In my research I aim to connect machine learning with classical methods like computer vision, complex optimization and taking account of physical constraints, to improve capability, efficiency and robustness. The resulting tools are used to solve complex visual problems, including industrial applications like interactive robot control.