The GT5 expectation thread (including preview titles)*

[Mintmaster]

Mintmaster, you cannot use distance traveled when comparing breaking at different speeds the way you are if you don't have the time it took to stop.

Why not? I am simply determining average acceleration. I could probably give you a proof with the mean value theorem or something, but if the average is lower for 100-0 than 60-0, then from 100-60 it has to be slower. 100-0 has times, and they match the calculations from the distances within a tenth of a second or so. Cyan's data above for the F1 car also matches. 130 to 0 in 55 metres calculates to 1.89 seconds. t = 2*d/v (for this scenario, assuming constant acceleration).

I'll admit, not the greatest example since the conditions were likely different.

Indeed, so I don't know why you brought it up. Two different publications on different tracks with different cars (even if the same model) is a pretty crappy experiment. I don't know what's so hard for you to accept about the data I showed. Threshold breaking is close to constant acceleration, and the equations are very simple.

For example, the amount of energy spent to move an object from 50mph to 60 is much less then 145 to 155.
...
Likewise the amount of resistance needed to slow an object moving at a very fast speed(155 to 95mph) is less then you need to slow the same object down from 60 mph to zero.

Uhh, what? How is that "likewise"? If you need more energy at high speed to accelerate, why do you need less resistance to decelerate? You stuck in the world of Aristotelian physics?

As I said before, you have to dissipate 4 times the energy in the former breaking scenario than the latter.

Sorry i dont want to sound like an ass but i thought this was common sense. More wind/air resistance plus more friction(tires to asphalt, axles and anything else on the car having to maintain a higher speed) = more energy consumed to maintain speed, thus less energy to decrease your speed by the same amount.

If you are travelling at 150 mph and go into neutral, how long will it take to slow down to 100 mph? Maybe 20 seconds or so? The retarding force is a small fraction of what can be achieved by the brakes.

You need more power to go fast for a reason.

You need more power to fast even in the frictionless confines of deep space. P=Fv. E=M*V^2. It's physics.

Anyway, you are completely missing the point. I am not saying that it's impossible for the Audi R8 to decelerate faster from 155 to 95 than from 60 to 0. The point is that the deceleration is roughly the same. It's going to be around 1g, not 1.6 or higher for this car without F1 style downforce.

However, Arwin brought up the issue of cars with net downforce at speed

Yet again, people aren't paying attention to my posts. I already said above that faster deceleration at higher speeds only happens with high downforce. You're preaching to the choir.

 

[Arwin]

Mazda Miata, or the Mx-5 like europeans call it, this car is super easy to drift with in real life. In GT4, not at all.

Yep, you hit the spot. This car was one of the biggest problems of GT4, and drifting in general was too hard in GT. So the main discussion has always been about grip. The time it takes to shift that car in the game versus real life is another complaint from real life owners of the MX-5. It was also a problem with the Evos in the game.

That's why I found it very interesting that we have the MX-5 and 2 Evos in GT:HD - it is almost as if they specifically chose cars for GT:HD that GT4 players had the biggest issues with. And sure enough, they seem to be fixed! Drifting the MX-5 in GT:HD, even with the stock tires and not using drifting mode, is easy. In fact, I don't think there is a physics driven difference between the drift trials in GT:HD and the regular Time Trials. Some big advancements have been made here.

 

[ihamoitc2005]

Braking

At high speed aerodynamic drag is greatest force. Even removing foot from the gas can have fast speed reduction without brake use. Tire grip for high speed braking is also more because hitting brakes makes vehicle "pitch" forward, causing increases of down-force, drag, and tire grip. Car with more wheel travel will have this "pitching downforce" effect more. Car with little wheel travel but big "wings" like Formula 1 already have much more tire friction at high speed because very high downforce. High speed downforce for Formula 1 car is more than weight of car so this is why Formula 1 can drive upside down on a tunnel ceiling. More hard it is to go faster more easy it is to slow down.

One car is 200 mph, one car is 50 mph. Both have same mass, shape and brake mechanism.

Momentum = mass x velocity.

Force = Rate of change of momentum

If Force goes up, rate of change in momentum goes up also.

Aerodynamic drag is proportional to square of speed so 200mph car has 4x sped but 16x aero drag of 50 mph car.

So, since total resistance Force for braking 200mph car (brakes+drag) is much more than resistance force for 50 mph car, rate of change of momentum is also much more for 200 mph car.

Since mass of cars is constant, higher rate of change = higher velocity change. So, 200mph car has faster velocity change than 50 mph car. So braking 200-150mph is less time than 50mph-0mph.

 

[Cheezdoodles]

From what you stated in the last topic, your method of testing left out quite a few rather important facts which pretty much make any results you gained from them useless.....

In an effort to end this, how about we just leave it at that, that Polyphony should have used less ideal conditions in the game which would result in less grippy cars?

I don't see why your so in denial about this. Even arwin has admitted that grip has been an issue for other cars aswell, like the Miata.

I told you, straight up, my Merc CLK500 Kleeman gets outhandled by a lot by the CLK in game, even tho i have the same stock suspension setup, and better tires than the CLK AMG has to begin with.

If PD is modeling tire grip based on the best ever grip you can have with the car ever. (And not taking into account tire heat etc) then there is to much grip. Period. End of story.

(Now, with Forza 2 on the other hand, my car has to little grip and braking power.)

 

[Mintmaster]

At high speed aerodynamic drag is greatest force.

Drag is nowhere near as big a force as can be applied by the brakes through the tires.

Aerodynamic drag is proportional to square of speed so 200mph car has 4x sped but 16x aero drag of 50 mph car.

At 100 mph drag does not give road cars any faster decelation than at 60mph, as proven by the data for 100-0 and 60-0 times and distances. Multiply that by 2.4 for 155mph, and the drag force is still very small compared to the brakes.

Want some numbers for the Audi R8? Remember that P = Fv.

A 1560kg car at 155mph braking at 1g is equivalent to -1421 horsepower. The R8 has a top speed of 187mph with a 420hp engine, so theres probably 200 hp needed at the wheels while cruising at 155mph using your square law for drag.

Thus air resistance is under 15% of the braking power at 155mph, and will taper off to ~5% at 95mph. There is no way that an Audi R8 will be braking at 1.6g.

EDIT: BTW, -1400 hp is over a megawatt of heat production. That's why high speed breaking is tough, despite the help from wind resistance.

 

[Mintmaster]

From what you stated in the last topic, your method of testing left out quite a few rather important facts which pretty much make any results you gained from them useless. Grip is a product of surface, surface temperature, tyre temperature, tyre compound, contact patch (tyre pressure, size), weight and the speed of the vehicle.

I mentioned a couple of times that

a.) surface on race tracks differ from those on public roads
b.) surface temperature is not variable in the game
c.) no game simulates all different tyres supplied with each car, especially if the focus is on racing tyres
d.) tyre pressure is very important in real life and makes a huge difference on grip level
e.) driving GT with the wheel or controller makes a huge difference in how easy it is to provoke oversteering

In other words, you could probably look at GT "simulating" the best case scenerio: Hot, grippy surface, ideal tyre pressures/temp etc.

All these factors do not make the kind of difference we're seeing. 1-1.1g braking and lateral g force is the result we get from supercars on idealized test surfaces with high end tires. We're just seeing way too much traction here throughout the game.

(BTW, Ostepop, braking and grip are the same issue in GT4, because too much of the latter makes the former too good.)

 

[Mintmaster]

I think you're underestimating how complex suspension, tire contact, and drivetrain models can be. Multi-link suspension, for example, tilts the wheels as a function of displacement and steering. This tilting will have an impact on tire contact, and cornering force will depend on this as well as the material of the road (dirt or asphalt, for example). When you're trying to model hundreds of cars according to factory specs, as well as under tuning, your model needs to be quite complex to be accurate.

A car's suspension is very easy to model, even if multi-link. For a given steering position, every suspension linkage in the world is a one degree of freedom system. You simply model the lateral displacement, wheel orientation, etc. as a function of vertical displacement. The moment of inertia and mass of all these elements are reduces to one effective mass. There's no need to simulate every element.

Similarly, you don't need to physically simulate every element of the drivetrain as that would be sheer stupidity. You just need a mathematical model from the pedals to wheel torque that uses some internal state variables like current engine RPM, turbo pressure, current wheel velocity, gear, etc.

A "complex model" capable of everything you said doesn't need to be computationally intensive at all. A hundred variable formula for each major part of the car can be handled with ease on a PS1. You just need a bunch of numbers to specify the effect of each vehicle/part/surface, and that's about content creation.

And while collision is simpler in a racer than in HL2, integration has to be a lot more accurate in a racer than an FPS. An FPS just has to look approximately right, but a racer better be able to reproduction acceleration times, stopping distances, cornering speeds, etc., accurately to real-world counterparts or players will gripe.

Sorry, I don't mean to be arguing with every other post of yours.

You don't need complex integration for reproducability. All the things you mentioned can easily match real-world values with simple Euler integration, let alone the far better yet still simple alternatives.

Physics is not hard because of accuracy. HL2 is harder because the types of physics are harder to solve. More objects have the possibility of hitting and being in simultaneous contact with each other. Free bodies have to react to impulse forces many orders of magnitude higher (relative to their mass) than what a car feels. Consider a stack of light, wide boxes, and dropping a heavy ball on top. What happens? In reality, the reason the stack doesn't move downwards is because the ground provides an instant upwards reaction force. You have to propagate the forces down through the boxes, then back up to the ball. These are all free bodies so you can't take any shortcut or preprogram a particular path through the contact points as I just described. It's not an easy thing to do, and requires many iterations.

Now, for a racing game, you simply model the motion of the wheels independently along the aforementioned single dimension (assuming the car body follows the same trajectory as specified by the last update), then apply forces on the car based on the wheel position and the forces they apply to the ground. You can iterate to make it more accurate if you want, but I guarantee you that nobody in the world will be able to feel the difference playing the game.

Even at a 360Hz update rate and a weak processor like a 386, you have a hundred thousand cycles to spend on a few formulas. The quality of car behaviour has everything to do with the parameters in those formulas, not their calculation accuracy.

I've coded a simple car simulation (though obviously without the car data that PD has) as well as a general free-body physics engine. I know what I'm talking about. For the driving simulation aspect of a racing game, it's ALL about the model. CPU power gets you nothing.

 

[Arwin]

You're really oversimplifying things. I stand by my earlier comments that you have to balance everything that is done in GT4 out against the limitations of the hardware AFTER everything else that is already being done, from the graphics, the sound synthesizer (too little memory to use a lot of samples), upscaling to 1080i (on the NTSC version anyway, even if tricked, still impressive), the dust and so on. Literally everything you add will mean no room for something else, and all aspects of the game have to compromise.

You can go on and on about Half-life 2 physics being more complicated, but that game has not been released for the PS2.

As for the rest, everything depends on how far you are willing to take the simulation. Grip is much harder to model than you claim it to be, if you want to be accurate. I won't claim to be an expert, but if you are going to simulate braking for instance, you have to consider not only the weight of the car, but the 3D location of that weight. Then the movement and decelleration impact has to be calculated against the pressure from the four individual wheels relative to the surface they are on, the angle at which they are on the surface, the grip levels of the surface and tires, the deceleration that the car's flywheel/clutch/engine themselves give to the car when it is coming down from a high level (which varies according to the gear you have the car in - you'll want to keep downshifting while braking to decelerate faster). Then there is air resistance to take care about taking not only downforce into account. Especially the fact that a lot of these factors work against each other continuously makes this very hard - you'd have to do a lot of them in paralel, and then recalculate them in paralel a little bit ahead of the actual experience to really do these effects justice. In that respect, the complexity can easily move towards your box problem. Maybe the latter is harder to do 'cheaply' than the former and it is easier to get to a relatively convincing level, but there is plenty of room for taxing your 'calculator' if you are striving for realism. Just think about the tires alone, the tire surface, pressure, hardness, sure, these can be caught in a few 'variables' per car, but then there is the matter of the 'resolution' of how you are going to model the wear and tear and grip of that round object against a variable underground, all the effects described above together determine on how much the actual wear is. And those results have to feed-back into the other calculations again. It's a continuous stream of data to be managed which can get pretty crazy too if you pick up speed in the car, and I don't think its easy or something any old 386 could easily manage.

Of course, I could be wrong and I can see how other situations could be more complex, but right now those other situations also aren't simulated with a similar level of fidelity. In the mantime you sound like someone from Codemasters working on Race Driver, stating that after a few weeks you were happy with the physics, claiming they were plenty realistic enough, so that you could now focus on other stuff like PR interviews.

But I guess you should be having this discussion with someone from PD. That would be much more interesting.

 

[Mintmaster]

Arwin, everything you say can be modelled with a trivial amount of CPU power. It doesn't matter if the CPU is doing other things. 5% of a PS2's processing power is more than enough.

CPU power is consumed by doing hundreds of iterations, recursing and searching through nodes, performing algorithms (esp. n^2 or n^3) on many (i.e. tens or hundreds) objects, etc. Nothing that you've described involves any of these. I'm telling you that if PD had to run the driving physics code using 10% of one SPU, they could do it without you noticing a difference. Rendering, sound, AI, locating the surface parameters below each wheel, and collision detection are far more difficult, and beyond rendering even those are pretty computationally easy for a racing game.

Weight transfer is already included in the stuff I mentioned above. Braking doesn't calculate the "3D location of that weight", it uses the downward force applied from the suspension. Air resistance doesn't do FEM simulation, it uses a formula with car speed and location relative to other cars for drafting. Flywheel momentum is determined by engine RPM, which is simply a state variable for the drivetrain equation that I already mentioned. Tire wear is quantified by a single scalar which the user sees as a colour from red to blue, and again is simply a model. NONE of these things can bring complexity even remotely close to the box problem I mentioned, which does require searching through nodes and lots of iteration to approach convergence.

Like I said, a hundred equations each with 100 variables (not just "a few") is so, so simple for a CPU. It's all about the model and its parameters. Your CodeMasters rant is irrelevant. Being happy with the physics has to do with your model and nothing to do with CPU power. I never said realistic driving physics is easy to model. I merely said it's piss easy for a CPU to calculate. The models and their parameters require a huge amount of money and labour to get right.

You, along with others here, have a very poor understanding of what a CPU is good at and what taxes it. Is it safe to assume you haven't done any performance critical progamming before? Why are you arguing with me on a topic that is so completely out of your league?

 

[MechanizedDeath]

Have you factored in engine-braking into the equation, MintMaster? Stuff like engine braking, brake fade/glaze and road incline also play roles in stopping. We should also factor in that unspring weight, the heavy rollers that while adding cornering stability also add rotational inertia that effects braking distance. Dropping a few inches on wheel diameter can make a great deal of difference in decel/accel.

It's an interesting argument, but I don't see any discripancies in the model that warrant such close inspection. By no means do any of the console games achieve realism, but I believe they also cater to a large number of vehicles (each tweaked to approximate their real life counterparts as best as possible) and the total package still manages to recreate similar lap times. Dare I say, sector times end up fairly realistic too, which means that somewhere between accel/decel and cornering, the "equation" balances itself out. ...and I'm rambling. PEACE.

 

[Mintmaster]

I'm already assuming perfect disc brakes and braking is limited by tire traction. Engine braking just helps to remove some load from them, but in the end the reduction in car speed is through the tires. The other factors you mention would only reduce braking ability, and would point to GT4 being even more wrong.

Like I said before, I could really care less and GT4 accuracy is good enough for me. I'm really just in this this part of the discussion to inject some sense into the physics talk.

 

[RobertR1]

If you want close to perfect simulation, have F1 engineers help design the sim/physics engine. Their simulator (like McLarens) are predicting car setup and qualifying within a second of actual laptimes when F1 hits new circuits. Now THAT is simulation. We're just playing simple video games

 

[phat]

You don't need complex integration for reproducability. All the things you mentioned can easily match real-world values with simple Euler integration, let alone the far better yet still simple alternatives.

From this statement, I can safely say that you haven't ever had to integrate a model for scientific or engineering purposes. Simple Euler integration is only for toy models that vaguely resemble the real thing but aren't useful for numerically predicting anything. Even a simple earth-sun planetary model needs better than Euler integration in order to predict an orbit that matches reality (or the closed-form results of the model). Trying to reproduce braking times, acceleration times, cornering speed, etc., all simultaneously accurate to real life numbers requires very good integration. Which is probably why you see racing games with sim aspirations run their physics model at much higher than the frame rate--because most of their integration is Euler and needs small deltas to be sufficiently accurate. And even with the high update rate, some aspects of a driving model will need more than Euler integration to deal with some of the impulse transients, like going over bumps and steps.

BTW, see http://en.wikipedia.org/wiki/Midpoint_method for an example of how inaccurate the Euler method is.

 

[phat]

I'm really just in this this part of the discussion to inject some sense into the physics talk.

Some of us have more sense in physics than you give us credit for.

 

[manux]

Some news from french magazine having interview with gran turismo makers. Seems great, 12-20 player online, car damage, changing weather. Sounds a bit too good to be true.

I seem to be 1 month late to this news.

 

[Mintmaster]

From this statement, I can safely say that you haven't ever had to integrate a model for scientific or engineering purposes. Simple Euler integration is only for toy models that vaguely resemble the real thing but aren't useful for numerically predicting anything. Even a simple earth-sun planetary model needs better than Euler integration in order to predict an orbit that matches reality (or the closed-form results of the model). Trying to reproduce braking times, acceleration times, cornering speed, etc., all simultaneously accurate to real life numbers requires very good integration. Which is probably why you see racing games with sim aspirations run their physics model at much higher than the frame rate--because most of their integration is Euler and needs small deltas to be sufficiently accurate. And even with the high update rate, some aspects of a driving model will need more than Euler integration to deal with some of the impulse transients, like going over bumps and steps.

BTW, see http://en.wikipedia.org/wiki/Midpoint_method for an example of how inaccurate the Euler method is.

Well, you'd be wrong, as I definately have used Euler integration and other methods for engineering purposes, and have seen first hand how they miss using trajectories of an orbit.

However, this has zero relevance to reproducability or ability to match lap times to real life for a racing game.

Euler integration will still be just as reproducable from user to user and machine to machine as Runge-Kutta. As for matching lap times, you simply change some parameters and voila - the end result is the same. Something as simple as slightly increasing the angle that the wheels turn for a given user input would be enough to match trajectories around a curve to those of a super-accurate integrator given identical user input. Increase brake force a bit, and again you match everything perfectly.

This is just one example of where a simple model is 99.999% identical to the perfect model for the end user simply by adjusting the parameters of that model. There is no way in hell that you or anyone else can tell what's going on behind the scenes simply by playing the game.

My point is not that Polyphony Digital actually does this. Runge-Kutta is so cheap and easy that you might as well use that. My point is that integration accuracy has little to no importance on the realism of a car simulator. More pertinent to prior discussion, it has nothing to do with the CPU load of the driving physics.

Some of us have more sense in physics than you give us credit for.

Sure, but some of you don't, and even those that do have occasionally proven to misapply it now and then.

 

[phat]

Euler integration will still be just as reproducable from user to user and machine to machine as Runge-Kutta. As for matching lap times, you simply change some parameters and voila - the end result is the same. Something as simple as slightly increasing the angle that the wheels turn for a given user input would be enough to match trajectories around a curve to those of a super-accurate integrator given identical user input. Increase brake force a bit, and again you match everything perfectly.

You're changing the topic. I wasn't talking about reproducibility from machine to machine or session to session. Computers are deterministic, after all. I was talking about reproducibility against the real physical systems being modeled. If you've modeled the system well enough, and use a sufficiently accurate integrator, then using real-world values for the parameters should give you reasonably accurate real-world behaviour. Fudging parameters to get correct behaviour in certain scenarios is a can of worms since it invariably means that other scenarios will not be modeled accurately by the same model. It's like changing the gravitational constant in a planetary model just so that the earth is in the correct orbit around the sun because your model or integrator wasn't accurate enough. (Using Euler, for example, will tend to have orbits spiral outward.) Doing so can correct the earth's orbit, but, look, now the moons of Jupiter are spiraling into the gaseous giant!

 

[Phil]

I'm already assuming perfect disc brakes and braking is limited by tire traction. Engine braking just helps to remove some load from them, but in the end the reduction in car speed is through the tires. The other factors you mention would only reduce braking ability, and would point to GT4 being even more wrong.

Like I said before, I could really care less and GT4 accuracy is good enough for me. I'm really just in this this part of the discussion to inject some sense into the physics talk.

Tire traction is the limiting factor when braking down to full stand still. It usually isn't the limiting factor at higher speed, which is what I think was being argued here when looking at the GT:HD videos - or am I missing something?

 

[Arwin]

Arwin, everything you say can be modelled with a trivial amount of CPU power. It doesn't matter if the CPU is doing other things. 5% of a PS2's processing power is more than enough.

If you say so. Pardon me for remaining skeptical though. My programming is not (that) performance critical, and involves mainly the nodes stuff. If I were doing any form of performance critical programming, I would most certainly never work with recursing / node searching in the way I do now, but program everything as datastreams. It's a matter of principle.

CPU power is consumed by doing hundreds of iterations, recursing and searching through nodes, performing algorithms (esp. n^2 or n^3) on many (i.e. tens or hundreds) objects, etc. Nothing that you've described involves any of these.

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.

locating the surface parameters below each wheel

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

Weight transfer is already included in the stuff I mentioned above. Braking doesn't calculate the "3D location of that weight", it uses the downward force applied from the suspension.

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

Air resistance doesn't do FEM simulation, it uses a formula with car speed and location relative to other cars for drafting. Flywheel momentum is determined by engine RPM, which is simply a state variable for the drivetrain equation that I already mentioned. Tire wear is quantified by a single scalar which the user sees as a colour from red to blue, and again is simply a model. NONE of these things can bring complexity even remotely close to the box problem I mentioned, which does require searching through nodes and lots of iteration to approach convergence.

Plenty of examples there on how we disagree on what is needed.

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.

But then a holiday is coming up, so who knows I may find some time to discuss this some more.

 

[Mintmaster]

Phil, you're basically making GT4 seem even more unrealistic. Tires traction generally limits your acceleration (whether lateral, braking, or whatever) to around 1-1.1g for good sports cars unless they have lots of downforce. If you weren't tire traction limited at higher speeds, then that means you are below that threshold and decelerating even slower than that. I agree with you that it's pretty easy not to be traction limited because the enormous energy dissipation would reduce their effectiveness.

The video, however, is showing much faster braking (~1.6g) than what should be possible even if you had perfect brakes and were tire limited.