Physics [split from network thread]

[rookie1]

With the change to 100fps, at the starting of a race, the cars do not seem to stick to the ground. They are shifting left and right, looks like floating.

[thelusiv]

I didn't notice this. If you'd like you can change it. Look in main.cpp for the line

Code:

#define FRAME_TIME 0.01

0.01 = 1/100 or 100 frames per second. Previously it was 0.004 = 1/250 or 250 frames per second.

[rookie1]

It's quite obvious with an F1 or SB at Kyalami.

[reece146]

After reading the Vamos docs and the stuff that Joe posted above I'm starting to really think a change of suspension model may be in order.

In order to model the "antis" you need a decent side view model that is "decoupled" from the transverse side model. Yes I am aware there is a coupling but a skewed IC model aint going to cut it.

If I am reading the stuff that Joe posted correctly, the instant center model used in Vamos does not vary camber with with suspension movement. This is a shame and is why I suggest it may be time to change the suspension model used.

In an ideal world car suspension would not change camber as the suspension cycled through its travel. But this is very rarely achieved in the real world. It's a shame the Vamos code doesn't do this because for a large number of cars that use solid axles and MacPherson/Chapman struts it would be pretty good.

Can someone post a few links to the other models that are being talked about?

[thelusiv]

I'm not sure if I'd call Vamos' suspension model shameful, just incomplete. As far as I can tell Vamos development is at a stand-still. There are a number of other things we'd like to have more detailed simulations of, and we're more or less on our own to do these.

From what's been said so far, inverse kinetic suspension modeling (IK) is the best way to go. First of all, if LFS really does use it, that's almost reason enough right there. LFS has excellent suspension feel, and also allows for full visualization of the suspension model. Not to mention it seems pretty easily adjustable. I find LFS's suspension setup quite impressive.

Second, it sounds like IK is fairly fast, so the performance doesn't have to be killed for realism. I think perhaps we should take stenyak's advice and not use ODE for suspension (although it will still probably serve quite well for the purpose of collisions and track objects).

[stenyak]

Here you have some interesting readings:
http://www.motorsport-sim.org/mediawiki/...ks#Physics

And don't miss this one, it's pretty good:
http://www.carbibles.com/suspension_bible.html


Camber changes are pretty important in a simulator, it's the reason why i called "up-down suspension" game-ish. Of course this is all relative, depends on how realistic you want your sim to be, and how much time you can put into it.
Maybe a constant-camber suspension is appropriate at the moment, knowing there's not even car to car collisions, which i think are more important if you're thinking on actually racing instead of hotlappging.


Regarding physics freq... the physics rate *always* changes how a simulation behaves. And, given a certain physics model, higher frequency always means a more realistic sim (except for, sometimes localized errors, which are not the norm, but a coincidence). The accuracy (or what we detect as being accurate by watching a screen) of the simulation doesn't linearly increase with frequency, which is what may lead you to think that 250Hz provides the same results as 125Hz.
But, if you take some time to test it, you'll see that a 1000Hz physics engine allows a F1 car to break its suspension over a bump, while using 125Hz the F1 car *might* not even feel it, tunneling through it. (this example can be avoided using certain techniques in collision detection, or even variable timestepping, but serves for the example).
There are some attempts at continuum collision detection, you might want to check the Bullet library. I haven't tried it personally (other than a few demos), but it doesn't look bad.


Regarding the tire model, as you point out, simply using certain algorithms doesn't mean the simulation will be realistic. ISI sims are known to use pacejka-like tire models, but use very wrong data for certain slip angles (i dont' want to start a flame war, it's just the way it is), which makes the sim be less realistic at times (or always). After all, it's very hard/expensive to get data for tire models, so it's usually about guesstimating values all over the place. And of course, there's no way to check the resulting simulation is accurate. This is what makes car simulator creation a nightmare when you want to reach certain levels of realism :lol: .

As usual, sorry for the offtopics; maybe a couple thread forks are in order?

[stenyak]

A small quote from the car bible suspension page, so that you know how important proper suspension geometry simulation can be:

"In 2006, OnCamber LLC patented their variable camber steering system which they launched at SEMA in Las Vegas.
[...]
On track, this system has shaved 3 seconds off the development vehicle's lap times in race conditions."

I think Mercedes developed a similar suspension for their SLK, adapting camber on the fly, so that the tire stays as perpendicular to the ground as possible always.
Edit: not an SLK hehe. http://www.topspeed.com/cars/mercedes/me...r1214.html

[thelusiv]

Thanks for the excellent links, stenyak. I wonder if we could simulate camber change in suspension by making the hinge points on the car between each set of wheels, so that suspension compression of one side will mean negative camber is increased, and extension will increase positive camber. It may still be tricky to get this right, but wouldn't that be more correct? Or have I got it backwards?

VDrift's tire model is indeed difficult to tune, and many of VDrift's tires have simply been copied from the tires the Racer cars came with when they were imported. It wouldn't hurt, on down the road, to experiment with alternative tire models and see if perhaps a simpler model would work as well and be easier to tune. For now I feel that what we've got isn't all bad, so I'd say upgrading the tire model is a low priority.

[stenyak]

Depends on which are your priorities. I don't know which are the goals of vDrift.

For example, nKpro is very focused on pure physics realism. Not knowing how the internal beta testing version Drivers Republic is, right now i'd say nKpro is the most realistic sim on the market.
ISI sims are more focused on reaching a larger audience, keeping it realistic. It has way more buyers than nKpro, but is less realistic (again, i don't want to start a flamewar, that's just the way it is).
LFS is focused on both things. It's very realistic but also has many "game" features.

So depending on what is vDrift goal, you'll want to implement a new tire and suspension physics now, or leave it until you got a "championship mode" working.

Personally, i wouldn't try to trick the vamos physics engine into behaving like something it's not supposed to simulate. I just don't like faking things, i prefer to do it all The Right Way ™, which is sometimes a suicide due to the complexity :lol: . But seriously, i think an IK system for suspension should be coded, instead of hardcoding values into vamos suspension system so that it behaves like every of those suspensions. Of course, supposing the tire model is perfect, which it isn't, so it may be more efficient to devote those hours to improve the tire model (which accounts for a larger percentage of the perceived sim realism).

I know, these are all just words; I, myself, don't know how to to code the IK thingy (i wish i hadn't chosen to study computer engineering..), but if you think you have the knowledge (or the time to learn the maths/physics involved), i'd say go ahead and forget about current suspension model.
Doing it The Right Way ™, you wont find yourself scratching your head when you later decide to include the 6-link suspension of whicever model, with rods positioned here and there, instead of over there and over here, which will anyways not be correctly simulated in all situatins even if you manage to get somewhat "correct" data.

Edit: this is not only about correct camber values (those could be precomputed using some 3rd party program like blender, a table with suspension compression vs. camber&distance to longitudinal car plane), but also about "toe", "caster", and front wheels rotation when steering (ackerman angles, for example)... basically, the whole 3d position and rotation of a wheel in all situations.

[reece146]

I going on the premise that no decisions have been made on this...

Does anyone have a link to the IK stuff? I did some quick googling and didn't find anything relevant.

[joevenzon]

I don't know anything about it besides how IK works in character animation... I've never heard it mentioned in relation to car suspensions except by stenyak, above.

I think the suspension model is alright at the moment... if we were going to change to something better, I'd want it to be easier to tune (perhaps by using real-world values) than the current method.

[stenyak]

IK is basically solving an equations system, so that all distance between certain points (edges of rods) stay constant (unless rods are elastic, like suspension turret).
Other than that, i have no idea how to implement it. Having a background in physics and maths would certainly help a lot here.

[thelusiv]

Here's a big list of comparisons of how lots of different auto simulators work, including lots of physics details: http://en.wikipedia.org/wiki/Comparison_...simulators

[stenyak]

Be careful with that list, many things are wrong; some people who don't really know, decide to fill with "No" instead of leaving "Dunno".

[thelusiv]

Yeah on a closer look I see what you mean. It's almost a helpful list though. One thing that's interesting to me, though, is how many sims use Pacejka or a Pacejka-like tire model. Of course most of them also buy proprietary tire data...