D3DX9 Vehicle Trouble

[jwhitt]

Hello, I am having difficulty getting a simple raycast vehicle to work in my application. It appears to be a coordinate problem because, among other things, the wheels face the wrong direction and apparently rotate on the wrong axis. It took me a while to realize that if I don't make the car's chassis into a compound it floats away opposite of gravity.

I've tried tweaking the wheel direction and axis vectors with no luck. Having tweaked every value available, I am finally at my wit's end.

See attachment.

[jwhitt]

Alternatively, I've flirted with using a hinge constraint and cylinder system but the car pops wildly while traversing a perfectly even and flat plane. This is fascinatingly frustrating for me.

Is there another way to go about this?

I perused the forklift demo of 2.75 breifly and it appears to be another raycast vehicle implementation. Am I right?

Surely there's another way and maybe there could be demo of that method also.

Sorry for double posting, I'm on my PSP which disallows proper productivity.

[jwhitt]

Spheres... using spheres as wheels seems to work. The sideward sliding is an issue, though.

Since this thread is mainly my own rhetoric, I'll post my progress.

A very rough vehicle simulation w/o raycasts. NOTHING spectacular.

[jrk]

I think you are having the same trouble as I had.

I think it is a bug:

http://bulletphysics.org/Bullet/phpBB3/ ... f=9&t=3963

[Erwin Coumans]

Currently only two configurations are supported, as shown in the Bullet/Demos/VehicleDemo: either Z axis up with X as right axis and Y forward

Code: Select all

		int rightIndex = 0; 
		int upIndex = 2; 
		int forwardIndex = 1;
		btVector3 wheelDirectionCS0(0,0,-1);
		btVector3 wheelAxleCS(1,0,0);

or Y axis up with X axis right and Z axis forward:

Code: Select all

		int rightIndex = 0;
		int upIndex = 1;
		int forwardIndex = 2;
		btVector3 wheelDirectionCS0(0,-1,0);
		btVector3 wheelAxleCS(-1,0,0);

You will need to modify the btRaycastVehicle source code for other configurations.

Thanks,
Erwin

[jrk]

If the change suggested in this post:
http://bulletphysics.org/Bullet/phpBB3/ ... f=9&t=3963

is made I don't think there would be a need for such a requirement.

Thanks,
John

[Erwin Coumans]

Unfortunately this change breaks the Bullet vehicle demos.

Do you have a solution with backwards compatibility?
Thanks,
Erwin

[jrk]

Ooops
Sorry, I thought I checked this and it didn't break the demos.
My bad, I should not have called this a "bug fix".
I'll look at it again, and maybe have a fix that does not break demos.

[jrk]

I think that I've fixed it.
Here is the diff.

< //wheel.m_worldTransform.setBasis(steeringMat * rotatingMat * basis2);
< wheel.m_worldTransform.setBasis(steeringMat * rotatingMat * btMatrix3x3(m_chassisBody->getOrientation()));
---
> wheel.m_worldTransform.setBasis(steeringMat * rotatingMat * basis2);
565,566c564,565
< wheelBasis0[2][m_indexRightAxis]);
<
---
> wheelBasis0[2][m_indexRightAxis]);
>
568,572c567,571
< //btScalar proj = m_axle.dot(surfNormalWS);
< //m_axle -= surfNormalWS * proj;
< //m_axle = m_axle.normalize();
<
< m_forwardWS = m_axle.cross( surfNormalWS );
---
> btScalar proj = m_axle.dot(surfNormalWS);
> m_axle -= surfNormalWS * proj;
> m_axle = m_axle.normalize();
>
> m_forwardWS[i] = surfNormalWS.cross(m_axle[i]);
574a574
>
Pertaining to the changes I made:
I can't understand why the surface normal times "proj" would be subtracted from the original axle direction.
Once you have the axle (right) and normal, the forward is the cross of the axle with the up.

My guess is the original author talked himself into needing the erroneous changes because the first error (described above) was missed.

I did run the vehicle & forklift demo and they work.
(Before forward was back and back was forward, I didn't notice)

All changes were made in btRaycastVehicle.cpp.

I tried to attach a patch file but the forum post wouldn't let attach it, so here it is in the post:

***************
*** 118,124 ****
right[2],fwd[2],up[2]
);

! wheel.m_worldTransform.setBasis(steeringMat * rotatingMat * basis2);
wheel.m_worldTransform.setOrigin(
wheel.m_raycastInfo.m_hardPointWS + wheel.m_raycastInfo.m_wheelDirectionWS * wheel.m_raycastInfo.m_suspensionLength
);
--- 118,125 ----
right[2],fwd[2],up[2]
);

! //wheel.m_worldTransform.setBasis(steeringMat * rotatingMat * basis2);
! wheel.m_worldTransform.setBasis(steeringMat * rotatingMat * btMatrix3x3(m_chassisBody->getOrientation()));
wheel.m_worldTransform.setOrigin(
wheel.m_raycastInfo.m_hardPointWS + wheel.m_raycastInfo.m_wheelDirectionWS * wheel.m_raycastInfo.m_suspensionLength
);
***************
*** 561,577 ****
m_axle[i] = btVector3(
wheelBasis0[0][m_indexRightAxis],
wheelBasis0[1][m_indexRightAxis],
! wheelBasis0[2][m_indexRightAxis]);
!
const btVector3& surfNormalWS = wheelInfo.m_raycastInfo.m_contactNormalWS;
! btScalar proj = m_axle[i].dot(surfNormalWS);
! m_axle[i] -= surfNormalWS * proj;
! m_axle[i] = m_axle[i].normalize();
!
! m_forwardWS[i] = surfNormalWS.cross(m_axle[i]);
m_forwardWS[i].normalize();

-
resolveSingleBilateral(*m_chassisBody, wheelInfo.m_raycastInfo.m_contactPointWS,
*groundObject, wheelInfo.m_raycastInfo.m_contactPointWS,
btScalar(0.), m_axle[i],m_sideImpulse[i],timeStep);
--- 562,577 ----
m_axle[i] = btVector3(
wheelBasis0[0][m_indexRightAxis],
wheelBasis0[1][m_indexRightAxis],
! wheelBasis0[2][m_indexRightAxis]);
!
const btVector3& surfNormalWS = wheelInfo.m_raycastInfo.m_contactNormalWS;
! //btScalar proj = m_axle[i].dot(surfNormalWS);
! //m_axle[i] -= surfNormalWS * proj;
! //m_axle[i] = m_axle[i].normalize();
!
! m_forwardWS[i] = m_axle[i].cross( surfNormalWS );
m_forwardWS[i].normalize();

resolveSingleBilateral(*m_chassisBody, wheelInfo.m_raycastInfo.m_contactPointWS,
*groundObject, wheelInfo.m_raycastInfo.m_contactPointWS,
btScalar(0.), m_axle[i],m_sideImpulse[i],timeStep);

[Erwin Coumans]

Can you attach a zipped patch (preferably using svn, for example tortoisesvn), or use the issue tracker at http://bullet.googlecode.com?

Thanks a lot,
Erwin

[jrk]

Here it is:

[jwhitt]

Thank you, Erwin and jrk, for your interest in solving my problem. I will look into this patch you've presented ASAP.

Bullet's working a lot better for me since my last posts here, though I had moved on without the satisfaction of a decent vehicle simulation, perhaps I can continue with my project. Thanks again.

[gjaegy]

Hi,

I am experiencing an issue that might be linked with that one... My coordinate system is D3D-like, so I am using the following :

int rightIndex = 0;
int upIndex = 1;
int forwardIndex = 2;
btVector3 vWheelDirectionCS0(0.0f, -1.0f, 0.0f);
btVector3 vWheelAxleCS(-1.0f, 0.0f, 0.0f);
m_pVehicle->setCoordinateSystem(rightIndex,upIndex,forwardIndex);

So basically this should be a supported system...

See the video below, and see how the wheel orientation jumps...

http://www.youtube.com/watch?v=k4eZFxR7Las

Is it related to the issue mentioned above ? should I apply the patch, even if the coords system used should be supported ?

Thanks
Greg

[gjaegy]

Hi, it seems the issue came from some of my other code.

Now, the only issue is that I need to rotate the wheel by 180° around Y, but I can live with that

[duncang]

I applied this patch just in case it was related to the issue I was having but no luck.

If I turn the steering left and right they seem to rotate correctly but if i move forward or backwards the wheels rotate on the wrong axis,
Any ideas?

I'm using

int rightIndex = 0;
int upIndex = 1;
int forwardIndex = 2;
m_vehicle->setCoordinateSystem(rightIndex, upIndex, forwardIndex);

and

btVector3 wheelDirectionCS0(0,-1,0);
btVector3 wheelAxleCS(-1,0,0);

Duncan.