Efficient Simulation Based Calibration of Automated Driving Functions Based on Sensitivity Based Optimization

Increasing demands on reliability and safety of automated driving functions require an augmented usage of simulation tools for the efficient calibration of these functions. However, finding an optimal solution can be costly, especially when the objective function is represented by scenario simulations. To face these challenges, a novel optimization scheme for simulation based calibration problems, that enables reduced computational effort is introduced. The approach is based on sensitivity analyses that provide scenario specific influential parameter spaces. Using these information, all parameter combinations are checked for reference candidates obtained in preceding iterations that are expected to have an equivalent solution as the new set. Thus, expensive simulation runs can be replaced by taking results from a reference set. The so called `scenario simulation reduction' approach is applied to the parameterization of an SAE level 3 automated driving function with a genetic algorithm as optimizer. In order to take modeling inaccuracies into account, a robustness analysis with respect to simulation model parameters is conducted. Finally, a validation of the optimization scheme is performed using an extensive sampling approach. Studies confirm that negligible errors occur that are not expected to disturb optimization progress.