Simulation of Test Drives by Using Police-recorded Accident Data and Combining Macroscopic and Microscopic Elements
With the development of autonomous driving functions, the evaluation of their functional safety is becoming increasingly important. Current vehicles are tested with separate simulations or test drives. In order to validate future autonomous vehicles by means of test drives, a substantial number of test kilometers are necessary. In addition, these test drives must be repeated for every new release of the system, which increases the expenses for validation. For this reason, programs that can simulate test drives have a high significance. Previous programs do not include the indispensable combination of routing simulation and accident simulation needed to represent a simulated test drive. Therefore, an approach to combining a macroscopic simulation (routing simulation) with a microscopic simulation (accident simulation) is used in this paper. When the start location and the destination are given, the macroscopic simulation can compute the test route by means of the OSRM (Open Source Routing Machine) routing application. While driving along the test route, the simulated vehicles pass various locations of real accidents. The relevant data is taken from the accident database compiled by the police of Saxony, Germany. A selection procedure ensures that only relevant accident situations along the test route are later simulated microscopically. Only if the accident situation is similar to the current situation of the simulated vehicle can the accident situation be simulated microscopically. Therefore, various boundary conditions are used to determine whether there are similarities regarding weather, traffic, light conditions and trajectories of the accident vehicles. To study different variations of the selection procedure, three different concepts are developed and evaluated. The first concept is based on a given test route between start location and destination and a realistic calculation of the travel time. The second concept is also based on a given test route but combines this with a time window for the entire route. The third concept combines an unknown test route, which is calculated between relevant accident locations during the simulation, with a realistic calculation of the travel time. After the evaluation of all three concepts, only the third concept is implemented in the simulation. Within the microscopic simulation by means of PC-Crash, a relevant accident situation is simulated twice, once without and once with the tested driver assistance system in action. With the help of a collision detection system, a conclusion about the efficiency of the driver assistance system is made. The result is a program that combines completed test kilometers with avoided accident situations to simulate a test drive. The current program can only be used in Saxony, Germany. For an expansion to all of Europe, comprehensive accident data is necessary. In addition, the selection procedure could be improved by means of georeferenced weather and traffic data. Because of the basic simulation tools, the actual simulation is not designed for quality but rather for quantity. However, high-quality simulation tools can be implemented with little effort. The simulation of test drives is an important challenge, and with the program developed here, an opportunity to solve it is introduced.