Use of a criticality metric for assessment of critical traffic situations as part of SePIA

Autoren

Dipl.-Ing. Matthias Lehmann, Dipl.-Ing. Maximilian Bäumler MBA, Prof. Dr.-Ing. Günther Prokop, Technische Universität Dresden;
Dipl.-Ing. (FH) Diana Hamelow, Verkehrsunfallforschung, Technische Universität Dresden

Jahr

2019

Zusammenfassung

The requirements for a driver driving through a traffic situation increase with the complexity of the infrastructure, the number of participants, their possible actions and the environmental conditions. Thereby, the human being as the controller of the vehicle is normally able to analyze complex situations and to act accordingly. Nevertheless, in some cases errors occur in the levels of:
– Information access,
– Information intake,
– Information processing,
– Target setting, and
– Action [1].
These errors lead to deviation from the safe driving behavior and can, in the worst case, end up in an accident. To prevent the driver from an accident, advanced driver assistance systems (ADAS) are designed to support drivers in their driving task. Simulations allow variable testing, but for a transfer modelling in the driver, vehicle and environment is needed. Especially the modelling of drivers and their behavior is a big challenge, which can be modelled stochastically by varying the different character traits [2].
Currently developed Level 3-systems can fully release the driver from the control loop (e.g. Audi traffic jam pilot) in certain situations. Thus, a main task in the development of such systems is to prove that the system can handle as many situations as possible and that it is better as a human being in order to increase the traffic safety. This also involves the evaluation and analysis of transfer scenarios between humans and ADAS that require the use of highly immersive driving simulators [3].
The system assessment is also important for the homologation and the periodic technical vehicle monitoring. For the latter, the assessment is intended to check if a certain ADAS can handle new situations caused by changed traffic events, e.g. due to a higher percentage of old and/ or new ADAS in road traffic or due to infrastructural changes [4].
Currently, the information needed for the generation of simulative testing scenarios of ADAS is mainly extracted from accident databases. These databases contain information about the initial critical conflict, but can never be fully realistic, because of uncertainties and assumptions made in the reconstruction process. Another way to obtain real world information is the use of real driving data. According to an estimation of Wachenfeld and Winner [5], 6.62 billion kilometers have to be driven before the safety of a fully automated highway pilot is proven.
Thereby, every change of the ADAS, e.g. of the software version, requires a completely new verification process. This high expenditures can be avoided with a scenario approach for ADAS testing. This approach is therefore very suitable for simulative testing of complete systems or parts of it. However, for the scenario set-up databases are needed, which can represent the real road traffic events. Besides the accident research data and police reports, natural driving studies (NDS) can represent the real road traffic events realistically and can so be useful as a basis for generating scenarios. The reason for this is, that NDS contain not only situations which can appear in real traffic but also situations, which have become critical or have led to accidents. But these situations
have to be extracted from the NDS data first.
Within the SePIA (Scenario based platform for inspection of advanced driver assistance systems) project, different databases, like NDS, are analyzed and combined to set up a representative prototypical platform containing rated and characterized scenarios in a simulable format for the assessment of ADAS. In the future, this platform can be evolved with updated and accessible data sources representing changes in traffic events.

Mitglieder-Login