Supervisor: Heiko Donat
Robots with many degrees of freedom describe often complex kinematics functions. To learn such functions with high precision a large and well distributed data set sampled from the whole workspace of the robot is necessary. Given the wear and tear of real robots and the time needed to move the robot to specific joint positions, such a data set can be hard to generate.
A solution to such a problem is given by the Kinematic Bézier Maps (KBM) a parametrizable model without the generality of other systems, but whose structure readily incorporates some geometric constraints of a kinematic function. The simplicity of the model reduces learning to soling linear least squares problems. KBMs show an excellent interpolation and extrapolation capability and relatively low sensitivity to noise.
