Hi all,

I am currently implementing a 3D interaction technique to manipulate objects. The first step is to enable 2DOF translation across the camera plane. The basic approach I used it to compute the intersection of the camera ray to the camera plane. I then use the intersection to determine where to place the object to. Just for clarity, if the user starts dragging a cube from, say the bottom left corner, then the offset will be taken into account.

Now, I'm not sure if there's anything wrong with my code but there seems to be noticeable variations in the resulting plane intersection point even though the original mouse ray comes from just a few pixel apart. Here's a brief output from my log to show what I mean:

[11:06:08] Info (UI) P[655,553] - T[0.22,0.09,0.00]
[11:06:09] Info (UI) P[655,552] - T[0.22,0.27,0.00]
[11:06:09] Info (UI) P[656,552] - T[0.24,0.27,0.00]
[11:06:09] Info (UI) P[656,551] - T[0.24,0.18,0.00]

As you can see, the second line is just one pixel above the first entry however the resulting intersection is 0.27 instead of 0.09. Visually this results in the object acting like it is jumping around. I hope this has something to do with precision issues. What I think would help is that if instead of using an absolute position to move the object each frame, it would be better to move the object towards the new position, instead of "teleporting" it to the new location.

I guess this could be done by interpolating the current position towards the new position. However I am not sure what kind of step parameter to use for the interpolation. Because in this situation I don't need it to arrive at its destination in a fixed amount of time, I just need to add the missing positions between frames.

One way to damp it is to move a proportion of the amount, e.g. move 7/8ths of the way from the current position to the desired position each frame. As long as the frame rate is fairly consistent that should look ok.

The basic approach I used it to compute the intersection of the camera ray to the camera plane.

Can you control the distance of the camera ray to the camera plane? I'm thinking that it's too far.

Thanks @Paradigm I'll try that!

@DonTzzy

The camera is placed at 0,5,20 while the object is at 0,0,0. The object I am trying to drag is a simple cube whose side is 1. Could this already be too far? I'm using SharpDX's own Ray.GetPickRay method. Just to make sure I tried with my own classical Vector3.Unproject but it doesn't change anything (and that's what that method is doing).

Here's an excerpt of my code:

// Camera.Orientation is a quaternion holding the current camera rotation
Vector3 vView = Vector3.Transform(new Vector3(0, 0, 1), CameraProvider.Orientation);
// vOffset is the exact intersection point expressed by the mouse on the surface of the cube
Plane plane = new Plane(vOffset, vView);
Vector3 vCur;
Matrix mViewProjection = CameraProvider.View * CameraProvider.Projection;
Ray pickRay = Ray.GetPickRay((int)vMoveCurrent.X, (int)vMoveCurrent.Y, CameraProvider.Viewport, mViewProjection);

if (pickRay.Intersects(ref plane, out vCur))
{
	// Compute translation delta 
	vTranslate = vCur - vOffset;
	// CurrentMatch is the object picked and translate changes the 
	// world matrix of the object to the one resulting from vPosition
	vPosition += vTranslate;
	CurrentMatch.Translate(vPosition);

	vOffset = vCur;
}

For this kind of problem i like to use Lerp, which is easy to implement.

When the points are very close, this is verry slow, and when you'll drag it you'll have a little smooth.

Vector2 position = new Vector2(100,100);
Vector2 destination = new Vector2(101,101);
position = Vector2.Lerp(position, destination, 0.8);

http://keithmaggio.wordpress.com/2011/02/15/math-magician-lerp-slerp-and-nlerp/

or you can use average time of frame before and actual frame time