The online racing simulator
How complicated can a tire model become
(143 posts, started )
Someone should tell him about look-up tables...
We've seen how awesome those work in other sims....
Quote from Be2K :... But a complete Dev. Team is developing a hole new Tire Model since 2012 and they aren't nearly done with it...

"(W)hole new" blah blah blah...

Most of the PCARS devs were not even born when the theory behind their model was made public, and what they are doing is a extremely simplified real-time version.

Quote from racon :someone should tell him about look-up tables...

Quote from dawesdust_12 :We've seen how awesome those work in other sims....

There appears to be some confusion here. Let's not mix-up physics problems with software engineering problems. Look-up tables are what you need if you go static, and if you go dynamic the model is totally different anyway.
I've been saying for a few years now: There has to come to a point 'soon' when the tire model is not perfect but better than the old model. I know Scawen is his own boss etc but surely he knows time is running out for everyone. Life goes too quick, and if he loved his sim like he used to it's better for all of us to be playing on a 'better' tyre model than we used to have even if it's not as perfect as he'd like.
Quote from Alric :I've been saying for a few years now: There has to come to a point 'soon' when the tire model is not perfect but better than the old model. I know Scawen is his own boss etc but surely he knows time is running out for everyone. Life goes too quick, and if he loved his sim like he used to it's better for all of us to be playing on a 'better' tyre model than we used to have even if it's not as perfect as he'd like.

two things: if i recall correctly, the decision to not release the Scirocco was not made by Scavier. it was more a compliance on their part to some agreement, so there may be something more than personal opinion on the model's performance. (my memory could be faulty here, or this is all hearsay...) also there might have been many revisions sent to some anonymous testers we know nothing about.

second and more importantly, and this is pure speculation, i don't think the tyre model Scawen wants can be 50% or 75% more functional than the current one. having decided which additional parameters need to be modeled, either the real-time calculation models them or it doesn't. (and possibly the parameters all need to be there or the result is no good :shrug
Quote from dawesdust_12 :We've seen how awesome those work in other sims....

The other sims with Scawen's frame, you mean?

Quote from Keling :Look-up tables are what you need if you go static, and if you go dynamic the model is totally different anyway.

Look-up tables don't care what model you use to do the pre-calculations
#32 - col
Quote from Racon :Look-up tables don't care what model you use to do the pre-calculations

Look-up-table(LUT) based systems are notoriously crap at covering the complete range of situations. They're great for a simulation where expected behaviour stays away from the physical limits of the system being modeled, each variable staying within a small range of stable fairly linear values, e.g. simulating a passenger jet in a flight sim where you don't go even close to stalling etc.
A racing sim is the worst possible for LUTs, a driver spends the whole time as close to the physical limits of the system as possible - for this to work well with a thorough LUT system without fudging, you would need astronomical amounts of data. That's why LUT based systems have unnatural feel during physical state transitions. LFS may not always seem 'realistic', but it always feels organic, 'natural' and predictable.
Scawen wants more realism without losing the natural feel. Can't do that with LUTs
I throw the dice and enter one more parameter to the discussion.

I thing that calculations sampling rate of 100 fragments per second may be not enough to approximate the behaviour of a car.

At a speed of 100 km/h about 27cm are being travelled every 100th of a second, thus increasing adjustments to the model.

Vector processing and resampling is a solution but increased sampling rate is the only way to calculate high frequency model changes.

I donnot know if the described issue has already being discussed, so excuse me if already done.

Finally Dynamic Look-up-tables is a way of caching data whenever processing power is available but I donnot thing this is the case in driving simulators.

For example on Grand tourismo they are using several convolution matrixes (drop pulses on real cars) to get samples of total deformations specific vehicles are having during a set of conditions. The rest is a good approximation on a model that gets data from these matrixes !!!

Result is average.

I donnot know what type of magic the creator of lfs has made to physics, if it was just a good guess, or a full blown mathematical model, but I believe he is the only single one that can really give some light of how difficult this can be.

And as time passes, age comes, brain slows down, consider sharing.....
Quote from tsichles :I throw the dice and enter one more parameter to the discussion.

I thing that calculations sampling rate of 100 fragments per second may be not enough to approximate the behaviour of a car.
[...]

AFAIK current tire physics run at 20.000 Hz (the rest of the simulation runs at 1.000 Hz). It's quite an impresive figure with that low CPU usage. Scawen may be having a tough time achieving this again with the new model.
Quote from col :Look-up-table(LUT) based systems are notoriously crap at covering the complete range of situations. They're great for a simulation where expected behaviour stays away from the physical limits of the system being modeled, each variable staying within a small range of stable fairly linear values, e.g. simulating a passenger jet in a flight sim where you don't go even close to stalling etc.
A racing sim is the worst possible for LUTs, a driver spends the whole time as close to the physical limits of the system as possible - for this to work well with a thorough LUT system without fudging, you would need astronomical amounts of data. That's why LUT based systems have unnatural feel during physical state transitions. LFS may not always seem 'realistic', but it always feels organic, 'natural' and predictable.
Scawen wants more realism without losing the natural feel. Can't do that with LUTs

Granularity vs data quantity is always a compromise. If you say that it's a compromise that blocks this particular case then I don't know any better, other than an inkling that the shortcuts that could be had when all the math fun happens in a small area would be particularly numerous.

Time to break out the neural net...
#36 - col
Quote from tsichles :

For example on Grand tourismo they are using several convolution matrixes (drop pulses on real cars) to get samples of total deformations specific vehicles are having during a set of conditions. The rest is a good approximation on a model that gets data from these matrixes !!!


Cool, I didn't think about the use of convolution for driving sim (I've implemented convolution for digital audio processing in NI Reaktor). Seems like that might be a good approach for simulating chasses flex. Maybe not so good for tyre physics though as convolution processing with an impulse response is an inherently linear process.
Another downside is that zero latency convolution is an expensive process even when using FFT approach.
Quote :
And as time passes, age comes, brain slows down, consider sharing.....

hehe, brain slows down, but wisdom increases, and intuition improves throughout middle age
Quote from Whiskey :AFAIK current tire physics run at 20.000 Hz (the rest of the simulation runs at 1.000 Hz). It's quite an impresive figure with that low CPU usage. Scawen may be having a tough time achieving this again with the new model.

I was with the wrong impression of 100Hz so forget that note. At 20khz maybe it is time for multithreading (hehe each tire a thread etc) and a thread orchestrating the communications and results. Transputer like if you remember...
Synchronizing multiple threads at 20 KHz actually takes quite a lot of CPU resources itself.
Quote from Be2K :Then what the problems are ??

Problem one: trying to create a perfect model without flaws?
This is a problem since no simulation will ever be perfect, you will always need some quick patches to make a simulation behave realistic.

But.... If you need too many fixes you might get it never perfect, but a perfect simulation does not exist......

So why is it taking five years, my guess is, someone is trying to be perfect, which is impossible.
#40 - col
Quote from Bluebird B B :
So why is it taking five years, my guess is, someone is trying to be perfect, which is impossible.

I doubt that. I think there are three objectives Scawen has for the new model. When they are all satisfied, he will release it.

#1 realistic setups: A big problem with the existing model is that realistic car setup values don't give correct behaviour. That means that restricting setups to realistic stock values in real cars e.g. Scirocco gave unrealistic handling, and made traction control(?) not work.

#2 consistency: The model must behave 'well enough' in all situations. The original update that was supposed to be released ages ago was found to have an unacceptable behaviour in some 'corner case' of inputs vs tyre physical state. That's why it was pulled. e.g. There's no point in having a sim that feels great unless you hit the brakes at 40.2kph with front wheels turned at some specific angle in which case it bounces like a vimto advert... or the tyres melt etc...

#3 improvement: It must be a clear improvement over the previous model, otherwise the result would be a massive cluster-f@ck amongst the forum haters, and serious jeopardy for the future of LFS. And there's the rub - It would be difficult enough to make a new model as good as the old one but with realistic setups and better consistency. To create one that also has a noticeably more natural and realistic feel is a major challenge.
Quote from PeterN :Synchronizing multiple threads at 20 KHz actually takes quite a lot of CPU resources itself.

The 4 tyres are connected via the body and suspension parts. If those things are sampled at 100Hz, the different tyres can only be synced at 100Hz, so it's not a big problem.
Quote from col :#1 realistic setups:
#2...
#3...

good points, col. i think part of the "magic" of lfs is that Scawen built a tire model around the capabilities of a target PC and discarded anything non-essential that stood in the way of the model working: e.g. motion ration can just be 1.0 all the time, shocks and springs don't have to necessarily have real-world values. as long as the thing feels realistic enough, people will like it because - as you said - having a unified dynamic model makes the sim "organic".

now it seems he has said "well, I got that far..." and is trying to do the opposite, build a tire model based on real life that somehow works on a PC too.

it might be interesting to think: screw the scirocco / "perfectly real"-life physics. it might be possible to evolve the lfs universe by going with the original idea and just adding new things like time of day, rain, a different damage model. you know, things that affect online racing and add depth to what is a great sim as is, even though a VW engineer could easily find flaws...

just saying
Quote from CarlLefrancois :it might be interesting to think: screw the scirocco / "perfectly real"-life physics. it might be possible to evolve the lfs universe by going with the original idea and just adding new things like time of day, rain, a different damage model. you know, things that affect online racing and add depth to what is a great sim as is, even though a VW engineer could easily find flaws...

just saying

Time to fork the uber-realism to LFS2 and get some S3 going on?
Quote from Racon :Time to fork the uber-realism to LFS2 and get some S3 going on?

Nope.
Amazing. So much to say about those 50 seconds.
Quote from col :I doubt that. I think there are three objectives Scawen has for the new model. When they are all satisfied, he will release it.

#1 realistic setups: A big problem with the existing model is that realistic car setup values don't give correct behaviour. That means that restricting setups to realistic stock values in real cars e.g. Scirocco gave unrealistic handling, and made traction control(?) not work.


That is where quick fixes are needed, cars roll/dive too much, tyres heat often too much up, innerside tyre tempatures is too critical, tyres loose grip/blow too fast when overheating, tyre wear is hardly affected by overheating(slightly overheated tyre will hardly loose grip, but will wear out faster). Using quick fixes will not make it perfect, but will make the experience MUCH better because! current tyre-moddeling is very good in fact i think lfs has still the best modelling of all. It just has some flaws, which is normal for any simulation.

Instead of having gamers wait 5 years for improvement, i would like to see some quick patches to get rid of the worst faults in current modelling. Then I would be very happy and go back to online-racing(and prevent me from buying an other race game next month).

Example of what real F1 tyres (pre-2013) can handle:
http://www.youtube.com/watch?v=lCXebCya6Q0
Quote from Sobis :Nope.

eloquent.

i assume you mean that in the spirit of putting the priority on the online racing experience, getting the model closer to reality is better than adding more diversity.

How complicated can a tire model become
(143 posts, started )
FGED GREDG RDFGDR GSFDG