Safety-critical scenarios and virtual testing procedures for automated cars at road intersections

2018-08-09T15:12:27Z (GMT) by Philippe Nitsche
This thesis addresses the problem of road intersection safety with regard to a mixed population of automated vehicles and non-automated road users. The work derives and evaluates safety-critical scenarios at road junctions, which can pose a particular safety problem involving automated cars. A simulation and evaluation framework for car-to-car accidents is presented and demonstrated, which allows examining the safety performance of automated driving systems within those scenarios. Given the recent advancements in automated driving functions, one of the main challenges is safe and efficient operation in complex traffic situations such as road junctions. There is a need for comprehensive testing, either in virtual testing environments or on real-world test tracks. Since it is unrealistic to cover all possible combinations of traffic situations and environment conditions, the challenge is to find the key driving situations to be evaluated at junctions. Against this background, a novel method to derive critical pre-crash scenarios from historical car accident data is presented. It employs k-medoids to cluster historical junction crash data into distinct partitions and then applies the association rules algorithm to each cluster to specify the driving scenarios in more detail. The dataset used consists of 1,056 junction crashes in the UK, which were exported from the in-depth On-the-Spot database. The study resulted in thirteen crash clusters for T-junctions, and six crash clusters for crossroads. Association rules revealed common crash characteristics, which were the basis for the scenario descriptions. As a follow-up to the scenario generation, the thesis further presents a novel, modular framework to transfer the derived collision scenarios to a sub-microscopic traffic simulation environment. The software CarMaker is used with MATLAB/Simulink to simulate realistic models of vehicles, sensors and road environments and is combined with an advanced Monte Carlo method to obtain a representative set of parameter combinations. The analysis of different safety performance indicators computed from the simulation outputs reveals collision and near-miss probabilities for selected scenarios. The usefulness and applicability of the simulation and evaluation framework is demonstrated for a selected junction scenario, where the safety performance of different in-vehicle collision avoidance systems is studied. The results show that the number of collisions and conflicts were reduced to a tenth when adding a crossing and turning assistant to a basic forward collision avoidance system. Due to its modular architecture, the presented framework can be adapted to the individual needs of future users and may be enhanced with customised simulation models. Ultimately, the thesis leads to more efficient workflows when virtually testing automated driving at intersections, as a complement to field operational tests on public roads.