A large part of the robotics community develops specialized tools to solve specific problems. If we want to compose these tools into a larger system that behaves puroposefully, then we need to understand how these methods can interact with each other. I want to find out how we can organize such processes that generate behavior.

The domain in which I examine this is robotic manipulation. Right now, I am interested in the relation between motion planning and motion control. These are problems and methods on different scales of time and space, yet time and space are continuous, so a hard cut between such methods may be deficient.

I believe that a system which generates behavior should reflect the discreteness and continuousness of its problem setting. The analysis and decomposition of problem structure can naturally lead to a suitable decomposition of a system that acts to solve such a problem. I try to understand how such decomposition can be done automatically.

A first step towards this was my master's thesis where I worked with MMC, a kind of neural network that decomposes the computation of robot kinematics into the simultaneous solution of multiple smaller problems.

I am also interested in pattern formation, procedural geometry, drawing and painting.