What exactly does Convex Cast return?

[Marc]

When I cast the upper shape through a world with two larger bodies like below:



Will I get
0. contacts for both larger bodies or
1. just for the upper one because the upper one stops the shape before it could reach the lower one
?

Also, what's the difference between NewtonWorldConvexCastReturnInfo::m_normal and NewtonWorldConvexCastReturnInfo::m_normalOnHitPoint ?

[pHySiQuE]

It should record all collisions.

I think the normalonhitpoint is the normal of the hit surface, which isn't always the same as the normal of the collision.

Note that Newton 3 presently is not reliable with convex casts against collision trees.

[Marc]

It should record all collisions.

Hmm, what are all collisions? For example, in the case from above, is it option 0 or option 1 ?

I think the normalonhitpoint is the normal of the hit surface, which isn't always the same as the normal of the collision.

Can you define the two? Or can you give an example when they are not equal?

Note that Newton 3 presently is not reliable with convex casts against collision trees.

I'm still on 2xx.

[pHySiQuE]

It should record all collisions.

Hmm, what are all collisions? For example, in the case from above, is it option 0 or option 1 ?

Both. If you're doing anything fancy, loop through all results and find the closest collision that occurs.

I think the normalonhitpoint is the normal of the hit surface, which isn't always the same as the normal of the collision.

Can you define the two? Or can you give an example when they are not equal?

If you run into a wall at an angle, the collision normal will not be the same as the direction the wall faces.

[Julio Jerez]

NewtonWorldConvexCastReturnInfo::m_normal is teh normal forme by teh mutual tangen of teh tow collisding shape at teh point of impast

and NewtonWorldConvexCastReturnInfo::m_normalOnHitPoint in teh nmorla of teh hit point at the point of impact

example say you can a sp[ager again a polygon, if teh spher hi on teh flat area, them m_normal and m_normalOnHitPoint are the same becaus ethe tangent of botth suiface are the same
however if the shape hit teh polygon in and edge or a vertex, tehn the norlam is teh norla that on teh suface of the sphere, but m_normalOnHitPoint teh norla on the plane

[Marc]

Hmm ok. Thanks for clarifying on the normals.

I experience something I still can't explain though. I have this situation:



For some reason, I get contact A with the correct normals, but I get contact B with some normals I can't explain. I would the contact normals of B point upwards - but they don't. Note: the normals of B are correct if the box is not there. But if it is there like that, I get different normals for B. Does the existence of contact A have an impact on contact B ?

[edit]
I'm not sure if it makes a difference or not, but it might actually be more like this:



The slope is a static body and the big box is dynamic, but rests there. If the big box and the slope touch like this, will I get contact B?
[/edit]

[Julio Jerez]

if the litle box is casted in the direction of the arrow, and box A is in the way, them you you only get contact betwinn box A and the litle box.
that is what the definition of convex cast means. If this is not how it is working then they is a bug.

[Marc]

Ok so just to make sure: The convex cast is different from the raycast. With the raycast I can get B as well, depending on the value I return in my callback function.

What I still don't get: The convex cast does not always just return one contact. In this case from above, what additional contacts will be returned besides A? If it doesn't report all contacts like the raycast can and it does not only report the first contact, what are the others?

[pHySiQuE]

if the litle box is casted in the direction of the arrow, and box A is in the way, them you you only get contact betwinn box A and the litle box.
that is what the definition of convex cast means. If this is not how it is working then they is a bug.

My mistake.

[Marc]

if the litle box is casted in the direction of the arrow, and box A is in the way, them you you only get contact betwinn box A and the litle box.
that is what the definition of convex cast means. If this is not how it is working then they is a bug.

Ah I just wanted to clarify I'm not saying something is wrong or buggy with newton. I just want to understand what to expect from the convexcast so I can predict what I'll get from it in certain situations (ignoring any current potential bugs).

I think what I experience is what Julio describes: I get contact A but I don't get contact B. The thing I'm still wondering about is, I get 3-5 contacts when I move around a bit, but never 1. What are those additional contacts? When the convex cast "casts" the small box downwards, it hits the large box and creates contact A, what happens from there on? Or does it work completely different and I'm thinking wrong? (And actually, I'm not sure if that matters, but the large box is a large box while the small box is a cylinder from the side.)

[Julio Jerez]

when to convex shape interset the intersetion point define a plane.
depene of the deatore of eth shape the first cone is contact you will get:
a single vertex/vertex, vertex/edge, or vertex/face , (for a conic shape like a sphere of a capsule)

or and edge/face or face face (for shape with hard edges like boxes) in this case the colliding in more that one place.
ex twp boxes colliding face to face. all teh contact fall in the same plane

[Marc]

all teh contact fall in the same plane

Ah, I think now I have an idea of what it returns. Thanks.

Is there a way to make it behave like the raycast? I could repeatedly call it and build a list of hit body and filter them until I hit nothing - but that's probably inefficient. [edit]Thinking a bit about it, I guess this hack/workaround does not necessarily have to work if the bodies are shaped appropriately. [/edit]

[Julio Jerez]

but the convest cats does works like a ray cast. the paremter t is the time at wich the contact happens. you cna use that just liek you use a ray cast. I do it many time.

[Marc]

but the convest cats does works like a ray cast. the paremter t is the time at wich the contact happens. you cna use that just like you use a ray cast. I do it many time.

Hmm. How can I use t so that I can get contact B as well then?

At http://newtondynamics.com/wiki/index.php5?title=NewtonWorldRayCast it says:

An all body ray cast can be easily implemented by having the filter function always returning 1.0, and copying each rigid body into an array of pointers.

But there is no such callback for the convexcast? The convexcast only has the NewtonWorldRayPrefilterCallback, while the raycast has NewtonWorldRayPrefilterCallback and NewtonWorldRayFilterCallback and the later one allows to return values from 0.0 to 1.0, where a value of 1.0 results in an all body ray cast. Is it possible to do that with the convex cast as well?

[Julio Jerez]

Because it is a convex cast, it will find the first body along teh trajectory that his the casting shape.