Automobilista 2 V1.5 Physics Development Update

Real world data is very important. The challenge for a lot of devs is how that data is interpreted using the physics and tire model engine in game. What we the users call the handling and feel of the car in game. The problem iRacing has, is that their lead tire model programmer (Dave) is trying to use a physical model (all formula based) vs an empirical model (formula and look up tables) which most other sims use.

And unfortunately the physical model, no matter how complex, is lacking in feel and realism because it can't properly simulate what happens to the tires as it breaks traction and starts to slide. Bottom line, not matter how complicated he's made his physical tire model formulas over the years it hasn't passed the sniff test for a lot of us that have had the privledge to driving cars to the limit at race tracks. Cars aren't that hard to drive at the limit in the real world, unlike iRacing portrays. Meanwhile people that drink the coolaid, keep giving iRacing money year after year for a subpar product IMHO.

On the other hand look at Kunos. Their sim is also based on real world data, but they understand the importance of feel in the sim. If the data they put into their physics engine doesn't correctly translated to the right feel, they update and modify their physics and tire model engine so that is does feel correct. They don't modify the real world data from telemetry. So the problem isn't if real world data is used but how it's correctly translated into the sim to create the correct feel. Saying real world data isn't needed is a fallacy.

 

AMS2 uses a physical model and it is better than both, of course is more complicated to get right a more realistic model but that is what we all want in a sim right?

 

I'm sorry to disappoint you, but perhaps what you miss is that unlike other vehicle simulation parameters, namely suspension, chassis, engine, and aerodynamics in part, it is precisely the pneumatic model in the field of simulation that is difficult to reproduce as a mathematical model or exact science. The parameters of manufacturers such as Pirelli or Michelin itself have their own simulation model, and they try to simulate reality but it is not a 100% exact science. currently Michelin itself uses a very advanced system to simulate their tyres, and the AMS2 itself has a very similar system but forget this thing about real tire data, the data goes so far to represent that tire in the simulation.

In a previous post of mine, I posted about the field of simulation at a scientific level to try to reproduce it in car simulation, you will find very exhaustive what I mean, and above all it is clear that it cannot be represented with reality simply by taking data from manufacturers as many claim.

here and here

 

This is peak physics ignorance. Real life data aren't that good for simulation unless they are info on construction and properties that are measurable physical properties that go into a physical model.
Real life data you are talking about such as accelerations, vertical and lateral loads, steering angle etc. are close to useless in order to achieve the "genuine equal to real life" tag you are talking about for many many reasons, but if nothing else because they are only valid in the conditions they were recorded, many of which aren't fully known: from track rubbering to wind, to spot track temp and so on.
Also "real life data" are clouds of points with variability due to a bazillion different matters at the same time: from traffic, to tires evolution, to track evolution to purely statistical variance, to driver not doing exactly the same thing and others. Whatever line of deterministic values you draw ends up being arbitrary from the developer as variables cannot be separated in the data variation by variability cause without again making assumptions and using approximations where the developer has the last and far more important word.
That means that any data you gather will still require a lot of developer craftmanship to close the holes into the information available (where the unknowns are far far bigger than the known points).
Posts like this are really peak nonsense marketing words that show how Kunos brainwashed users through a bunch of streamers.

 

Real-world data is required so you can have an air of credibility and users cannot dismiss your simulator. Then when your chosen tyre model drives terribly (ACC circa 2018) you go and tweak all the values to be completely different so that you have a driveable sim (ACC circa 2023), but the users still are under the impression that it's "all based on real-world data".

 

Pretty sure the AMS2 which uses the Madness SETA tire model is Empirical based (using complex formulas and look up tables). Because it's almost impossible to simulate a tire correctly once they start to slip using just formulas due to the complexity of what is happening. From my understanding it's why the empirical model seems to have an edge on the pure maths based physical model that iRacing is persuing.

 

A while back I read an article on an interview with Marco (Kunos) where they talked about exactly this. It's been a while and I don't remember what site I read it on but they did mention that instead of tweaking the "real world" values to work with their current physics engine and make ACC feel good. They would tweak the physics engine side of things, leaving the real world values alone, to make it feel correct for the testers.

I'm not race game dev and what they are doing is it some very complex stuff beyond me. But if it is true, hats off to them for going about things the hard way.

 

You have quite a confused idea of what is an empirical model vs a physical one.
A physical model is a model based on variables that have a physical meaning combined into formulas with physical meaning to provide an output. This generates a set of matrixes that get used as database for the quick computation purposes.
There is still a limit to what you can measure physically and what you can address with physical variables and therefore there is still a degree of craftmanship even here.
An empirical model does not use variable with physical meaning, they use formulas with parameters that have NO physical meaning to approximate a target curve that either the user recorded in a given condition or just suits the taste of the developer. Changes to the said parameters do not immediately picture a recognizable change in the physical tire, i.e. you change the parameter to an exponential formula but you can't tell what that change would mirror in the real world tire. The results are again put into look up tables also in this case.
The degree of craftmanship and arbitrary choices from developers in this case is far higher if not almost complete.
Secondly the limit of this approach is that it becomes increasingly "wrong" as soon as you move out of the "recorded points" and it is extremely difficult to make it properly predictive as usually the corrective formulas for non physical models are even less accurate.
In other words, even assuming you are perfectly right in the points you measured as soon as you are not on those your accuracy drops far faster than a physical model.
FYI, no sim really perform tires calculations in real time.

 

LOL that is 110% marketing BS... If he said that, it has no meaning whatsoever. Just pure BS.

 

Maybe I've confused which is which as I'm not a sim racing title dev. I'm more interested in what correlated with my real world driving experience and less about what kind of complex coding goes into these titles. I leave these things to the smart guys and tend to believe what they tell me (perhaps naive).

So help break it down for me in layman's terms.
Which one model uses purely formulas? Physical or Emperical?
Which one uses lookup tables to transform?
Which model does iRacing use?

 

All use formulas at some point: the difference is one uses formulas with a physical meaning behind (variables are related to the physical properties of tires to a very large extent) like AMS2, iracing and RF2 do, the other uses parameters that are set into certain range by experience but can't be measured or reconducted to physical properties (AC, ACC, RR). As such when you adjust the variables "to suit" you are not really sure about what you are assuming that would change in the real tire. There is no more link between your change and the properties of the tire.
Lookup tables are numerical calculation enablers that take the above mentioned results (whichever approach you use, physical or empirical) and pre-calculate values for a large number of variables values. All simulators use them, like really all. The purpose is to avoid real time calculation based on the instantaneous inputs so that you don't need a NASA PC to run calculations real time when you can pre-compile all the values you need.
iRacing uses a physical model.

 

Also another point worth knowing: usually physical models by construction use a large set of nodes in the thread as points at which the equations are solved and the formulas calculated. I have not seen first hand what developers do, but I believe I read RF2 or AMS2 physical models would calculate over something like 30 nodes (if someone has first hand numbers is welcome to correct this), while empirical models by nature have a very limited amount of so called "points of contact", I believe in the range of 3-5 over the contact patch.
This is no small matter as the inherent higher number of "points of contact" in the physical model has a decisive potential advantage over the empirical model in terms of proper organic feeling of the tires as well as smoothing out calcs over step changes in surface like kerbs, aprons and other similar situations .
All this being said, whatever model you use if it is not well tuned it will not give good results, which is what SMS skipped before jumping ship and Reiza has spent years correcting.

 

Tested Just F Ultimate Gen2 in 1.5.3 built.

Amazing. Now the tyres have less margin to keep the grip in a total control agressive slip, but when you loose rear or front, the loose and regain of grip I feel much more natural and progressive.

Now you can't have a lot of lap time drifiting a F1 car, but can save one from crash into the barriers sometimes.

 

In the development of the SETA tire model, SMS in the first pCARS used 32 knots, in the second chapter I remember that we were talking about this of the new tire model with 128 contact points. Unfortunately internal WMD forum is closed and you can't check it again. I think the only ones who can give us some info is @Renato Simioni

 

Sure.. Well I prefer to have my sim feel like every other sim instead being an outlier.

 

Either way, "better" can mean different things to different people so . . . probably a pointless discussion until (never) we settle on which title is the true winner that sets the bar.

 

Why is it that whenever you approach a corner and you have max steering angle and you are literally understeering since there is no more grip to turn in more, you still can just steer more than a quarter of the wheel and suddenly there is more grip and the car turns in. Fastest way for me around tracks is just abuse this and just turn max in every corner regardless of what understeer im feeling.

And why does every car slide, there is never a point when you approach a corner with full grip and then it progressively looses grip, you always seem to have some controllable slide in every corner.

 

What makes you think that a car at a high steering angle should turn less than when it's at a quarter of the steering input?
Also, by definition of understeering, if you can still turn the car and even faster than before it means you are not understeering really, but it's rather your perception of it. Tertium non datur.

 

Any other game or karting there is a point when you turn the wheel nothing more happens, you can turn it until the wheel literally locks, nothing will happen with the tyres. But in AMS2 to a point obviously it still makes the car turn in a bit more when you naturally would already be fully on the limit of what the car can turn anyways.

 

On some level, I might agree but, can we be sure if we haven't had RL experience with those cars under the very same conditions . . . and what happens to the tire temps when you repeatedly abuse them by over-driving?