean, dive, and squat are very different between the two views. There's another video of the R8 in GTHD from the external view that looks exactly like Tiff's video.

Edit: Here's a link to the video with good external views of the R8 in GTHD (best is the replay near the end). http://divx-333.vo.llnwd.net/stage6vid/1320845.divx

 

[sir doris]

Didn't someone already say that the GTHD Le Mans version was running what were effectively slicks?

That would produce more body roll, dive and squat due to the higher grip levels and increased forces on the suspension, compared to the car Tiff drive with road tyres.

EDIT: to say I haven't yet seen either video clip

 

[Mintmaster]

You're making the mistake of comparing an internal cam view of GTHD with an external cam view of Tiff driving an R8. The sense of lean, dive, and squat are very different between the two views. There's another video of the R8 in GTHD from the external view that looks exactly like Tiff's video.

Excellent point.

However, I hated how GT4 (and I guess GTHD now) treat the internal camera this way. Since when does a driver have his head rigidly strapped to the headrest? Humans tend to have this thing called a neck, as odd as it may sound to PD.

If you want to convey a sense of body movement in the pitch axis from inside the car, do it by showing motion of the dashboard and A-pillars. The camera should be level (at least in the fore-aft direction) and only rotate with the car. Roll axis is tolerable since the movements are less abrupt and the center of the screen is minimally affected.

 

[tha_con]

That's some nicely exaggerated body roll, dive and squat. Wasn't aware the Audi R8 comes on pogo sticks! http://youtube.com/watch?v=_ro30yBUUjs notice how flat the car is in cornering with Tiff driving wildly on a track.

Seriously?

Is it THAT hard to figure out that a few degree's on a bumper cam (which is normal for any sports car) can be GREATLY exaggerated because EVERYTHING in the distance moves exponentially?

Draw a 2 degree line with your imagination, and carry it out 500 feet. I promise you it will be much greater than you perceive 2 degree's to really be.

Even if you put a cam on the bumper of a car (like the said car in the video) it would probably look the same. Really now, it would be nice if you'd put two and two together.

 

[DJ12]

However, I hated how GT4 (and I guess GTHD now) treat the internal camera this way. Since when does a driver have his head rigidly strapped to the headrest? Humans tend to have this thing called a neck, as odd as it may sound to PD.

There has never been an "drivers eye" view.

There's bumper and roof (Maybe bonnet also) then the chase views.

I haven't seen my drivers with their heads on the bumper while driving

So you basically don't like PD for not doing something they shouldn't be doing.

The only view this should even be entertained as an idea is cockpit view and no one outside a few media types and the staff at PD (and Sony) have even seen it in action at this point.

 

[Mintmaster]

When switching between cameras it always felt to me like this camera that I'm talking about (which phat called "internal" and you call "bumper" or "bonnet") is approximately located where the driver is. Too high to be a bumper cam, and if it was on the hood then by the time another car entered the field of view it would already be past you. During licenses, I also remember the ghost car's rear jumping into view when it got about half a car length ahead.

It's been a while, though, and I don't have a PS2 on me right now.