GUIDANCE VECTOR FIELD

A good friend of mine is an artist. Last year when he did some kind of an installation he asked me if I could help him on an interesting problem. For an interactive installation he basically wanted to work with reflections on a three-dimensional surface. The general idea was quite simple and the setup basically as follows:
He wanted to have two sculptures in a room which could change their position in space. Pretty much in the middle of the room there is a three-dimensional surface, which could be deformed by using some servos with Arduino. This surface responds to the movement and position of the visitor and get modified in such a way that the two sculptures are always visible as reflection on the surface. Even if they are not visible directly. Well, that was the plan. He already had great ideas how to build the reflecting surface in reality and also knew how to build the servo-rig. Tracking the visitors wasn’t a problem either but what he didn’t know was how to define the shape of the surface based on the incoming data, which in turn he needed to control the servos. That was the problem I was supposed to take care of since I promised him to help.
So, just to summarize the “surface-problem”:

  • The surface should be deformed in such a way that both sculptures are visible as reflection as much as possible
  • The surface should be as smooth as possible because of aesthetic reasons
  • Due to material constraints, the surface area should stay pretty much the same
  • Since it’s all about a responsive surface in an interactive installation, it has to work fast

So, because I had no clue how to solve this problem I started doing some prototyping in Houdini. I first set up a simple scene with two spheres representing the sculptures, a grid representing the surface and of course, a “visitor” (in fact, another sphere). After some unsuccessful tests I started to compute various vector fields describing visual rays and reflections in the scene. And suddenly it made “click” – I knew how to do it. Looking at the setup I had so far, it became quite apparent that I could just use the vector field as gradient field in the Poisson equation. I tried it immediately and it worked.

The vector field was easy to compute. For every polygon of the grid I had to compute the vectors pointing to both “sculptures” by using a weight-function based on distance and reflection angle to finally blend between them. This way I got the direction in which visual rays needed to be reflected and hence I also knew the optimal orientation of each polygon. After that I applied a local rotation to each polygon of the mesh and obtained the new orientation. By doing so, every polygon turned to it’s “best” direction and the mesh was torn apart. Subsequently it needed to be reconstructed based on the new orientation of its polygons. And this is exactly the point where the Poisson equation came into play.

The setup was quite simple. Based on the new orientation I calculated a gradient for x, y and z, computed the divergence for each of these and used the result as new gradient field in the Poisson equation. After solving the Poisson equation, which is in fact a sparse linear system, the result is a piecewise continuous function. In other words, it is the final mesh reconstructed in the least squares sense.

 

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

w

Connecting to %s