Calibration and Assessment of Urban Microscopic Traffic Simulation as an Environment for Testing of Automated Driving

The development and testing of automated driving systems currently pose a challenge for the automotive industry. Since real world tests of automated driving systems are time consuming, often infeasible and potentially dangerous, especially the evaluation of urban automated driving must begin with simulation in a realistic virtual environment. Therefore, it has been proposed to employ traffic simulation to provide an environment of traffic participants for an automated driving system to interact with. For the achievement of this goal, this work presents two contributions. Firstly, we provide an optimal parameterization of existing driver behavior models for urban traffic based on real-world vehicle trajectory data gathered in Ingolstadt, Germany. Secondly, the result of the calibration is assessed for potential improvements and central design aspects of future models for environment simulation. Using a simulation of the city in SUMO, the best calibration result is achieved with the Krauss model with slightly modified parameters compared to the default parameters. Deviations between the calibrated simulation and the real-world data are evaluated for individual roads. Assessment of residuals between real-world data and calibrated simulation suggests that behavior could be dependent on characteristics not yet represented in the simulation models, such as number of lanes or speed limit.