PDF: ''tires in race simulations'' some thoughts..

N

Niels_at_home

Registered
Hello,

I was thinking a lot and decided to write somethig down.

Some observations how ''real tire data'' is (or is not) useful, and what I want from a tire model from a 'modders' point of view.

Feel free to read or ignore,

Niels
 
Nice! Thanks Niels! *really need a thumbs up emoticon here*

edit:
I've read the PDF and from what I read, there is one aspect that will make modding much easier (and faster?) in the long run: just create the tire generated model (.tgm) file once, and use it over and over again, even by different car mods, as long as the tire's physical attributes are the same or similar (width, radius, etc).

Or maybe not. :rolleyes:
 
Last edited by a moderator:
good work niels. you did a very nice sum up of the situation.

i just would add that i don't think, in a simulator project, we have to choose between the 2 kinds of model you describe.

my bad english prevents me to explain well what i mean, so sorry guys ;)
as we know, pacejka is a semi empirical model anyway. i think a simulator can make his own "semi empirical model". the magic formula make a link between real facts and theoricals elements to finish in a result close to reality.

IMO a tire model of a simulator, like rf2 here, have to provide his own "magic formula" to make a link between what we can find as datas and what the engine need.

if ISI can't provide it directly, we need a better tool to find our own formulas to be able to match to reality, as much as possible.

this is not an opinion of engineer, or developer, just a modder one ;)

ty niels, this is a good base for a discussion. i hope ISI guys will look this thread and post their opinion.

cheers
 
You’re not having a slip curve as input, the model actually works out how the tire grips in each situation based on its current deformation, construction and materials properties. From a programming point of view this is mightily impressive. But is it as good and flexible for the people who end up working with it? It is too early to tell though I think I am slightly worried.

The best model in the world will not result in good tires if it is not easy to work with by end users. While I am sure the model will do logical things most of the time, how can I design certain load sensitivity into a tire and go for a target cornering stiffness versus tire load curve? Can these be predicted or do I have to wait hours for the new tire to be computed before I can test it out? How can I ever draw these curves that I need to interpret the performance of the tire?

With this type of model it is less up to the guy making the mod and more about the model deciding how tires respond to load, slip, temperature etc. It may be harder to do something badly but a lot of trust has to be placed into the model.
Click to expand...

And this is the key thing with the rF2 tire model. A "physical mode" can work in a way like iRacing, where the programmers/physics engineers etc know what are they doing, but in rFactor2 I'm not sure if it's a proper way, because the physics guys (in the best scenario they are a bunch of guys with some charts from Internet) don't have the knowledge required, and the process of making tires looks difficult and time consuming.
 
To take real measured data and put them straight into a game is rather useless. The physics models in games are not that good, especially not on tires. I'm not saying that real data is useless in modding, but you have to look beyond the numbers and capture what properties the numbers stand for and try to achieve that with the tools available in the game. Matching slip curves number by number isn't necessarily the best way to do that. Modeling cars, in for example rFactor, with success is not a science but more like a black art. Yes, you need understanding about real world physics but also about what works best in the game. Once we get some experience with the new tire model it will get better.
 
Novis summed it up nicely. Too many people get hung up on inputting 'real' numbers in mods. What is important is inputting numbers that will give the most accurate reaction from the tires. The numerical values you have are irrelevant as long as the car responds correctly.
 
But how do you conclude that it reacts correctly?
There is necessary feedback of real world drivers of the car and maybe some telemetry.
It is not easy at all...
I guess the new tire model is better for modeling the tire reaction on strange things like being on the curbs or jumping or sudden movements, so I guess it is worth the effort and trial and error.
The modders that are not willing to spend hours testing and testing them, then maybe they could reuse the tires from another mod.
 
How many mods are there for rF 1 with tyres not based on real-world data, which are at least "good"? Not much (including some of those "made based on real world drivers input" ) ;).

No matter how "off" all those real world tyre data sheets are, they are still a good representation of that tyre and from my own experience, it is definatelly better to use them (and in the end maybe tweak the values just a bit) than trying to make own tyres from scratch becase of "I think they are ok and they feel good to my teste" kind of attitude.
Of course there are guys out there who actually can do nicely working tyres from scratch, but they have enough book knowledge to do that stuff properly.
 
Ivan Baldo said:
The modders that are not willing to spend hours testing and testing them, then maybe they could reuse the tires from another mod.
Click to expand...
I'm pretty sure it will work out exactly like it did in rFactor 1. Rfactor 2 will just provide us with a diffferent tool to make crap mods.

The majority of the mod makers will continue with old habits from rF1 and just quickly copy the files from another mod. Maybe they spend some time to edit the credits and randomly change some numbers to not get caught and then release it. Obviously this will not work and they deliver a horrible mod.

Some mod groups however will spend some time to understand the game model. They will research data for weeks and put them into the mod and still produce bad mods as the real world and game model is not the same.

A few modders will look past the real world data, test to capture the feel and get it right.

:D
 
If a simulation cannot handle real world numbers correctly, it is faulty simulation. Altering data to fudge results is like treating the symptoms of an illness. You will not fix the actual problem, you just end up masking the illness. When the simulation is then repaired, the work done to create a mod increases as you will then have to go back through your faked data and input what should have been there to begin with. Furthermore, there are a bunch of variables, by using real world numbers you can change over some of those to constants. Faking the numbers to get a result is somewhat of lazy modding. There are still plenty of unknowns to play around with to bring those results in line with reality, but altering known values is just silly.

Judging by the NHTSA document referenced to about the tire model, I'd say the Devs got it right, and that worries should be put aside. I only hope all aspects of the simulation are detailed to that extent.
 
Niels,
I do not fully agree with how much control you can have with the rf1 tire model. I do not know what tools you use, but what rf1 actually does with the slip curves is somewhat of a mystery to me. It seems to work fine in a small window, but when you try to do something radical to work around bugs, it becomes somewhat unpredictable.
As you said, the bugs and other implementation issues is maybe not be directly related to the type model used, but can the model limitations be avoided? If you increase the amount of functions and variables, sooner or later you will need to interlink more and more of them and then why not go to a physical model?
When you have to create a tire with a model like rf1, it is already way too easy to get lost in all the variables. This is quite obvious if you look at all the different solutions mods use to try to simulate the same tires. As you said, real data is pretty much useless, so everything is pretty much modders blend. From slip curves and angles to camber effects to speed sensitivity. Then into the black box that is rf1 and something comes out. I do not think this optimal.
An advantage with going to a physical model is that the model should be able to take care of all that. You only need a tool to create a reasonable tire. I understand that trying to fit a tire to data might be an interesting task, but for general modding there is definitely some advantages.
 
Hello hello,

Thanks for the replies.

Andreas, I actually really agree with you. After all these years there is CarFactory which is a sort of tire tool, and I have developed tools myself. Theoretically, rF1 type tire models are 'easy'. However, ISI didn't make it very easy for us giving us only small bits of information. If they were more forward with this, or if it was Pacejka, it would be a matter of a couple days working in Excel and you'd have the model there. So in theory, Pacejka / rF1 models are 'simple', in practise, rF1 isn't the best example of 'transparancy'. :)

But even if we had good tools and information about the physics engine, I still think a major part is the subjectivity. You would still get bad and good mods, because one guy feels the cars should be spinning out driven at 101% and others want a 30 degree slide in every corner. You see this in rF1, where half the simracers like one type of handling, and the other half likes another type of handling. Both claim theirs is more realistic.

Even if the physical tire model is more likely to make a decent tire, it will always have input variables. I really don't know much at all about rF2 but it must get its data somewhere and I'm sure that in some way we can make quite a different behaving tire with this physical model, just like rF1 mods show a difference from mod to mod.

And there is a lot more than tires of course, with funky suspension geometry or aero easily affecting a mod that has good tires. In short I think there is no getting away from the fact that physics modding is very complex and will always be largely down to the guy doing it, with the physics engine and tools being perhaps at most 50% of the story.

iRacing is a good example where they have plenty of budget and people yet after what, 4 years, they're still struggling with the handling. I don't mean to talk about iRacing but it shows that good cars just aren't build easily regardless of the physics engine or the amount of money and people you throw at it.

I don't think there will ever be agreement on physics either. Players have just too different expectations, talent, computers, controllers, latency in the screens etc etc. And they will always drive cars made by 'physics guys' like me who also have certain expectations, lack of talent and a tendency to be obnoxious from time to time. The question 'physical model or not' plays only a small role here.
 
i don't know anything about tires and its not a mod for rf, it's built with its core engine though, and the ARCA racing software uses actual real tire data direct from Hoosier corporation! their software credits an engineering firm too. not really fair to compare to modders but to defend the software i think is appropriate and shows what is capable with data not available to public.
 
Niels_at_home said:
But even if we had good tools and information about the physics engine, I still think a major part is the subjectivity. You would still get bad and good mods, because one guy feels the cars should be spinning out driven at 101% and others want a 30 degree slide in every corner. You see this in rF1, where half the simracers like one type of handling, and the other half likes another type of handling. Both claim theirs is more realistic.
Click to expand...
New tire system is really magnificent when you look at software level, but does it allow too much to play with to have most modders to gain from it or is it so free and complex that there is only 1% that can actually get useful quality out from it?

So I think that what good is Bösendorfer when you can't play Yamaha? I think that new tire model and everything around it is indeed great, but it might be too great for 99% of those who try to use it? I think that would something less free and limited be something that would give bigger amount of quality content?

Hmm, here I read document Niels made, I guess we think bit same way then? :P



With perfect tire model, you would input type of tire and size + other stuff from side of tire, software would do the rest. Type of tire would be something like 70's era standard M+S or modern sports car tire and so on. Model of course is not perfect, but I would think that if we take average of all mods, such limited version would actually average a higher level of mods. Sure one could throw in tweaks for grip to adjust it a bit, but within limits of tire type. But it would require quite bit of planning from model maker, to allow enough tire types, from R/C car to dump truck used in mines.

There I think our opinions might differ of course, but I think that from different opinions we can always learn something.

I'm however pretty much against any limitations, so I think anything simple should be an option, there is under 10 persons who actually understand and can do anything they like with rF1 model, that 10 is probably very high, being closer to 5, also there is maybe under 100 modders? It is something to keep in mind.

edit: Bummer, forgot lines that I quoted already, old age and all that...
To that quote, I would like to say both camps being wrong, both can happen and even at same day on same car and at same corner :D
 
First problem with the tire data sheets is that the slip curve from tests are for one specific condition, and slip curve of the tire is constatly changing due to different loads, different surface on what tire is moving etc :) Now we dont input tire slip curve, but we model actual physical tire in virtual world :) For me this is awsome :)

Example from book:

ux - coefficient of adhesion
Fx - peripheral adhesion strength
Fz - tire load
sx - relative circumferential slip road wheel moderated
V - vehicle velocity
Rd - dynamic radius
w(omega) - angular velocity

ux = Fx/Fz
sx = (V-Rdw)/V for V < Rdw

Slip curve is function ux(sx)
 
Very interesting thoughs on tire model dated from 2012. Back in 2012 there was no large experience with the new physic based tire model - I would be interested in the assessment of the physical tire model after having roughly 3 years more of experience.

Furthermore I would be interested - because it sounds that in the physical tire model you are not defining any slip, grip and slip curves whether there is any opportunity to get this data out of rF2 tire definition files?
 
You must log in or register to reply here.
Back
Top