Determining Required Simulation Model Fidelity for Developing an Advanced Driver Assistance System for Automated Lane Change Decision Making

Abstract:
Simulation data generation recently gains increasing attention, as the development of automation systems often involves machine learning algorithms. Those algorithms need a large amount of data to work properly. A big challenge regarding simulations is the trade-off between model fidelity and computation costs. This contribution proposes a method to determine the required model fidelity for a simulation by training support vector machines to map the usage boundaries of regarded simulation models. This method is validated by developing an advanced driver assistance system for automated lane change decision making with help of a support vector machine as well.