Integrierte Werkzeugkette zur Nutzung von realen verkehrlichen Szenarien für simulationsbasierte Absicherung des automatisierten Fahrens

Abstract

In mixed traffic of the future, automated and non-automated road users will interact with each other. The design and testing of automated driving functions are essential for safe traffic. Due to the rarity of certain scenarios, not all test cases can be covered by test drives. In this context, simulative testing provides a solution. A major challenge is to depict a realistic range of traffic events in normal, but also rare and critical scenarios in the simulation. This is where the present work begins, with the aim of presenting an integrated tool chain that enables the use of real-life scenarios to realistically design the simulation. The structure of the tool chain for scenario extraction from real data and its simulation is presented and an insight into individual use cases is given. At the beginning of the tool chain real driving behavior data of motorized and non-motorized road users is collected. For this purpose, infrastructurally installed camera technology is used, which allows continuous (24/7) traffic monitoring. This is carried out stationary along the A39 motorway (Testbed Lower Saxony) and at an urban intersection in Braunschweig (AIM Research Intersection). It is also made possible in a mobile manner with a measuring trailer, so that roundabouts or level crossings can also be observed. The movement paths (trajectories) of road users are determined from the recorded video footage using object detection and tracking. At the same time, it is possible to view and plausibilise the traffic event in the video recordings. Subsequently, relevant traffic scenarios must be extracted from the trajectory data (so-called scenario mining). Algorithms have been developed that identify interesting scenarios based on certain parameters such as strong braking maneuvers or traffic safety metrics such as time-to-collision (TTC). Based on this information, traffic events can be modelled with regard to driving and interaction behavior in normal, critical and atypical situations. Selected scenarios are collected in a database together with the range of scenario-related parameters (e.g. driving speeds, distances between relevant road users or the position and orientation of a central vehicle at a certain point in time). In the next step, the scenarios are processed for use in the simulation using the OpenSCENARIO industry standard. This paper provides an insight into the conversion and modelling into concrete and logical scenarios. The prepared OpenSCENARIO descriptions are then transferred to a DLR in-house developed simulation tool chain, which enables the task of simulation- and scenario-based validation of automated driving functions. This is the simulation toolchain OSTAR (Open Source Toolchain for Automotive and Railway Research), which Carla uses for the central simulation of vehicle physics, sensor simulation and visualization. OSTAR enables manufacturer-independent model integration (e.g. of sensors, driving dynamics or a driving function) using the industry standards Functional Mockup Interface (FMI) and Open Simulation Interface (OSI). The tool chain is freely available as an open source implementation. The output of the simulation chain is the model and simulator output in the form of OSI-compliant messages, which can then be used for analysis and evaluation. In this presentation, examples of use and evaluation are given (UNECE R157) as well as prospects for further developments and research projects (scenario exploration, model validation, infrastructure conceptualization). The main innovation in the work presented here is the comprehensive and direct integration of findings from traffic observations into simulation-based testing. There are incremental innovations on the scenario mining side and in the implementation of the simulation tool chain.