Physic discussion thread

Simplification doesn't necessarily mean you're making steps backwards, the opposite is the case. Of course you need a minimum of complexity in a simulation game as you're trying to do nothing less than coding real life physics into a piece of software. A 5 point contact model is better than a 1 point contact model. Just to be clear. I'm not saying 'throw everything out of the window and make Trackmania type of tyre physics'.

I think the argument is that you don't need to re-build every aspect of reality in the game. As an example you can look at how people started flying without doing it exactly as the birds do. It is much easier to just have some kind of propulsion and use lift.

If you can get a similar or even better result of realistic tyre physics with a simpler method without compromising the physics itself, then that is probably what Einstein was meaning. Overall simulation games will get more complex as the tech will get better as well, but maybe you can optimise certain parts by using a simpler method. Optimisation through simplification if you will.

Anyway, I don't want to talk down what Reiza's doing, AMS2 is amazing and I love it!

 

Not aiming at arguing or attack anyone, I just think some light needs to be shed on what a physical model brings to the table vs the empirical models IMHO.

From what I can gather from my previous involvement with automotive engineering, the conversation over rendering tires behaviors is basically on two levels.
1. Rendering tire performance in terms of pure grip from a global standpoint (where the meaning of global is basically disregarding what happens "inside" the tire) : in simple word which forces is the tire exchanging with the ground in different conditions, whether it's tire type or track temperatures and rubbering conditions and so on
2. How these forces are are transformed into torque on the steering wheel, which bottom line is what we call FFB (although some additional effects are also overlayed later on in simracing).

When it comes to empirical models the conceptual idea is to have a certain amount of data that could be effectively collected through telemetry and adjust the empirical equations parameters to fit the data the best way possible. The conceptual downside of this approach is that the data available, no matter how comprehensive the database, will always be limited to the conditions those data were collected in. So, to make an example that everyone understands: let's take the last Monaco GP, F1 teams will be able to collect data for the race at full car weight under wet conditions and later damp and green conditions for dry tires. There won't be available data for that track (i.e. aero, suspensions etc. settings) in a fully rubbered hot and full tank conditions.
Therefore conceptually, the data you can empirically measure will always be limited to the conditions you encountered and the car settings you were running.
Everything else has to be extrapolated by use of corrective equations with again more parameters that the user/developer must populate either by experience or literature.
Generally, the empirical equations or corrective equations user defined parameters also won't have any immediately identifiable physical meaning, e.g. you have for instance an exponential function whose exponent is a parameter that is determined to match empirical data, but if data are not available you have to set it by experience or literature and in both cases there won't be any guarantee that the result is in fact close to what the real life would have been.
So bottom line, when you go empirical the correspondence between real life and the model is theoretically only guaranteed for the exact conditions in which data were collected and everything else (which is the vast majority of conditions) is likely to be quite off and anyway pretty much based on engineering judgement at best or personal feeling at worst if you are talking about people with less experience like a modder could be.

In this scenario, what a physical model brings to the table is to base the properties and behavior of tires on equations with physical meaning and much more on actual "architecture" of the tire in its sections and components.
Generally the parameters that the designer/developer uses have a physical meaning and could potentially even be measured (although they are not always measured).
Going this route, means the complexity necessary to achieve a good match with global data of the tire grip and forces is far bigger as the amount of variables is generally far bigger and at times the sensitivity of the models to slightly incorrect data can be significant. It generally requires a deeper understanding of the tire mechanics as some of this variables can have unexpected impacts down the line.
However the advantage is that by using a physical model, once your basic parameters are about right, your model will be able to "swallow" much bigger excursions of its variables (whether it's tire/track temps or pressures, or suspensions angles etc.) keeping results much closer to real life than empirical models would do, especially if excursions are big.
Also, you could imagine the advantage in using physical models for very old tires for which there is little to no empirical data available.

Finally, when you move from point 1 (determining the tire forces) to point 2 above (determining the FFB or the "feeling" of the tire at the steering wheel), the accuracy of the forces, the pneumatic trail, the tire elasticity required to pass on a correct feeling to the driver is the highest. Very few percentage mistake on those or the suspensions/steering arms geometry can ultimately affect a lot the feeling and the user will recognize it immediately.
This is really where a physical model can potentially shine: having a far more accurate representation of what happens "inside" the tire, with its structure and elasticity (vertical and lateral) because those translate into recognizably different steering wheel feelings.

The reasons above is why a physical model has higher potential than an empirical model IMHO, and despite RD has a few "invested" guys who keep saying the opposite, empirical models will never be capable to match a physical model accuracy over a wide range of conditions, because they are far more "actual data" dependent to depict any conditions and those data are simply not available.
Yes, the physical models require a very tough learning curve, yes they require a lot of tuning to figure out the best parameters and yes there will always be some of those parameters that developers have to guess by themselves as they are not public knowledge, but what happens beyond that is engineering equations not "artist impression" of a certain matter as is far more the case of extensive portions of the empirical models.

 

I still wonder why Niels doesn't want to mod the physical tire model. It's one of two reasons:

He refuses to move with the times and use a superior (physical) tire model
or
The empirical tire model is superior

Maybe just maybe.... and i am playing devils advocate here .... You cannot emulate a tire by constructing a virtual tire but rather a slip curve and pure calculations comes closer to the real feel of a car tire grip?? I mean i never wanted to be toxic this is why i removed my comments and replaced with "...." but this still really annoys me because is hard to get concrete answers seeing as they compared rfactor 1 to reality back in the day and it was exceptionally close to the lap time as well as driving feel so im confused and thus wondered if it was a marketing gimmick.

How close is close enough?
How close is close enough? – Part 2

 

I do hope the physical tire model is more realistic when comparing to real life lap times and car behaviour.

 

Summarizing to make thing overly simplistic, the empirical model has the upside that the developer can directly "design" the tire behavior to his liking or to the data they may have collected (I suspect this is what Niels may have liked particularly but who knows for sure). The downside is that it's a stiff model: its accuracy gets degraded fast as soon as the conditions in which the model operates are not anymore those under which the data were collected, which is an inevitable occurrence.

The physical model does not give a direct control of the tire grip curves to the developer, but it is rather an indirect product of tire properties set by the developer through certain engineering equations. This is obviously a nuisance for the developer, but allows for much bigger excursion from the data that may eventually have been collected ahead of time while remaining closer to the IRL tire behavior, which empirical models will not be able to do.

Which one is better depends also on the application: if you are an F1 team and all you want to know is which changes you have to apply to your already known setup on an already known track and track conditions and tires because you just finished your Friday Free Practice 1 & 2 you are probably unlikely to enjoy the pros of a physical model. Also you won't give a heck about FFB fidelity in your simulator. All you want to understand is the trend in lap times by making a change in setup.

If you want to sell a commercial simulator that needs to provide a realistic tire/car behavior for a very wide range of conditions and tracks for which data will be covering an extremely small portion of those conditions and you are after FFB fidelity, then probably the physical model is a much better choice in the long term if you are persistent enough to go after its tuning.

For the sake of clarity, lap times are not the ultimate test bench of fidelity of a model. You can predict laptimes for a circuit with knowledge of a car and its properties almost with a Matlab routine alone with a certain accuracy, but that doesn't make Matlab a simracing software.

 

I can't speak for Niels on the exact reasoning, but he sees shades of grey and would be first to say that he doesn't know everything - for certain anyway.

Philosophically speaking I think you're putting too much value on what a single person may or may not think about it. It's not good for everyone's own analytical thinking. You need a lot of it for such a subject as it is very multi-faceted.

Also to cut Niels some slack: it's not so black and white that I'd say he hasn't moved on with the times or anything snarky like that. At the end of the day he has a great system for reproducing cars and tires for the models that he is familiar with. But just because he has a great system for what culminated in AMS1, doesn't mean everyone has to follow him.

Yes the old faithful ISI engine as a whole is indeed a masterpiece in many ways (reminder: Madness physics are an extension of ISI engine). It's not just the tire model, but the sort of physical chassis and suspension model ISI came up with that AMS2 is also still using to great effect. All of it adds up to create a general framework for vehicle dynamics which gets close to real lap times if you tweak it the right way.

Putting aside the tire convo for a moment: it has to be said that it's also limited for consumer simulation without additions. Not to criticise it, it's from 2005 after all. AMS1 already had some important additions like the turbo model that rF1 was sorely lacking, and I'm sure Raceroom has added plenty. Madness added the SETA tire model, a thorough driveline model/solver, barometric pressure, Livetrack, dynamic weather, DRS, and a boost model for different needs and rules (P2P, ERS, turbo pressure override).

These are all things that you don't necessarily need to replicate a real lap time or race pace, because in the end it all works on averages. This is a bit of an extreme example just to underline how misleading accurate lap times can be as a metric for sim accuracy. The same is true when we talk about tires in optimal conditions and within limits and when you throw them into crazy scenarios - which tires go through in practice lap after lap to give us an end result.

In the end with consumer sims I like to think more along the lines of: yes it gets close enough to the real lap time. But it did it on a cold Sunday evening at historic Hockenheim on bias ply tires. What each person values in a sim (race car engineers have veeery different priorities for example) changes how we view each one available.

That is also why there isn't a concrete answer. It depends. You just have to live with that.

 

Live with it or use games with empirical model.

 

The karts at rain are totally disfunctional. The things will strongly aquaplane even within low speed (as 30km/h). Even with wet tires it is incontrollable, as if totally disconnected to the ground.

These folks were running over some big puddles way faster than I and their karts were barely disturbed by that. You need to be way faster to feel like hitting a wall like I did.

A lad racing with a faster kart. It's slippery, but not disconnected. Just see how the guy always had some control, the transitions are gradual and some big&deep puddles barely disturb the kart.

Maybe kart's tires still need some love.

PS: Just as a reminder because it's my favorite car ever (in RL, not the AMS2 version), but the McLaren F1 LM still drives as if it have tires made of glass. In the rain I would be faster driver a Polo and in the dry it just don't match the fame of being a friendly car to drive (in RL a lot of folks reported to use this car as an usual daily vehicle as it behaves as a death trap even if you drive it slow and stead (following real road rules).

Abração e obrigado pelo update!