Forums

Full Version: Directional stability
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
I've spent some time trying to figure out the reason for unstable car behavior during braking.

My first suspicion has been tire lateral force symmetry. Making sure that the tires are left right symmetric hasn't fixed it though.

Next was the center of mass offset due to driver position. Adding a balance weight still didn't help.

What I've learned though is that the effect is caused by positive feedback of current slip angle. As soon as there is a minimal slip angle it will generate forces that depending on center of mass and weight transfer(braking) will act to increase the slip angle.

Why does it not happen with real cars? From my current understanding the reason is steering caster which allows wheel self centering and gives cars/bikes directional stability, counteracts to slip angle variations. VDrift doesn't simulate this self centering, the wheels are seen as fixed relative to chassis at any time.

How to fix it? I had some success (have got the CO to stay stable during hard braking) by carefully adjusting the toe and camber values of the front and rear wheels. But it also means that with any change to camber, toe or center of mass one has to check the car for directional stability, which is not optimal.

What I haven't found out yet is whether the reason for the initiating slip angle is due to numerical instability or some simulation bug. I can clearly affect the direction by messing with front and rear camber/toe, make it pull to left or right.
While comparing current with new tire model I've noticed that beckman force combining method is only sampling the positive side of the force curves. It doesn't support asymmetric tires, neither can it handle camber effects properly from what I see.

What does it mean? The direction instability issues will be gone. I am looking for a way to keep camber effects but will have to disable them for now.
I think I've got something to make camber work properly with beckman. Will upload a build today.
Congratulations on solving the puzzle! Smile
Sounds cool.
So does this mean the new tire model is now better than the older one and it will be used ?
So far I've seen it has fewer parameter for curve generation but much more for combining both. I guess that similar cuvers can be generated right ?
Can you tell what changes in terms of cars handling and game feel ? Any idea how that relates to gravel ?
I've understood only few parameters in Pacejka's Magic Formula and it still feels magic to me Smile I mosly had this problem that there were many parameters, but none of them was changing what I wanted.
The new model is strictly optional. You have to compile with VDRIFTN define to enable it. It is better in the sense that it is using a data set from a tire manufacturer and includes combined forces. So if anything is behaving incorrectly I can at least be sure that the tire forces are correct. My ultimate dream is to reduce parameter tweaking as much as possible, just plug in some realistic values and get somewhat realistic car behavior.

It has got more parameters, but at least the naming is more logical. The aligning torque is a bit too complex now imho, still don't fully understand what they are doing there. But I guess it is a rather complex effect, thus the complex model.

The handling will ultimately depend on the parameters. The data I've got has less grip and a bit lower ideal slip angle, but better combined handling in general I think (full combined model vs beckman).

To get a better feeling for the parameters I am trying to look at the main variables (this is better with the new model):
Example old model Fx = D * sin(C * atan(B * S - E * (B * S - atan(B * S))))
D(b1, b2): Peak factor, force maximum (scales Fx)
C(b0): Shape factor, falloff after peak
B(b3, b4, b5): Stiffness factor, slope towards peak
E(b6, b7, b8): Curvature factor, width/smoothness of the peak

I think you could get away with a simpler model, by using fixed C, B, E. But with D(b1,b2) as the force should scale with tire load.

I am not sure what is the best way to deal with gravel. I remember seeing some curves in a paper, will see if I can find it. I think it had reduced slope, smooth peak with low falloff (or almost no peak). Should result in quite slippy but smooth handling I guess.

I'll upload my tire tools soon. They allow to load a reference data set, which I found quite helpful when tweaking values. And I have got a version which allows loading old tire data as reference to compare it with the new model.
great news this... and to test the VDRIFTN define, have you got a sample tire parameters?
(05-18-2013, 10:37 AM)ghiboz Wrote: [ -> ]great news this... and to test the VDRIFTN define, have you got a sample tire parameters?

Pull the latest data. The game will load new tire data sets instead of the old ones when built with VDRIFTN.

There are four sets in data/carparts/tire:
vsim.tiren: original tire data
touring.tiren: a bit more grip, smoother peak compared to vsim
sports.tiren: 20-25% (I think?) more grip than touring
racing.triren: 40-50% more grip than touring
Tire data editors are available here (for now): https://github.com/logzero/vdrift-tools

Consider them beta quality atm. Parameter limits need adjustments (coeff_min and coeff_max values in the scripts).
great thing....
a question:
the tiren is the new... tire no? and wich version of python you use? with 3.3 gives some errors loading the files...
tireno.py loads old tire files as reference, will add some text to the readme. I am still on 2.7, will have to check why 3.3 doesn't like the scripts.
thanks Nan...
last question, the scale of the axis, what is?
(05-22-2013, 07:34 AM)ghiboz Wrote: [ -> ]thanks Nan...
last question, the scale of the axis, what is?

Ah yeah, might be a good idea to add some labels. The scale is 30 degrees for slip angle and 1.0 for slip and (normalized) forces.
thanks!
http://screencloud.net/v/4YYq
in this, for exaple fy is almost 0.5 or 1.0?
1.0

And I have to correct myself, see now what might cause confusion. Slip and slip angle have a scale of 0.5 and 15 deg, whereas forces are 1.0 (gray grid). I'll add labels asap.