Cylinder wheels question

[CrystalH]

This is just my question, not a feature request or anything.
VDrift has still 1 raycast for 1 wheel right ? This is also in Stunt Rally.
Of course it works good for small wheels hidden inside car.

I was recently thinking about having cars like a buggy or a monster truck in SR, which have big wheels that can collide with anything on track.
Do you have any advice on how to implement it ?
There is a convex/cylinder sweep method working similary like raycast but IMO this isn't yet what's needed. For example it wouldn't work if I flip a monster truck to side right ? Sweep works good if it's cast down not if the side or top of wheel already collides.
There was something in older topic, I can't remeber, but you did try cylinder wheels before ?

I think the only good way having those would be just to have usual rigid body cylinder shapes. But then is it possible to get their contacts with other collision objects from bullet, somehow filter them to get only 1 contact and input this into tire simulation etc. ? Would it work ?
And since I want to use suspension form VDrift not a joint in bullet, I would have to set this wheels shape position on every simulation update. Which makes me think what will happen if it's set already inside terrain, sounds dangerous, although there is skin width margin for colisions.

[NaN]

VDrift has been using cylinder wheel shapes for collisions only for quite some time already, since I've added fracture code I think. It is very primitive though and is certainly an area that needs improvement, won't work for monster truck type wheels I think. I've got a few ideas but not the time to implement them atm. Other things have higher priority.

[CrystalH]

Sure, I'm just asking because I want to add this in StuntRally.
I've seen in code that you add cylinders for wheels:

Code:

        btCollisionShape * shape = new btCylinderShapeX(size);

I can't find though where do you set its position or rotation ? Does it move with suspension ?
It does rotate with steering angle, I've seen that in game.

The WheelContactCallback is interesting, do I get it right that it ignores contacts from the upper half of wheel ?
Is it a possible implementation (could it work) to have wheel cylinder contacts (from bullet callback) feeding wheel contacts in cardynamics ? (ofc in SR, not in VDrift, since you have bezier patches).

[NaN]

If you have seen it rotate with steering angle you should also have seen it move with suspension .

Push the wheel cylinders into your compound collision shape object first. This way you know that the first 4 shapes are the wheels. Update their position relative to compound shape each time you update your wheels.

My original idea has been to get rid of wheel rays by using wheel contact data, but more for performance reasons. Just make sure that the wheels are always in contact, let them sink into surface by 3-4 cm or so. The problem has been the normals reported by bullet, if I remember correctly. It didn't handle mesh edges very well.

I'd say give it a try. Maybe it'll work better for you.

[CrystalH]

Ah so wheel shape pos/rot is being set here:
CARDYNAMICS::UpdateWheelTransform()
body->setChildTransform(i, btTransform(rot, pos));

Hmm, so cylinder wheels are shapes inlcuded in chassis rigid body.
This means if I collide with wheel, it acts just like it was chassis.
I.e. no unsprung mass from wheel acting through suspension on chassis.

Would it be possible (or make sense) to have wheel cylinders as dynamic rigid bodies, get their contacts from bullet callback, feed one of those to tire contact, compute simulation like always, then get wheel position from suspension and use it to set wheel cylinders pos and rot.

Can we maybe use a computed normal by just vector of contact point - wheel center ?

[NaN]

What is the point of having them as rigid bodies if you'll override their contacts and transform anyway? The normal should be the surface normal at contact.

---

What you could try is, getting the contact face(triangle) and calculate the normal yourself.

One idea I had was to use the normals of the mesh and interpolate the contact normal. This should give you smooth normals as a bonus, assuming your mesh normals are smooth too.