Autonomous driving

Model predictive control is a numerics-based controller, that relies on repeated open-loop online computations in feedback fashion. The ability to solve increasingly complex optimal control problems has been made possible thanks to the increased, high-speed computers in recent decades. In this context, model predictive control represents a promising engineering solution that systematically accommodate for input and state constraints while performing receding horizon control allowing for feasible trajectories, which is of particular relevance in autonomous driving. In autonomous driving, our goal is to guide a vehicle towards safe trajectories autonomously via the execution of manoeuvers such as accelerating and braking for the vehicle under control, while satisfying the constraints specified for given traffic scenarios. With a constantly changing and possibly unpredictable driving environment, it is crucial that the vehicle trajectory takes into account the dynamic surrounding while satisfying physical and temporal constraints.

We investigate model predictive control (MPC) schemes to solve the challenges arising in autonomous driving in uncertain environments. 

MPC schemes for uncertainty prediction for testing of autonomous vehicles 

We leverage MPC schemes to design a framework for uncertainty description based on vehicle models and current measurements. For this, we leverage advanced MPC schemes to accomodate for uncertainties arising from unknown environment, while ensuring constraint satisfactions.

Distributed MPC schemes for testing of autonomous vehicles

We investigate different methods that aim to reduce the computational effort associated with solving finite-time optimal control problem. For this, we leverage distributed MPC schemes to reduce the computational time of solving the optimization problems for trajectory prediction, first assuming full control over the vehicle under test. In a second step, we design robust approaches that are resilient to the apriori unknown trajectory of the autonomous vehicle under test and implement this numerically on specified driving scenarios.