Time: 9:00 am-10:00 am
Venue: Room "Girasole", Via Sommarive, 9 - Trento
- Matthias Gerdts, Institute of Mathematics and Applied Computing - Department of Aerospace Engineering Universität der Bundeswehr, München
Modern vehicles offer a number of passive and (semi-)active assistance systems, which are supposed to support the driver especially in crucial situations. The degree of autonomy increases continuously and eventually ends in fully autonomous driving. Next to challenges on technical and legal levels, also algorithmic challenges arise. Algorithms are particularly needed to perform control tasks that are usually performed by a human driver. This applies to both, real and virtual vehicles in a simulation environment. The talk focuses on mathematical methods and concepts to control vehicles or vehicle components by means of optimal control and model predictive control. To this end, a core task is to provide trajectories which optimize a given performance criterion (fuel consumption, comfort, safety, time, etc) and obey constraints (collision avoidance, safety regions, control bounds, etc). In order to solve these problems numerically, we typically use direct discretisation schemes and sensitivity updates to predict optimal solutions in the presence of perturbations. Herein, discrete-valued controls or decisions cause an additional level of difficulty and require tailored solution approaches. Some approaches to handle such mixed-integer optimal control problems will be presented.