What Is the Best Way to Optimally Parameterize the MPC Cost Function for Vehicle Guidance?

Model predictive control (MPC) is a promising approach to the lateral and longitudinal of autonomous vehicles. However, parameterization MPC with respect high-level requirements such as passenger comfort, well tracking, challenging. Numerous tuning parameters conflicting need be considered. In this paper, we formulate task multi-objective optimization problem. Its solution demanding for two reasons: First, MPC-parameterizations are evaluated in computationally expensive simulation environment. As result, algorithm needs sample-efficient possible. Second, some poor parameterizations, cannot completed; therefore, useful objective function values not available (for instance, learning crash constraints). work, compare sample efficiency particle swarm (MOPSO), genetic (NSGA-II), multiple versions Bayesian (BO). We extend BO by introducing an adaptive batch size limit computational overhead. addition, devise method deal constraints. The results show that works best small budget, NSGA-II medium budgets, none optimizers superior random search large budgets. Both proposed extensions are, shown beneficial.