Parameterization of Standard Test Scenarios of Automated Vehicles Using Accident Simulation Data
Martin Urban Martin Urban,,Jorge Lorente Mallada Jorge Lorente Mallada,Maria Pohle Maria Pohle,Christian T. Erbsmehl Christian T. Erbsmehl,Pablo Puente Guillen Pablo Puente Guillen,Satoshi Taniguchi Satoshi Taniguchi,Tom Landgraf Tom Landgraf,,,,,,
DOI: https://doi.org/10.46720/f2020-pif-038
2021-09-30
Abstract:Research objective In February 2019, the Japanese society of automotive Engineers published a white paper titled "Proposal for the Autonomous Driving Vehicle Test Scenarios Standards". This white paper presents in a comprehensive way the current development state of a validation method for safety performance, with the aim of safely introducing automated driving systems to the market. As part of the validation method, the white paper proposes thirty highway scenarios that should be used to validate automated driving. The scenarios are defined by combining the road geometry (main road; marge; branch; ramp), the ego vehicle behaviour (lane keep; lane change) and the surrounding environment or behaviour of other vehicles (Cut In; Cut out; Acceleration; Deceleration; synchronic behaviour). The objective of this research is to define the important parameters regarding the behaviour of the ego vehicle in relation to the opponent vehicle, in addition to several geometric and physical parameters needed for the analysis. Methodology Based on police recorded accident data, the Fraunhofer IVI has developed a methodology for creating simulation files in order to recreate and obtain the pre-crash phase of these accidents. More than 8,000 cases are available in the current simulation database with around 2,000 of this are highway cases. The simulation results contain detailed trajectories for each participant involved in the accident with speed, acceleration and trajectory information. Each of the 2,000 highway accidents is assigned to one of the thirty scenario groups using a three-step database analysis. After the assignment process, parameters such as longitudinal and lateral distance, velocity and acceleration or deceleration are calculated for each accident. Subsequently, a statistical analysis is executed for each group that delivers the mean values and standard deviation for each parameter. The methodology used to obtain the important parameters for each scenario can be recreated and applied to any other database containing similar information to the database used for this study. The result of the analysis presented in this abstract is a statistical description of the highway scenarios proposed by the white paper. It allows configuring test scenarios for automated vehicles and systems with real world parameters from accident simulations based on accident data recorded by the police in Saxony. Around 2,000 reconstructed accidents are the basic data for the statistical description of thirty highway scenarios. The police recorded accidents on highways in Saxony are the basic data of this study. Therefore, using data from different topographic regions, other traffic regulations or other driver education requirements, may modify the parameters and the results of the analysis. Key Words: automated driving, test definition, accident scenario, simulation