What limits vehicle physics ?

This is a question for the developers. It's something i've often wondered.

What actually limits the quality of vehicle handling physics in games ? Over time, slowly but surely, handling models do improve in racing sims. But it seems to take a very long time. There could be several factors, and I don't really know which one is correct.

1) Is is it computational power ? are there things devs want to implement and could implement but that could never be run in real time on a gaming PC with current CPUs/GPUs ?
2) Is it knowledge based issue ? is it a simply an issue of not enough people happen to know all the complicated mathematics and physics behind car handling while also being competent programmers ?
3) Maybe it's just time ? I can't imagine there is anything quick about making a realistic handling simulation.
4) Game engine limitations


Plenty of games claim to be realistic. But they also feel inherently different. The cars in rFactor 2 handle differently to those in iRacing. But both are considered very realistic. Which brings up the question, if they really were realistic.. should they not both have similar feeling and handling to vehicles? which kind of points to either one of them being wrong, or both being wrong (in terms of how they simulate certain things) - probably the latter.

I was running off on a tangent towards the end with regards to the question in hand, but whatever. I'd like to hear opinions (or preferably facts!) on the matter.

edit:

To clarify, CPU power is obviously going to limit some things. You're not going to be able to simulate the molecular excitement and expansion of the rubber molecules as they heat up. I get that. But that kind of simulation is what's going to be required to finally get car simulations physically perfect way off into the future. I'm just concerned with the meat of the simulation, and it being 98% correct most of the time - you don't need that kind of crazy detail to achieve that. I never expect to see perfect true to life reproductions in my life time.

 

The beam and nodes system in BeamNG for one require a lot of computational power, but the most limiting factor in games, I find personally, is and remains the fact that everything needs to be programmed beforehand.
I'm programming a top-down racing game and I've got a huge amount of trouble getting the physics to work, let alone work properly.

If you want an extremely sexy real-as-real-life racing game it has to be programmed. Every single force acting on the car must be programmed, every reaction the car is supposed to make to that force has to be programmed.
Programmers, humans essentially, are the limiting factor when it comes to realism.

 

I think the best way at the moment to simulate true-to-life physics is using the node and beam structure from BeamNg/Rigs of Rods... Real suspension is made up of various "beams" of metal hinging from eachother. You simply cannot replicate that with pre-coded values. On top of that, real car bodies flex their steel a little during driving, something that doesn't happen in any other game as far as I know. Problem here is that you need to, both make the PERFECT n/b skeleton, and need to get the values for the beams and weight of nodes 100% accurate, which is very unlikely to ever happen in any n/b-based physics simulator within the next 20 years, as far as I can see... Yes, people will get close in BeamNG, but 100% accurate is hardly possible without sophisticated measuring devices.

Once you have that done, I believe the best way to continue is using the Live for Speed tire model. As far as I know that is the only game that simulates tires as they are in real life: With wires running through them.
Yes, I know it's probably not the most realistic game out there, but they simulate tires in the right way. That means, they *can* eventually get to a true-to-life-ish tire model once all the values are 100% accurate. I believe Live for Speed also uses beams to simulate the suspension, but it doesn't have chassis flex and the values they use are probably not 100% correct for a lot of stuff. And there will probably never be a similar tire model used in games that's actually using this *and* uses the right values for litterally everything...


Then there's the engine and gearbox coding... This *might* be the simplest thing to get right, since I think it will be easier to get this very close to being right than it is with the other parts of the simulation. I'm not saying it is easy, but I think it's easier than it was developing the LfS tire model, for example...


So now imagine how taxing BeamNG probably still is on the CPU on it's own... Then realise that the LfS game isn't exactly "Pong" for your computer either. Then combine both these types of physics... You can see where this is going: All gonna be very hard to simulate. I would imagine within a few years it will be possible for high-end computers to combine both though. However, with all the data having to be correct, it will be even more taxing on computers. So it might take another few years for it to be completely possible, I guess.

Long story short: Computers will probably be able to get very close, at least in like 5-10 years. At the moment it's just VERY hard to get data from real cars to make both an n/b that's extremely accurate AND a tire model that's extremely accurate, both for suspension positions, spring/damping rates, and all that stuff.

And there's also not enough knowledge to do this, and people who do have the knowledge AND access to a ton of different cars(creators of Forza and Gran Turismo, for example) usually don't really care to get it right.
I am sure that within 5 or so years it SHOULD be possible to get pretty close to real life though... I hope.

 

Tyre model is indeed the most limiting factor. That's why I've LFS has always felt the most realistic to me. RFactor 2 seems to adopt some of their model, but I haven't played it.
When you have a good tyre model, IMHO next think to make a big difference in handling is flex in suspension components. I take correct geometry for granted. Way after that may be flexing chassis.
Unfortunately it's really not possible to simulate tyres precisely. It has so many components and many factors that affect adhesion. Complex FEM is not doable online, so you're left with measuring as many characteristics as possible and model rather results of real physics instead of "the real physics", if you know what I mean. Nevertheless, it can both perform and feel close enough. You know, even driving similar road cars may feel quite differently.
What I dislike about N/B wheels is their mass distribution. Most of it is on its circumference giving it huge moment of inertia.