Kurzbeschreibung: Optimization is a key tool for the control of dynamical systems. In autonomous vehicles, for instance, optimization algorithms allow autonomous planning of optimal and dynamically feasible routes. Our approach is based on a combination of classical optimization methods and graph-based planning: Pre-computed motion primitives form a library, from which a planning problem during runtime is solved via sequencing. By extending turnpike theory from optimal control, we can show when it is optimal to move along a primitive. Moreover, in autonomous driving, the library of motion primitives shall represent the behavior of human drivers. Thus, we apply machine learning techniques in order to design a representative motion primitives library in terms of an automaton.